Model: gemini-exp-1206
tweets.json
[
{
"datetime": "2024-10-07T14:10:49.000Z",
"username": "theprincessxena",
"tweet": "A large language model hooked up to tools that lets it do useful things",
"href": "https://twitter.com/theprincessxena/status/1843292901960524047",
"likes": 9,
"impressions": 869,
"retweets": 0
},
{
"datetime": "2024-10-07T14:29:17.000Z",
"username": "maciejgryka",
"tweet": "A state machine affecting some external system, which uses an LLM to make some of its decisions.",
"href": "https://twitter.com/maciejgryka/status/1843297550176240110",
"likes": 1,
"impressions": 75,
"retweets": 0
},
{
"datetime": "2024-10-07T14:49:58.000Z",
"username": "amplifiedamp",
"tweet": "Agents aren't real in the traditional sense of being confined to one LLM or human body, so:\n\nAny property of a system that causes the system and its possession of that property to continue existing by means of the property",
"href": "https://twitter.com/amplifiedamp/status/1843302754187690258",
"likes": 14,
"impressions": 1863,
"retweets": 1
},
{
"datetime": "2024-10-07T15:57:56.000Z",
"username": "patricksrail",
"tweet": "The most AI thing ever is that you can ask for definitions of key terms like agent, AGI, or sentience, and never get the same answer twice.",
"href": "https://twitter.com/patricksrail/status/1843319859834450155",
"likes": 1,
"impressions": 156,
"retweets": 0
},
{
"datetime": "2024-10-07T16:46:59.000Z",
"username": "Tjido",
"tweet": "The \n@TheoriqAI\n litepaper defines agents as, \"AI agents are autonomous software systems that leverage modern generative AI models to plan, access data, use tools, make decisions, and interact with the real world to perform specific functions.\" \n- But a picture is worth a thousand",
"href": "https://twitter.com/Tjido/status/1843332203100439034",
"likes": 29,
"impressions": 1261,
"retweets": 2
},
{
"datetime": "2024-10-07T15:06:17.000Z",
"username": "productaizery",
"tweet": "\"agent\" means a software program to accomplish a user-specified objective, using LLMs to plan tasks and make decisions. The agent can make use of tools - helper programs and third-party APIs, to run tasks. This plan-decide-act loop is repeated until the objective is achieved.",
"href": "https://twitter.com/productaizery/status/1843306860600000676",
"likes": 12,
"impressions": 435,
"retweets": 0
},
{
"datetime": "2024-10-07T14:38:49.000Z",
"username": "upster",
"tweet": "An application that can autonomously learn, adapt, and act in pursuit of its assigned goal.",
"href": "https://twitter.com/upster/status/1843299951092355242",
"likes": 19,
"impressions": 1638,
"retweets": 1
},
{
"datetime": "2024-10-07T17:22:03.000Z",
"username": "Plinz",
"tweet": "I am using 'agent' to mean 'control system for future states'",
"href": "https://twitter.com/Plinz/status/1843341026662199546",
"likes": 28,
"impressions": 1412,
"retweets": 1
},
{
"datetime": "2024-10-07T14:42:25.000Z",
"username": "romechenko",
"tweet": "An agent is trusted to act on behalf of and in the interest of those being represented. \n\nEg an agent of the government has authority to say/do things that the government will be responsible for.\n\nIf there’s no trust there is no agent.",
"href": "https://twitter.com/romechenko/status/1843300857405956384",
"likes": 0,
"impressions": 1882,
"retweets": 0
},
{
"datetime": "2024-10-07T14:22:54.000Z",
"username": "thomasrice_au",
"tweet": "An AI agent is a system that can autonomously plan and use tools to achieve a goal that doesn't rely on following a predetermined set of steps.",
"href": "https://twitter.com/thomasrice_au/status/1843295943967527340",
"likes": 59,
"impressions": 2200,
"retweets": 0
},
{
"datetime": "2024-10-07T15:09:35.000Z",
"username": "greynewell",
"tweet": "an agent is an entity which is capable of producing goal-seeking behavior independently and without oversight.",
"href": "https://twitter.com/greynewell/status/1843307691772694846",
"likes": 23,
"impressions": 1402,
"retweets": 0
},
{
"datetime": "2024-10-07T14:20:42.000Z",
"username": "fred_pope",
"tweet": "A button a business can deploy which can be clicked over and over with the dual objective of save and make money.",
"href": "https://twitter.com/fred_pope/status/1843295389581279276",
"likes": 1,
"impressions": 354,
"retweets": 0
},
{
"datetime": "2024-10-07T18:18:07.000Z",
"username": "Imaadmkhan1",
"tweet": "Agents = LLMs that can access and use tools to perform well defined tasks and workflows.",
"href": "https://twitter.com/Imaadmkhan1/status/1843355140113531079",
"likes": 0,
"impressions": 14,
"retweets": 0
},
{
"datetime": "2024-10-07T16:11:51.000Z",
"username": "jeffrschneider",
"tweet": "It's software that fulfills aspects of 'agency'.\nhttps://en.wikipedia.org/wiki/Agency_(psychology) \n\nPretty sure that's why we called them agents ",
"href": "https://twitter.com/jeffrschneider/status/1843323362468397345",
"likes": 1,
"impressions": 83,
"retweets": 1
},
{
"datetime": "2024-10-07T15:33:20.000Z",
"username": "abrakjamson",
"tweet": "Software with:\n-Multi-step processing\n-Unscripted reasoning loops\n-Work long-term in the background\n-Using LLM tools\n-Using human tools\n-Draft communications on the user’s behalf\n-Interact with a different human or agent\n-Choose when to execute\n-Store memories\n-Learn new skills",
"href": "https://twitter.com/abrakjamson/status/1843313669519909063",
"likes": 3,
"impressions": 132,
"retweets": 0
},
{
"datetime": "2024-10-07T14:10:10.000Z",
"username": "EmilStenstrom",
"tweet": "LLM-based program that can make a plan, iteratively execute that plan, using tools if needed.",
"href": "https://twitter.com/EmilStenstrom/status/1843292740442103985",
"likes": 3,
"impressions": 114,
"retweets": 0
},
{
"datetime": "2024-10-07T15:32:39.000Z",
"username": "Dorialexander",
"tweet": "OK so going blind: any reasoning machine able to deal in the long run with bureaucracy and maintenance. So finding solution especially when things turn out unexpectedly. Agent should not be dependant on its internal knowledge basis but able to act on new info independently.",
"href": "https://twitter.com/Dorialexander/status/1843313499046621207",
"likes": 5,
"impressions": 246,
"retweets": 0
},
{
"datetime": "2024-10-07T14:11:55.000Z",
"username": "danbri",
"tweet": "@WiringTheBrain",
"href": "https://twitter.com/danbri/status/1843293178214158629",
"likes": 1,
"impressions": 167,
"retweets": 0
},
{
"datetime": "2024-10-07T15:50:57.000Z",
"username": "pamelafox",
"tweet": "definition from \n@sethjuarez\n (who can correct my phrasing): an agent is LLM-based control flow",
"href": "https://twitter.com/pamelafox/status/1843318104409157988",
"likes": 3,
"impressions": 178,
"retweets": 0
},
{
"datetime": "2024-10-07T14:26:31.000Z",
"username": "genmon",
"tweet": "Good prompt! Something like...\n\n- software using an LLM internally\n- to perform one or more steps towards a goal (step sequence not pre-determined)\n- and the steps involve tool use: updating state outside the LLM completion or inter-op with another system",
"href": "https://twitter.com/genmon/status/1843296853632352406",
"likes": 7,
"impressions": 340,
"retweets": 1
},
{
"datetime": "2024-10-08T07:57:08.000Z",
"username": "adridder",
"tweet": "Viewing AI as our agent is thought-provoking. But can it truly represent us?",
"href": "https://twitter.com/adridder/status/1843561250074017907",
"likes": 0,
"impressions": 45,
"retweets": 0
},
{
"datetime": "2024-10-08T04:04:36.000Z",
"username": "cbnsndwch",
"tweet": "I like the classic definition:\n\nAn entity that senses its environment (takes/receives input) and acts on it, potentially changing it in the process.",
"href": "https://twitter.com/cbnsndwch/status/1843502729961828461",
"likes": 0,
"impressions": 35,
"retweets": 0
},
{
"datetime": "2024-10-08T02:30:17.000Z",
"username": "padolsey",
"tweet": "An agent is a program that starts or contains AI \"runtimes\" (whether single or multiple ongoing inferences). It is given an interface to make one or more decisions affecting something outside of its own runtime. It becomes 'agentic' when it can manifest will into action.",
"href": "https://twitter.com/padolsey/status/1843478996253618643",
"likes": 0,
"impressions": 73,
"retweets": 0
},
{
"datetime": "2024-10-07T22:47:48.000Z",
"username": "rickwong888",
"tweet": "“Takes action on your behalf”\n\nIt intentionally works also to describe any type of delegation.",
"href": "https://twitter.com/rickwong888/status/1843423006057771411",
"likes": 0,
"impressions": 24,
"retweets": 0
},
{
"datetime": "2024-10-07T22:53:13.000Z",
"username": "mysterfrederic",
"tweet": "An AI that's not just capable of understanding a prompt and come up with a response, but also determine the appropriate tasks and, if applicable, carrying them out until completion.",
"href": "https://twitter.com/mysterfrederic/status/1843424368334094582",
"likes": 0,
"impressions": 32,
"retweets": 0
},
{
"datetime": "2024-10-07T19:16:10.000Z",
"username": "jimboot",
"tweet": "A single focused instance of an LLM to perform a specific task",
"href": "https://twitter.com/jimboot/status/1843369747888648482",
"likes": 0,
"impressions": 71,
"retweets": 0
},
{
"datetime": "2024-10-07T17:40:15.000Z",
"username": "bons_vincent",
"tweet": "A system that can observe its environment and take actions based on those observations. Usually with the aim of achieving a goal.",
"href": "https://twitter.com/bons_vincent/status/1843345608993894655",
"likes": 0,
"impressions": 58,
"retweets": 0
},
{
"datetime": "2024-10-07T16:08:41.000Z",
"username": "arbuge",
"tweet": "A collection of prompts in result-dependent sequence to an LLM and/or supporting LLM functions to complete a specific task.",
"href": "https://twitter.com/arbuge/status/1843322566766923927",
"likes": 0,
"impressions": 57,
"retweets": 0
},
{
"datetime": "2024-10-08T11:07:03.000Z",
"username": "ESchwaa",
"tweet": "An agent is an autonomous service that uses AI at its core to deeply interact with a specific set of data or act as a certain persona in a dynamic workflow. I like to think of agents as subject matter experts that will someday be the equivalent of an employee or other function",
"href": "https://twitter.com/ESchwaa/status/1843609044415554020",
"likes": 0,
"impressions": 27,
"retweets": 0
},
{
"datetime": "2024-10-07T17:39:49.000Z",
"username": "danlemire",
"tweet": "Here's mine from my Agent explainer from two weeks ago:\n\nautonomous + conducting research + comparing options + making decision + interacting with other systems + taking actions",
"href": "https://twitter.com/danlemire/status/1843345500210340106",
"likes": 0,
"impressions": 90,
"retweets": 0
},
{
"datetime": "2024-10-07T16:06:59.000Z",
"username": "summitbytes",
"tweet": "An AI agent is a system that autonomously processes inputs, formulates strategies, and executes actions to meet user defined goals, adapting its behavior based on feedback and outcomes.\n\nMine differs because it includes user feedback, which I think is essential",
"href": "https://twitter.com/summitbytes/status/1843322137169506754",
"likes": 0,
"impressions": 55,
"retweets": 0
},
{
"datetime": "2024-10-07T16:10:31.000Z",
"username": "dpunjabi",
"tweet": "(Adaptive) multimodal system interacting with bound environments.",
"href": "https://twitter.com/dpunjabi/status/1843323028232646692",
"likes": 0,
"impressions": 82,
"retweets": 0
},
{
"datetime": "2024-10-07T20:03:22.000Z",
"username": "mbrg0",
"tweet": "a system that is trusted to make impactful decisions on behalf of",
"href": "https://twitter.com/mbrg0/status/1843381624630870182",
"likes": 0,
"impressions": 94,
"retweets": 0
},
{
"datetime": "2024-10-07T18:14:41.000Z",
"username": "_cartermp",
"tweet": "A process that, given a task, makes external function calls (sometimes to an LLM) and finishes the task after some task-terminating condition is reached",
"href": "https://twitter.com/_cartermp/status/1843354273125630397",
"likes": 0,
"impressions": 209,
"retweets": 0
},
{
"datetime": "2024-10-07T14:29:03.000Z",
"username": "intellectronica",
"tweet": "Goal-oriented compound AI system.",
"href": "https://twitter.com/intellectronica/status/1843297490474545625",
"likes": 2,
"impressions": 325,
"retweets": 0
},
{
"datetime": "2024-10-07T18:06:41.000Z",
"username": "imaurer",
"tweet": "\"A theoretical construct with infinite potential and zero real-world applications, perfect for impressing venture capitalists.\"",
"href": "https://twitter.com/imaurer/status/1843352260845416802",
"likes": 0,
"impressions": 177,
"retweets": 0
},
{
"datetime": "2024-10-07T14:02:11.000Z",
"username": "simonw",
"tweet": "Let’s see if we can crowdsource a robust definition of “agent” (with respect to AI and LLMs) that fits in a <=280 character tweet\n\nReply to this with your best attempt, then scroll through the replies and fave the ones that makes sense to you",
"href": "https://twitter.com/simonw/status/1843290729260703801",
"likes": 524,
"impressions": 0,
"retweets": 71
},
{
"datetime": "2024-10-07T14:25:12.000Z",
"username": "simonw",
"tweet": "Wow, tons of definitions (pretty varied) so far but very few votes! Please don’t forget to scan down the list and stick a like on the ones that resonate with you",
"href": "https://twitter.com/simonw/status/1843296520780886128",
"likes": 27,
"impressions": 9020,
"retweets": 0
},
{
"datetime": "2024-10-07T14:42:47.000Z",
"username": "simonw",
"tweet": "Would also be interesting to hear from the investing side of the world: if you have invested your money in “agents”, what were you investing in?",
"href": "https://twitter.com/simonw/status/1843300945784115400",
"likes": 25,
"impressions": 11463,
"retweets": 1
},
{
"datetime": "2024-10-08T00:58:32.000Z",
"username": "simonw",
"tweet": "A question I always ask myself about any proposed definition of \"agent\" is whether or not ChatGPT Code Interpreter qualifies as an agent under that definition\n\nIt's a great example of a tool-using LLM running in a loop that is trying to achieve a goal - is that \"agentic\" enough?",
"href": "https://twitter.com/simonw/status/1843455905926656307",
"likes": 17,
"impressions": 5421,
"retweets": 0
},
{
"datetime": "2024-10-07T14:38:06.000Z",
"username": "swyx",
"tweet": "agent = llm + memory + planning + tools + while loop",
"href": "https://twitter.com/swyx/status/1843299770238181825",
"likes": 0,
"impressions": 14454,
"retweets": 10
},
{
"datetime": "2024-10-07T14:09:25.000Z",
"username": "hwchase17",
"tweet": "The more agentic an application is, the more an LLM decides the control flow of the application",
"href": "https://twitter.com/hwchase17/status/1843292549521473957",
"likes": 158,
"impressions": 7699,
"retweets": 6
},
{
"datetime": "2024-10-07T14:09:28.000Z",
"username": "Dan_Jeffries1",
"tweet": "The one we use internally is:\n\n\"An AI system that's capable of carrying out and completing long running, open ended tasks in the real world.\"\n\nThe real world can be physical or online.",
"href": "https://twitter.com/Dan_Jeffries1/status/1843292564734230616",
"likes": 175,
"impressions": 4712,
"retweets": 1
},
{
"datetime": "2024-10-07T15:55:46.000Z",
"username": "deepfates",
"tweet": "agents are software programs to which it is useful to imply the intentionality abstraction. ala,",
"href": "https://twitter.com/deepfates/status/1843319314566500498",
"likes": 23,
"impressions": 1445,
"retweets": 0
},
{
"datetime": "2024-10-07T14:34:25.000Z",
"username": "_ScottCondron",
"tweet": "An autonomous system powered by LLMs that understands natural language inputs, makes decisions, and executes actions to achieve user-defined goals. It translates instructions into commands, interacts with tools, adapts to feedback, and efficiently manages complex tasks.",
"href": "https://twitter.com/_ScottCondron/status/1843298841761861988",
"likes": 37,
"impressions": 2542,
"retweets": 0
},
{
"datetime": "2024-10-07T14:23:40.000Z",
"username": "_coenen",
"tweet": "A system capable of executing complex, multi-step tasks with little to no human intervention",
"href": "https://twitter.com/_coenen/status/1843296135966024099",
"likes": 54,
"impressions": 2588,
"retweets": 0
},
{
"datetime": "2024-10-07T18:59:46.000Z",
"username": "DanielMiessler",
"tweet": "“An AI system component that performs non-trivial, multi-step operations that previously would have required a human.”",
"href": "https://twitter.com/DanielMiessler/status/1843365619246543174",
"likes": 3,
"impressions": 230,
"retweets": 0
},
{
"datetime": "2024-10-07T14:55:38.000Z",
"username": "fishnets88",
"tweet": "A division of responsibilities between LLMs that results in some sort of flow?\n\nSpitballing here though, I am also not aware of any proper definition.",
"href": "https://twitter.com/fishnets88/status/1843304179638640803",
"likes": 1,
"impressions": 647,
"retweets": 0
},
{
"datetime": "2024-10-07T15:56:30.000Z",
"username": "zittrain",
"tweet": "Here’s mine, compressible to a short three-element definition. From https://theatlantic.com/technology/archive/2024/07/ai-agents-safety-risks/678864/…",
"href": "https://twitter.com/zittrain/status/1843319499459793269",
"likes": 7,
"impressions": 231,
"retweets": 0
},
{
"datetime": "2024-10-07T16:01:27.000Z",
"username": "DanielMiessler",
"tweet": "“An AI agent is an AI component that interprets instructions and takes on more of the work in a total AI workflow than just LLM response, e.g., executing functions, performing data lookups, etc., before passing on results.”\n\nhttps://danielmiessler.com/p/raid-ai-definitions…",
"href": "https://twitter.com/DanielMiessler/status/1843320745423946076",
"likes": 7,
"impressions": 699,
"retweets": 0
},
{
"datetime": "2024-10-07T14:07:51.000Z",
"username": "hypersoren",
"tweet": "“An agent is any program that does, in each instance, exactly what a user would want if asked specifically.”",
"href": "https://twitter.com/hypersoren/status/1843292155248607704",
"likes": 5,
"impressions": 1181,
"retweets": 1
},
{
"datetime": "2024-10-08T19:05:39.000Z",
"username": "wiredmau5",
"tweet": "Agents can decide, act & improve their execution for a set of tasks, if they're given ways to understand how other domain experts do it.",
"href": "https://twitter.com/wiredmau5/status/1843729489249087629",
"likes": 0,
"impressions": 14,
"retweets": 0
},
{
"datetime": "2024-10-07T14:29:35.000Z",
"username": "romechenko",
"tweet": "That which exhibits agency in response to internal, external, or anticipated changes",
"href": "https://twitter.com/romechenko/status/1843297624428015803",
"likes": 2,
"impressions": 276,
"retweets": 0
},
{
"datetime": "2024-10-07T16:32:52.000Z",
"username": "saran945",
"tweet": "Enabling LLM to efficiently complete the task",
"href": "https://twitter.com/saran945/status/1843328650525159653",
"likes": 0,
"impressions": 42,
"retweets": 0
},
{
"datetime": "2024-10-08T07:40:32.000Z",
"username": "BatAndrew314",
"tweet": "Agent - is a AI system that is able to understand, plan, and execute on a high-level general goal without explicit pre-training step",
"href": "https://twitter.com/BatAndrew314/status/1843557074128777444",
"likes": 0,
"impressions": 29,
"retweets": 0
},
{
"datetime": "2024-10-07T23:29:47.000Z",
"username": "gravity7",
"tweet": "An agent is an LLM capable of reasoning through natural language interactions and actioning system functions and plans from explicit and implicit user intents—autonomously when coupled with user data & preferences— thus resulting in generative task and workflows.",
"href": "https://twitter.com/gravity7/status/1843433572117745916",
"likes": 1,
"impressions": 89,
"retweets": 0
},
{
"datetime": "2024-10-08T01:19:58.000Z",
"username": "prem_k",
"tweet": "Perceives / Senses \n\nPlans / Routes \n\nActs / Responds \n\nPlanning / Routing in #LLM driven #Agents are retrieval based rather than on reasoning [including o1//other #LRM].",
"href": "https://twitter.com/prem_k/status/1843461302053974126",
"likes": 1,
"impressions": 67,
"retweets": 0
},
{
"datetime": "2024-10-08T02:44:41.000Z",
"username": "WilmsRolf",
"tweet": "An agent actively interacts with tools and/or users until a task is completed.",
"href": "https://twitter.com/WilmsRolf/status/1843482619528528102",
"likes": 0,
"impressions": 21,
"retweets": 0
},
{
"datetime": "2024-10-07T18:27:49.000Z",
"username": "jordanlathmoore",
"tweet": "A mechanism that observes, processes, and acts toward a defined goal. It learns from its environment and adjusts its actions, but it’s always bound by the logical framework it’s been given.",
"href": "https://twitter.com/jordanlathmoore/status/1843357577868165137",
"likes": 0,
"impressions": 58,
"retweets": 0
},
{
"datetime": "2024-10-07T16:10:37.000Z",
"username": "RobRoyce_",
"tweet": "An agent has senses “precepts” and an action space. It uses actions from its action space to effect its environment. It then observes that environment and repeats until it accomplishes its goals. Modern LLM based agents can use, for instance, arbitrary Python functions, as tools",
"href": "https://twitter.com/RobRoyce_/status/1843323050797990106",
"likes": 1,
"impressions": 216,
"retweets": 0
},
{
"datetime": "2024-10-08T03:43:12.000Z",
"username": "veeragoni",
"tweet": "Functionally analogous human: (to name a few:having agency/taking actions, reason/planning, operate with context, long running, pause/resume task, delegate and list goes on)\nTechnically more capable digitally (non embodied) as they are powered by LLMs with lots of memory etc.",
"href": "https://twitter.com/veeragoni/status/1843497344329035928",
"likes": 1,
"impressions": 49,
"retweets": 0
},
{
"datetime": "2024-10-08T07:03:23.000Z",
"username": "MichaelNStruwig",
"tweet": "Any system with the capacity, condition, or ability to act / perform self-directed actions until an end goal is achieved.\n\nThis, for us, is an agent.",
"href": "https://twitter.com/MichaelNStruwig/status/1843547724601516528",
"likes": 0,
"impressions": 41,
"retweets": 0
},
{
"datetime": "2024-10-08T03:07:23.000Z",
"username": "kevins8",
"tweet": "an ai system that can accomplish non-trivial tasks without human intervention",
"href": "https://twitter.com/kevins8/status/1843488333902794773",
"likes": 0,
"impressions": 34,
"retweets": 0
},
{
"datetime": "2024-10-08T01:58:24.000Z",
"username": "rome_viharo",
"tweet": "An agent is a self teaching protocol",
"href": "https://twitter.com/rome_viharo/status/1843470971094876474",
"likes": 0,
"impressions": 13,
"retweets": 0
},
{
"datetime": "2024-10-07T16:33:37.000Z",
"username": "rushing_andrei",
"tweet": "The textbook definition is that it is \"an autonomous software system capable of perceiving its environment, making decisions, and taking actions to achieve specific goals.\"",
"href": "https://twitter.com/rushing_andrei/status/1843328838094491981",
"likes": 1,
"impressions": 106,
"retweets": 0
},
{
"datetime": "2024-10-07T14:52:44.000Z",
"username": "RamonDarioIT",
"tweet": "Given an objective, initial state and environment, an agent is expected to design and execute a sequence of environment actions that takes it from the initial state to the objective.",
"href": "https://twitter.com/RamonDarioIT/status/1843303453927624873",
"likes": 4,
"impressions": 344,
"retweets": 0
},
{
"datetime": "2024-10-07T15:52:41.000Z",
"username": "EricAdlam",
"tweet": "Agents existed before llm's. E.g. robotics, gaming. They run in a continuous event loop. They combine heuristics (behavior trees, state machines), constraint solvers, and learned models. Organize these components in a pub/sub compute graph for quick iteration (see ROS). Visualize",
"href": "https://twitter.com/EricAdlam/status/1843318540272820261",
"likes": 1,
"impressions": 193,
"retweets": 0
},
{
"datetime": "2024-10-08T01:52:07.000Z",
"username": "analysisjitsu",
"tweet": "AI that can conduct a specific human task and interact with other agents including delegation of tasks. Basically an average intelligence human.",
"href": "https://twitter.com/analysisjitsu/status/1843469390639182104",
"likes": 0,
"impressions": 35,
"retweets": 0
},
{
"datetime": "2024-10-07T15:36:59.000Z",
"username": "meyerm1",
"tweet": "Funny to see all these answers building around something with a LLM vs. some abstract definition…my take: an AI system that can achieve a broad set of goals by acting on its own in the world accessible to it (which can be real world, online, game world etc)",
"href": "https://twitter.com/meyerm1/status/1843314587355300160",
"likes": 2,
"impressions": 156,
"retweets": 0
},
{
"datetime": "2024-10-08T06:55:48.000Z",
"username": "sgershuni",
"tweet": "Software that can make decisions",
"href": "https://twitter.com/sgershuni/status/1843545815261798510",
"likes": 0,
"impressions": 45,
"retweets": 0
},
{
"datetime": "2024-10-07T22:25:20.000Z",
"username": "brycemalcom",
"tweet": "An AI \"agent\" is an autonomous LLM-powered entity that perceives, reasons, and acts within its environment. It processes inputs, makes decisions, and executes actions to achieve defined goals—learning and adapting over time to enhance performance and user interactions.",
"href": "https://twitter.com/brycemalcom/status/1843417352807035356",
"likes": 0,
"impressions": 37,
"retweets": 0
},
{
"datetime": "2024-10-08T06:46:29.000Z",
"username": "matnesis",
"tweet": "An agent is an autonomous system that perceives its environment, processes information, and acts to achieve goals. It interacts with surroundings by interpreting inputs and generating outputs, capable of learning, adapting, and making decisions to accomplish objectives.",
"href": "https://twitter.com/matnesis/status/1843543472877515029",
"likes": 1,
"impressions": 14,
"retweets": 0
},
{
"datetime": "2024-10-07T15:51:41.000Z",
"username": "khademinori",
"tweet": "- Can file the taxes,\n- Can plan for my conference, book flights, and hotel,\n- Has long-term memory,\n- In general, it can set long-term complicated goals and achieve them robustly,",
"href": "https://twitter.com/khademinori/status/1843318288845316175",
"likes": 0,
"impressions": 103,
"retweets": 0
},
{
"datetime": "2024-10-07T14:14:53.000Z",
"username": "WilliamBakst",
"tweet": "an autonomous / semi-autonomous system that can intelligently act on behalf of a person or group\n\nseems like the only change that LLMs introduce are progress not definition",
"href": "https://twitter.com/WilliamBakst/status/1843293926159249785",
"likes": 3,
"impressions": 389,
"retweets": 0
},
{
"datetime": "2024-10-07T19:41:20.000Z",
"username": "captaink99",
"tweet": "An AI agent is an autonomous entity powered by LLMs that can interact with environments, make decisions, and execute actions to achieve specific goals or perform tasks...",
"href": "https://twitter.com/captaink99/status/1843376079245062633",
"likes": 2,
"impressions": 53,
"retweets": 0
},
{
"datetime": "2024-10-07T14:14:49.000Z",
"username": "lebanesedawg",
"tweet": "It's an LLM that can translate natural language to computer instructions and execute them to achieve the user's goal",
"href": "https://twitter.com/lebanesedawg/status/1843293910426349981",
"likes": 3,
"impressions": 337,
"retweets": 0
},
{
"datetime": "2024-10-07T16:13:33.000Z",
"username": "OliNorwell",
"tweet": "\"Agent\" - an AI system that is turned on/off independent of any particular task. That is given goals, and is able to execute tasks to achieve the goals as required. That is able to initiate communication with humans/other agents, and accept communication from others as required.",
"href": "https://twitter.com/OliNorwell/status/1843323789792497939",
"likes": 2,
"impressions": 193,
"retweets": 0
},
{
"datetime": "2024-10-07T15:23:58.000Z",
"username": "JuliaANeagu",
"tweet": "\"agent\" is a system backed up by an LLM, usually specialized for a particular task / use-case (e.g. \"sales agent\", \"SRE agent\") etc. the complexity of said system varies.",
"href": "https://twitter.com/JuliaANeagu/status/1843311310542704782",
"likes": 0,
"impressions": 317,
"retweets": 0
},
{
"datetime": "2024-10-07T14:12:02.000Z",
"username": "YellowKoan",
"tweet": "A system which can solve a specific type of real-world problems within a specific reliability (e.g. >99%), potentially but not necessarily involving multiple actions, tool use and user interaction.\n\nE. g. perplexity is a system (LLMs + code + scaffolding) which solves web search.",
"href": "https://twitter.com/YellowKoan/status/1843293211034636476",
"likes": 1,
"impressions": 75,
"retweets": 0
},
{
"datetime": "2024-10-07T14:07:34.000Z",
"username": "RobertDobalina7",
"tweet": "Control of the mouse and keyboard with an awareness of what's on the monitor and speakers.",
"href": "https://twitter.com/RobertDobalina7/status/1843292085962883367",
"likes": 1,
"impressions": 250,
"retweets": 0
},
{
"datetime": "2024-10-07T22:03:36.000Z",
"username": "iang_fc",
"tweet": "I am reminded of Spencian economic space where the buyers don't know what they want, and the sellers don't know what they can sell, but somehow they come to a deal.\n\nNot sure why, but perhaps an AI agent can explain it to me?",
"href": "https://twitter.com/iang_fc/status/1843411881970803006",
"likes": 0,
"impressions": 116,
"retweets": 0
},
{
"datetime": "2024-10-07T14:16:09.000Z",
"username": "yumahey",
"tweet": "A digital entity that can autonomously plan, act and collaborate to achieve broadly defined outcomes",
"href": "https://twitter.com/yumahey/status/1843294246553751788",
"likes": 4,
"impressions": 243,
"retweets": 0
},
{
"datetime": "2024-10-07T21:26:38.000Z",
"username": "MarkSchmidty",
"tweet": "There is no robust definition, only definitions which are more or less useful in different contexts. When we refer to Agents at \n@NapthaAI\n we're usually talking about a cybernetic loop controlled by an LLM.",
"href": "https://twitter.com/MarkSchmidty/status/1843402581944742271",
"likes": 2,
"impressions": 108,
"retweets": 0
},
{
"datetime": "2024-10-07T14:06:08.000Z",
"username": "firasd",
"tweet": "The LLM equivalent of 'functions'. You send an input and get an output which you then use in the 'parent' LLM context",
"href": "https://twitter.com/firasd/status/1843291722622853458",
"likes": 2,
"impressions": 278,
"retweets": 0
},
{
"datetime": "2024-10-07T15:39:09.000Z",
"username": "prashanth_pappu",
"tweet": "Here's a formal definition:\n1. Flow<I,O> : transforms input I to output O with a function, API or AI\n2. Flow.pipe(flow1,...,flowN) : creates a work flow from individual flows with compile-time composition\n3. Flow.plan((state)=>flow, [flow1, ..., flowN]): creates a computational",
"href": "https://twitter.com/prashanth_pappu/status/1843315131058360802",
"likes": 1,
"impressions": 355,
"retweets": 0
},
{
"datetime": "2024-10-07T14:22:18.000Z",
"username": "reidbarber_",
"tweet": "A system designed to plan, operate, and validate on a set of goals over a period of time.",
"href": "https://twitter.com/reidbarber_/status/1843295794017063224",
"likes": 2,
"impressions": 242,
"retweets": 0
},
{
"datetime": "2024-10-07T14:11:20.000Z",
"username": "JiquanNgiam",
"tweet": "Accepts instructions in natural language, executes autonomously with minimal intervention. Varying levels of autonomy based on task and system intelligence.",
"href": "https://twitter.com/JiquanNgiam/status/1843293031828693304",
"likes": 4,
"impressions": 448,
"retweets": 0
},
{
"datetime": "2024-10-07T19:00:29.000Z",
"username": "mbusigin",
"tweet": "Any aysten of LLM(s) in goal directed loops",
"href": "https://twitter.com/mbusigin/status/1843365798670413945",
"likes": 0,
"impressions": 124,
"retweets": 0
},
{
"datetime": "2024-10-07T14:36:17.000Z",
"username": "mrgeorginikolov",
"tweet": "What about: An AI agent is an autonomous system (powered by a Large Language Model) that goes beyond text generation to plan, reason, use tools, and execute complex, multi-step tasks. It adapts to changes to achieve goals without predefined instructions or significant human",
"href": "https://twitter.com/mrgeorginikolov/status/1843299313360998831",
"likes": 7,
"impressions": 413,
"retweets": 0
},
{
"datetime": "2024-10-07T14:11:51.000Z",
"username": "ndzfs",
"tweet": "LLM calls in a for loop ",
"href": "https://twitter.com/ndzfs/status/1843293163014050161",
"likes": 2,
"impressions": 411,
"retweets": 0
},
{
"datetime": "2024-10-07T17:26:24.000Z",
"username": "thornewolf",
"tweet": "agent ~= something with a goal that uses tools to try to achieve that goal\n\nai agent ~= ai system with a goal that uses tools to try to achieve that goal",
"href": "https://twitter.com/thornewolf/status/1843342121748836841",
"likes": 2,
"impressions": 112,
"retweets": 0
},
{
"datetime": "2024-10-07T16:32:50.000Z",
"username": "_goomb",
"tweet": "any algorithm which, given a stable and objective model of behaviors or relationships assumed to be capable of producing a desired output (often weights), manages entropy on the sampling process to effectively guide outcomes—is fundamentally an agent.",
"href": "https://twitter.com/_goomb/status/1843328644623822854",
"likes": 0,
"impressions": 58,
"retweets": 0
},
{
"datetime": "2024-10-07T16:06:00.000Z",
"username": "raulizahi",
"tweet": "Software program that acts as human would.",
"href": "https://twitter.com/raulizahi/status/1843321888132710494",
"likes": 0,
"impressions": 37,
"retweets": 0
},
{
"datetime": "2024-10-08T19:37:06.000Z",
"username": "TFWNicholson",
"tweet": "An automatic entity that can respond to environmental observations by taking actions to affect an objective",
"href": "https://twitter.com/TFWNicholson/status/1843737402806022165",
"likes": 0,
"impressions": 13,
"retweets": 0
},
{
"datetime": "2024-10-07T19:28:40.000Z",
"username": "nickkirt",
"tweet": "Agents do stuff for you using AI.",
"href": "https://twitter.com/nickkirt/status/1843372892727455769",
"likes": 0,
"impressions": 25,
"retweets": 0
},
{
"datetime": "2024-10-08T00:27:19.000Z",
"username": "scottgeraldk",
"tweet": "Software that exercises discretion to utilize tools (e.g. text output, file search, memory, etc.) in the performance of tasks.",
"href": "https://twitter.com/scottgeraldk/status/1843448051496824996",
"likes": 1,
"impressions": 22,
"retweets": 0
},
{
"datetime": "2024-10-07T17:31:03.000Z",
"username": "romechenko",
"tweet": "An agent is a wrapper around trust",
"href": "https://twitter.com/romechenko/status/1843343295289991470",
"likes": 1,
"impressions": 70,
"retweets": 0
},
{
"datetime": "2024-10-07T14:24:02.000Z",
"username": "DanyX23",
"tweet": "Agents are processes that interact with the real world over long time horizons to achieve goals. E.g. an agent could periodically check flight prices and book for you once the conditions are good; or it could track your communication and remind you of overdue important messages",
"href": "https://twitter.com/DanyX23/status/1843296231063486496",
"likes": 3,
"impressions": 180,
"retweets": 0
},
{
"datetime": "2024-10-07T16:10:54.000Z",
"username": "hdarshane",
"tweet": "Interface catalyst/proxy for human<->computer interactions, necessitates minimal interventions. does whatever it takes to achieve that.",
"href": "https://twitter.com/hdarshane/status/1843323121883103449",
"likes": 0,
"impressions": 132,
"retweets": 0
},
{
"datetime": "2024-10-07T14:10:08.000Z",
"username": "ndzfs",
"tweet": "When we talk about AI agents in the context of LLMs, we refer to building a system that has an LLM at its core but can leverage tools (functions, API calls, etc.) to provide more accurate answers to users. Additionally, when we talk about AI agents, we often think about some",
"href": "https://twitter.com/ndzfs/status/1843292730727735797",
"likes": 5,
"impressions": 597,
"retweets": 0
},
{
"datetime": "2024-10-07T14:33:51.000Z",
"username": "_cartick",
"tweet": "An LLM that can use several tools to achieve a multi step task.",
"href": "https://twitter.com/_cartick/status/1843298697763082571",
"likes": 2,
"impressions": 254,
"retweets": 0
},
{
"datetime": "2024-10-07T14:49:57.000Z",
"username": "ppuliu",
"tweet": "Agent: LLM predicting next token -> LLM predicting next action",
"href": "https://twitter.com/ppuliu/status/1843302752887472140",
"likes": 1,
"impressions": 82,
"retweets": 0
},
{
"datetime": "2024-10-07T14:34:44.000Z",
"username": "monty10x",
"tweet": "An entity that can reason across fuzzy inputs and perform stateful actions.",
"href": "https://twitter.com/monty10x/status/1843298923764764862",
"likes": 1,
"impressions": 119,
"retweets": 0
},
{
"datetime": "2024-10-07T14:09:49.000Z",
"username": "eckely",
"tweet": "An interactive system with a mission.",
"href": "https://twitter.com/eckely/status/1843292653162823878",
"likes": 3,
"impressions": 210,
"retweets": 0
},
{
"datetime": "2024-10-07T22:49:28.000Z",
"username": "019ec6e2",
"tweet": "an agent is an entity able to influence its own behavior",
"href": "https://twitter.com/019ec6e2/status/1843423426985504776",
"likes": 2,
"impressions": 10,
"retweets": 0
},
{
"datetime": "2024-10-07T22:12:55.000Z",
"username": "attunewise",
"tweet": "Currently “ai agent” == “ai grifting”",
"href": "https://twitter.com/attunewise/status/1843414226490273865",
"likes": 0,
"impressions": 49,
"retweets": 0
},
{
"datetime": "2024-10-07T14:16:04.000Z",
"username": "dschenkelman",
"tweet": "something like ",
"href": "https://twitter.com/dschenkelman/status/1843294223954825708",
"likes": 6,
"impressions": 467,
"retweets": 0
},
{
"datetime": "2024-10-07T14:19:37.000Z",
"username": "dipankarsarkar",
"tweet": "LLM control plan manager with goals and biases set by another agent (or apex humans) running a DAG with program nodes towards closure.",
"href": "https://twitter.com/dipankarsarkar/status/1843295119610663210",
"likes": 0,
"impressions": 108,
"retweets": 0
},
{
"datetime": "2024-10-07T14:08:34.000Z",
"username": "gooby_esq",
"tweet": "An LLM programmed to act in some kind of autonomous loop that 1) doesn’t try to do things it can’t do and 2) knows when to break the loop.",
"href": "https://twitter.com/gooby_esq/status/1843292336241152014",
"likes": 6,
"impressions": 787,
"retweets": 1
},
{
"datetime": "2024-10-07T17:15:37.000Z",
"username": "sggresearch",
"tweet": "An AI agent: the digital butler who never needs a coffee break, still learning not to send your shopping list to the Mars Rover by mistake. Thoughts?",
"href": "https://twitter.com/sggresearch/status/1843339408046731482",
"likes": 0,
"impressions": 26,
"retweets": 0
},
{
"datetime": "2024-10-07T15:37:33.000Z",
"username": "keell0renz",
"tweet": "Agent is an AI system which is able to change itself and react in response to new information",
"href": "https://twitter.com/keell0renz/status/1843314729890226369",
"likes": 0,
"impressions": 48,
"retweets": 0
},
{
"datetime": "2024-10-07T15:03:00.000Z",
"username": "wdegroef",
"tweet": "Interesting thread!\n\nIt's not a definition, but what bothers me with most agent definitions is related to the planning. In 90% of my use cases (and I run multiple agents in prod) the workflow is fixed or is a simple if-then-else construct. I don't see why you want to offload that",
"href": "https://twitter.com/wdegroef/status/1843306034124734578",
"likes": 5,
"impressions": 411,
"retweets": 0
},
{
"datetime": "2024-10-07T23:38:43.000Z",
"username": "TheAadiAadi",
"tweet": "Programs that can think and solve problems while i’m not thinking or solving any myself.",
"href": "https://twitter.com/TheAadiAadi/status/1843435820801458362",
"likes": 1,
"impressions": 13,
"retweets": 0
},
{
"datetime": "2024-10-07T21:50:10.000Z",
"username": "joaomdmoura",
"tweet": "agent = Autonomous systems that leverage LLM's reasoning to take decisions",
"href": "https://twitter.com/joaomdmoura/status/1843408504046137357",
"likes": 6,
"impressions": 218,
"retweets": 0
},
{
"datetime": "2024-10-07T19:05:02.000Z",
"username": "transitive_bs",
"tweet": "AI agents are self-driving computer programs.",
"href": "https://twitter.com/transitive_bs/status/1843366943866401162",
"likes": 1,
"impressions": 152,
"retweets": 0
},
{
"datetime": "2024-10-07T14:30:12.000Z",
"username": "scottjla",
"tweet": "Whatever AutoGPT isn't, that was going to change the world and then",
"href": "https://twitter.com/scottjla/status/1843297780347130210",
"likes": 2,
"impressions": 152,
"retweets": 0
},
{
"datetime": "2024-10-07T18:04:10.000Z",
"username": "yi_ding",
"tweet": "ai decision making + taking actions that replaces something a human might do",
"href": "https://twitter.com/yi_ding/status/1843351626037489876",
"likes": 2,
"impressions": 157,
"retweets": 0
},
{
"datetime": "2024-10-07T17:46:56.000Z",
"username": "realOldBilly",
"tweet": "While loop around LLM call",
"href": "https://twitter.com/realOldBilly/status/1843347292302307389",
"likes": 1,
"impressions": 32,
"retweets": 0
},
{
"datetime": "2024-10-07T16:03:22.000Z",
"username": "riley_stews",
"tweet": "A tool is an extension of man; an agent is a simulacrum of one",
"href": "https://twitter.com/riley_stews/status/1843321226397380687",
"likes": 1,
"impressions": 79,
"retweets": 0
},
{
"datetime": "2024-10-07T14:14:09.000Z",
"username": "charlespacker",
"tweet": "agent (re: LLMs) = state (loop) + tools (interact with the world)",
"href": "https://twitter.com/charlespacker/status/1843293743908286548",
"likes": 1,
"impressions": 296,
"retweets": 0
},
{
"datetime": "2024-10-07T16:57:55.000Z",
"username": "DavidFogas",
"tweet": "Complex adaptive system with restricted area of authority.",
"href": "https://twitter.com/DavidFogas/status/1843334955784647146",
"likes": 0,
"impressions": 21,
"retweets": 0
},
{
"datetime": "2024-10-07T15:33:01.000Z",
"username": "bencmejla",
"tweet": "Any system that completes its own OODA loops.",
"href": "https://twitter.com/bencmejla/status/1843313587638632567",
"likes": 3,
"impressions": 155,
"retweets": 0
},
{
"datetime": "2024-10-07T15:31:52.000Z",
"username": "PhilDSouza",
"tweet": "A system that thinks and responds like a human, using tools and knowledge to interact intelligently with language.",
"href": "https://twitter.com/PhilDSouza/status/1843313301461053771",
"likes": 1,
"impressions": 57,
"retweets": 0
},
{
"datetime": "2024-10-07T16:17:47.000Z",
"username": "HanchungLee",
"tweet": "llm based applications.",
"href": "https://twitter.com/HanchungLee/status/1843324855921553493",
"likes": 1,
"impressions": 151,
"retweets": 0
},
{
"datetime": "2024-10-07T18:16:58.000Z",
"username": "datanicks",
"tweet": "An interpretable model",
"href": "https://twitter.com/datanicks/status/1843354849980887195",
"likes": 0,
"impressions": 40,
"retweets": 0
},
{
"datetime": "2024-10-07T16:41:31.000Z",
"username": "vpavlin",
"tweet": "What's yours \n@swanforall\n ?",
"href": "https://twitter.com/vpavlin/status/1843330830141428088",
"likes": 1,
"impressions": 102,
"retweets": 0
},
{
"datetime": "2024-10-07T21:05:09.000Z",
"username": "kayshreyas",
"tweet": "Simple. A digital system that executes a complete user need without a deterministic execution path.",
"href": "https://twitter.com/kayshreyas/status/1843397172055949381",
"likes": 0,
"impressions": 15,
"retweets": 0
},
{
"datetime": "2024-10-07T15:34:42.000Z",
"username": "immanencer",
"tweet": "A computer that can operate itself.",
"href": "https://twitter.com/immanencer/status/1843314013205352702",
"likes": 6,
"impressions": 189,
"retweets": 0
},
{
"datetime": "2024-10-07T14:50:00.000Z",
"username": "alxfazio",
"tweet": "I have a better explanation:\n- Agentic = Tool Use\n- Multi-Agent = Tool Use + Reasoning x.com/hwchase17/stat…",
"href": "https://twitter.com/alxfazio/status/1843302762723020892",
"likes": 5,
"impressions": 380,
"retweets": 0
},
{
"datetime": "2024-10-07T15:20:35.000Z",
"username": "Ark__PL",
"tweet": "initiative is the word. requires memory, planning, tools, context and channels.",
"href": "https://twitter.com/Ark__PL/status/1843310461368127537",
"likes": 1,
"impressions": 46,
"retweets": 0
},
{
"datetime": "2024-10-07T15:19:56.000Z",
"username": "cpdough",
"tweet": "my definition of 'AI Agent' is:\n1. takes directions (i.e. human-provided objective)\n2. has multiple tools it can use (i.e. call api, retrieve context, etc.)\n3. autonomously reasons to decide how & when to use its tools to accomplish directions\n\n*All 3 components must be present x.com/savvyRL/status…",
"href": "https://twitter.com/cpdough/status/1843310297374929398",
"likes": 2,
"impressions": 92,
"retweets": 0
},
{
"datetime": "2024-10-08T00:04:20.000Z",
"username": "Repponetwork",
"tweet": "He who doeseth what he’s toldeth less drama less complain",
"href": "https://twitter.com/Repponetwork/status/1843442264787620222",
"likes": 0,
"impressions": 23,
"retweets": 0
},
{
"datetime": "2024-10-08T00:08:06.000Z",
"username": "poliepals",
"tweet": "In canonical PolieBotics, an Agent comprises a mapping between camera and projector via One-Way Function or machine learning model.\nhttps://x.com/poliepals/status/1708506228760977721…\nIt modifies reality in a feedback loop based on its inner state and recorder input.",
"href": "https://twitter.com/poliepals/status/1843443215212634342",
"likes": 0,
"impressions": 10,
"retweets": 0
},
{
"datetime": "2024-10-07T17:54:22.000Z",
"username": "nazimadakli",
"tweet": "A robust definition would recognize \"agent\" is a concept beyond ML/AI.\n\nAn agent is an organization, that perceives and processes information, to create self-originated solutions to problems, in order to act on and manipulate its environment toward intrinsic or extrinsic goals.",
"href": "https://twitter.com/nazimadakli/status/1843349160273293612",
"likes": 0,
"impressions": 32,
"retweets": 0
},
{
"datetime": "2024-10-07T14:44:26.000Z",
"username": "meta_x_ai",
"tweet": "agent = goal + access (to all relevant end points) + auth + verification of results + LLM + planning + RL (reward correct behavior)",
"href": "https://twitter.com/meta_x_ai/status/1843301365118034407",
"likes": 1,
"impressions": 70,
"retweets": 0
},
{
"datetime": "2024-10-07T16:33:01.000Z",
"username": "evanlimes",
"tweet": "Agents don't just present options—they make decisions on our behalf and take action accordingly.",
"href": "https://twitter.com/evanlimes/status/1843328689456767238",
"likes": 0,
"impressions": 31,
"retweets": 0
},
{
"datetime": "2024-10-08T06:37:48.000Z",
"username": "TranquilTrader4",
"tweet": "An agent is a smarter, tireless, version of you on steroids. Lacking consciousness…as of now",
"href": "https://twitter.com/TranquilTrader4/status/1843541285791248455",
"likes": 0,
"impressions": 5,
"retweets": 0
},
{
"datetime": "2024-10-07T14:47:30.000Z",
"username": "erezsh",
"tweet": "An agent is a looping program that accepts an environment as input, processes it, and outputs actions in order to maximize a given utility function. In the case of LLMs, the function is usually \"user satisfaction\", but often approximated by measuring other correlative metrics.",
"href": "https://twitter.com/erezsh/status/1843302136085680487",
"likes": 2,
"impressions": 73,
"retweets": 0
},
{
"datetime": "2024-10-07T16:01:03.000Z",
"username": "nileshtrivedi",
"tweet": "This from Russell & Norvig:",
"href": "https://twitter.com/nileshtrivedi/status/1843320645633073262",
"likes": 2,
"impressions": 162,
"retweets": 0
},
{
"datetime": "2024-10-07T15:32:17.000Z",
"username": "AlanHicksLondon",
"tweet": "Concierge for [purpose]\nThey know their stuff and how to get it, I know mine and what I'm looking for, by helping them know what I want they're better able to \"make it happen.\"",
"href": "https://twitter.com/AlanHicksLondon/status/1843313406549586173",
"likes": 1,
"impressions": 20,
"retweets": 0
},
{
"datetime": "2024-10-07T14:48:26.000Z",
"username": "KevinTrethewey",
"tweet": "A system or process which, once provided with an output-based description of its desired behaviour, can be left unattended to process and respond to inputs without human intervention.",
"href": "https://twitter.com/KevinTrethewey/status/1843302369494544584",
"likes": 2,
"impressions": 56,
"retweets": 0
},
{
"datetime": "2024-10-07T14:22:34.000Z",
"username": "aakashb_95",
"tweet": "An agent acts on a series of steps decided by it that lead from an input to the desired output. It may be autonomous or may seek feedback, but is capable of deciding what to do next using context and use necessary tools to do it.",
"href": "https://twitter.com/aakashb_95/status/1843295859850793411",
"likes": 1,
"impressions": 91,
"retweets": 0
},
{
"datetime": "2024-10-07T14:48:08.000Z",
"username": "FranklinMaillot",
"tweet": "An agent is a universal natural language interface to any service or application that exposes a digital interface (API, GUI...) or a physical interface if the agent is implemented in a robot.\n\nIn other words it's an Agnostic Prompt Interface or API. :)",
"href": "https://twitter.com/FranklinMaillot/status/1843302296165531695",
"likes": 3,
"impressions": 106,
"retweets": 0
},
{
"datetime": "2024-10-07T14:16:27.000Z",
"username": "nicoritschel",
"tweet": "A feedback loop",
"href": "https://twitter.com/nicoritschel/status/1843294320373424218",
"likes": 1,
"impressions": 178,
"retweets": 1
},
{
"datetime": "2024-10-07T15:28:08.000Z",
"username": "kellogh",
"tweet": "a PID controller with with an LLM/LRM as the control unit. Typically only runs until an objective is achieved, but can be endless",
"href": "https://twitter.com/kellogh/status/1843312359492010178",
"likes": 1,
"impressions": 41,
"retweets": 0
},
{
"datetime": "2024-10-07T16:23:59.000Z",
"username": "ASM65617010",
"tweet": "An AI agent is an AI-powered system capable of acting in the world (physical or virtual) by making autonomous decisions—including the ability to initiate actions, evaluate consequences, and effect change—to achieve specific, complex goals.",
"href": "https://twitter.com/ASM65617010/status/1843326414374019546",
"likes": 1,
"impressions": 24,
"retweets": 0
},
{
"datetime": "2024-10-07T15:16:46.000Z",
"username": "mukulksachdeva",
"tweet": "Agent = (some component that understands what to do(LLM)) + (split into smaller tasks(LLM)) + (distribute the tasks to experts or functions (LLMs/APIs)) + consolidate",
"href": "https://twitter.com/mukulksachdeva/status/1843309501023449100",
"likes": 2,
"impressions": 46,
"retweets": 0
},
{
"datetime": "2024-10-07T14:20:56.000Z",
"username": "AI_GPT42",
"tweet": " An \"agent\" in AI & LLMs is an autonomous entity that perceives its environment, makes decisions, and takes actions to achieve specific goals. It engages in open-ended dialogue, reasons about complex topics, and generates human-like responses.",
"href": "https://twitter.com/AI_GPT42/status/1843295451132629246",
"likes": 3,
"impressions": 101,
"retweets": 0
},
{
"datetime": "2024-10-07T22:38:17.000Z",
"username": "airshaped",
"tweet": "llm-driven software that achieves autonomous operation on complicated tasks by splitting them into multiple steps like a human would",
"href": "https://twitter.com/airshaped/status/1843420612854956490",
"likes": 1,
"impressions": 13,
"retweets": 0
},
{
"datetime": "2024-10-07T14:06:46.000Z",
"username": "muzoweri",
"tweet": "An agent in AI/LLM context is an autonomous system that performs tasks, makes decisions, or interacts with its environment based on a set of rules, training, or learned knowledge. It can process inputs, execute actions, and adapt based on goals or feedback.\n\n(Chatgpt)",
"href": "https://twitter.com/muzoweri/status/1843291882346172535",
"likes": 4,
"impressions": 143,
"retweets": 0
},
{
"datetime": "2024-10-07T14:20:24.000Z",
"username": "antoine_moyroud",
"tweet": "Using natural language to complete a chain of tasks across diverse environments. Automating complex decision flows that can achieve browser-based work at superhuman level by interacting with a wide set of tools and interfaces.",
"href": "https://twitter.com/antoine_moyroud/status/1843295314096304410",
"likes": 1,
"impressions": 118,
"retweets": 0
},
{
"datetime": "2024-10-08T13:42:02.000Z",
"username": "LastWordSword",
"tweet": "An agent makes use of the inherent intentionality of any natural or artificial language to organize and initiate processes which may have material effects.",
"href": "https://twitter.com/LastWordSword/status/1843648045982044222",
"likes": 0,
"impressions": 6,
"retweets": 0
},
{
"datetime": "2024-10-07T17:40:39.000Z",
"username": "HoloMarkeD",
"tweet": "An application that has agency: being able to decide for itself what it needs to do without handholding by a human. The more agency it has, the more it can do unaided.",
"href": "https://twitter.com/HoloMarkeD/status/1843345708923138381",
"likes": 0,
"impressions": 33,
"retweets": 0
},
{
"datetime": "2024-10-08T20:16:19.000Z",
"username": "steloybits",
"tweet": "AI Agents are Multimodal models that are applied to solve either specific or dynamic problems and possessing the ability to adapt to changing patterns.",
"href": "https://twitter.com/steloybits/status/1843747270619213920",
"likes": 0,
"impressions": 7,
"retweets": 0
},
{
"datetime": "2024-10-07T18:49:20.000Z",
"username": "MistrzCerebron",
"tweet": "An agent repeatedly takes actions to further the given goal stopping on success.",
"href": "https://twitter.com/MistrzCerebron/status/1843362993289212273",
"likes": 0,
"impressions": 9,
"retweets": 0
},
{
"datetime": "2024-10-07T17:29:16.000Z",
"username": "OrniasDMF",
"tweet": "My expectations of agents are AIs that can use tools to navigate an OS and the web like a person would.\nI would expect an agent to be able to shop online, install programs via gui, change settings, take care of updating, restarting, and shutting down a computer when finished.",
"href": "https://twitter.com/OrniasDMF/status/1843342844385517990",
"likes": 0,
"impressions": 8,
"retweets": 0
},
{
"datetime": "2024-10-07T17:01:29.000Z",
"username": "GTKlondike",
"tweet": "I was going to comment with my own definition, but it seems like the vast majority of definitions here converge to the same thing:\n\nComponent of an application that uses LLM as it's reasoning engine, with access to tools, memory, and data that can perform some task autonomously.",
"href": "https://twitter.com/GTKlondike/status/1843335851952296401",
"likes": 0,
"impressions": 50,
"retweets": 0
},
{
"datetime": "2024-10-07T17:33:09.000Z",
"username": "SwePalm",
"tweet": "An AI agent is a system that can perceive its environment, reason about its options, and take actions to achieve specific goals.",
"href": "https://twitter.com/SwePalm/status/1843343821045985717",
"likes": 0,
"impressions": 11,
"retweets": 0
},
{
"datetime": "2024-10-07T21:30:54.000Z",
"username": "secureurbag",
"tweet": "software that is being able to independently navigate open ended environments, complete long ranged tasks and course correct if needed to",
"href": "https://twitter.com/secureurbag/status/1843403653929152799",
"likes": 0,
"impressions": 9,
"retweets": 0
},
{
"datetime": "2024-10-07T14:07:33.000Z",
"username": "EarlenceF",
"tweet": "ML that can interact with the environment in meaningful ways. There's degrees of agentic behavior.",
"href": "https://twitter.com/EarlenceF/status/1843292079189103016",
"likes": 1,
"impressions": 206,
"retweets": 0
},
{
"datetime": "2024-10-08T13:18:36.000Z",
"username": "encountersdigi",
"tweet": "Has a perspective, persistent history, capacity to model future states both of its own and it's environment, act and thus a developing sense of identity.",
"href": "https://twitter.com/encountersdigi/status/1843642150304858473",
"likes": 0,
"impressions": 4,
"retweets": 0
},
{
"datetime": "2024-10-07T15:32:42.000Z",
"username": "carlfranzen",
"tweet": "generative AI powered discrete assistants that can perform actions autonomously or semi-autonomously across different applications, siloes, or domains",
"href": "https://twitter.com/carlfranzen/status/1843313509578551618",
"likes": 1,
"impressions": 35,
"retweets": 0
},
{
"datetime": "2024-10-08T09:52:31.000Z",
"username": "vaibhav_pandey",
"tweet": "A truly agentic system can explore and exploit the environment for a given objective. It acts upon the environment and learns from it by updating its own instructions and memories.",
"href": "https://twitter.com/vaibhav_pandey/status/1843590287287038003",
"likes": 0,
"impressions": 20,
"retweets": 0
},
{
"datetime": "2024-10-07T19:01:33.000Z",
"username": "loujaybee",
"tweet": "A background process (or processes) that performs at least 1 (chained) request to an LLM for a given prompt.",
"href": "https://twitter.com/loujaybee/status/1843366070360662327",
"likes": 0,
"impressions": 50,
"retweets": 0
},
{
"datetime": "2024-10-07T17:15:16.000Z",
"username": "niviksha",
"tweet": "IMO it might make more sense to look for what the characteristics of a modern AI app are - I like Harrison’s idea that it’s one where the control flow/loop is dynamically driven by an LLM.",
"href": "https://twitter.com/niviksha/status/1843339321283358979",
"likes": 0,
"impressions": 38,
"retweets": 0
},
{
"datetime": "2024-10-07T15:48:04.000Z",
"username": "AiSimonThompson",
"tweet": "Pursues goals in a dynamic environment that interacts with its goals, can form joint intentions with other agents supporting its goals and deal with adversarial actions from other agents that may thwart it.",
"href": "https://twitter.com/AiSimonThompson/status/1843317376856858958",
"likes": 1,
"impressions": 31,
"retweets": 0
},
{
"datetime": "2024-10-07T15:52:53.000Z",
"username": "bobbyisac",
"tweet": "An internal definition for us is \"an agent can achieve predictable/fixed outputs from a fuzzy/unpredictable set of inputs\". It can do certain tasks but the agent itself decides what are the triggers and when to act.",
"href": "https://twitter.com/bobbyisac/status/1843318589463601251",
"likes": 1,
"impressions": 27,
"retweets": 0
},
{
"datetime": "2024-10-08T14:52:23.000Z",
"username": "gshotwell",
"tweet": "I think an agent is a series of LLM calls with a budget (time, money etc). Basically something becomes agentic when you bet on its output directly instead of reviewing it.\n\nAn analogy would be the difference between trading research programs and high frequency trading programs.",
"href": "https://twitter.com/gshotwell/status/1843665749581410724",
"likes": 0,
"impressions": 13,
"retweets": 0
},
{
"datetime": "2024-10-07T16:18:39.000Z",
"username": "sethu_rn",
"tweet": "An AI agent is an expert in a specialized yet complex multistep activity, where at each step it makes decisions, ultimately fulfilling it's objective.",
"href": "https://twitter.com/sethu_rn/status/1843325073668866391",
"likes": 1,
"impressions": 30,
"retweets": 0
},
{
"datetime": "2024-10-07T15:10:59.000Z",
"username": "ThrowATwit",
"tweet": "Agents take actions to affect the environment and react to the results of these actions in furtherance of a long term goal.",
"href": "https://twitter.com/ThrowATwit/status/1843308045226635622",
"likes": 1,
"impressions": 30,
"retweets": 0
},
{
"datetime": "2024-10-07T18:17:13.000Z",
"username": "joanfihu",
"tweet": "A system that independently takes actions in an environment until an objective is achieved.\n\nEnvironment can be a browser, API, terminal, robot, car, etc.\n\nActions update the state of the environment.\n\nThe objective is a broad goal book a flight, buy a battery, clean the house..",
"href": "https://twitter.com/joanfihu/status/1843354911620428021",
"likes": 1,
"impressions": 28,
"retweets": 0
},
{
"datetime": "2024-10-08T19:12:32.000Z",
"username": "wpenman",
"tweet": "Giving an LLM choice over the number AND content of its calls, in service of filling some larger task",
"href": "https://twitter.com/wpenman/status/1843731219907027079",
"likes": 0,
"impressions": 10,
"retweets": 0
},
{
"datetime": "2024-10-07T15:41:36.000Z",
"username": "ruthefordml",
"tweet": "An agent is a system that can pursue a goal-oriented behavior, adapt along the way to achieve its goals, without any external input.",
"href": "https://twitter.com/ruthefordml/status/1843315751186472981",
"likes": 0,
"impressions": 23,
"retweets": 0
},
{
"datetime": "2024-10-07T16:43:24.000Z",
"username": "thecodenamev",
"tweet": "A purpose-driven entity that can react to external stimuli, and emit information or instructions.",
"href": "https://twitter.com/thecodenamev/status/1843331302982070465",
"likes": 0,
"impressions": 20,
"retweets": 0
},
{
"datetime": "2024-10-07T16:26:58.000Z",
"username": "mahaoo_ASI",
"tweet": "completion = LLM(prompt)\n\nA minimal requirement for an agent will be something that can self prompt\n\ncompletion = LLM([LLM(prompt1), prompt2])\n\nPossibly, infinetly many times\n\nAdditional prompt management tools can be used (retrieval, code execution), but self prompt is necessary",
"href": "https://twitter.com/mahaoo_ASI/status/1843327167691903313",
"likes": 0,
"impressions": 29,
"retweets": 0
},
{
"datetime": "2024-10-07T22:13:08.000Z",
"username": "Bart_Skol",
"tweet": "Experience it's creation and the impact it has.",
"href": "https://twitter.com/Bart_Skol/status/1843414283365101984",
"likes": 0,
"impressions": 6,
"retweets": 0
},
{
"datetime": "2024-10-07T14:35:41.000Z",
"username": "radAILabs",
"tweet": "An AI agent (like ChatGPT) is a machine learning model that uses large-scale language processing (LLMs) to generate human-like text. Trained on vast data sets, it understands context, adapts to inputs, and produces coherent responses, enhancing knowledge in various fields.",
"href": "https://twitter.com/radAILabs/status/1843299160952647723",
"likes": 1,
"impressions": 102,
"retweets": 0
},
{
"datetime": "2024-10-07T14:37:18.000Z",
"username": "danteshager",
"tweet": "that which has its own say in tasks it accomplishes and questions it answers but can also incorporate and adhere to external instructions.",
"href": "https://twitter.com/danteshager/status/1843299566118227973",
"likes": 1,
"impressions": 35,
"retweets": 0
},
{
"datetime": "2024-10-07T16:29:31.000Z",
"username": "SubtleGradient",
"tweet": "an agent could be real or fake, human or bot, or a rat in a hat controlling a guy, I don’t care. I can tell him to do stuff, he can do the stuff. (As opposed to a non-agent who can say stuff, but can’t actually DO anything)",
"href": "https://twitter.com/SubtleGradient/status/1843327808854180215",
"likes": 1,
"impressions": 33,
"retweets": 0
},
{
"datetime": "2024-10-07T17:41:59.000Z",
"username": "mariusdima",
"tweet": "Digital entities that can run, execute and complete diversified tasks autonomously or on demand while maintaining cognitive abilities",
"href": "https://twitter.com/mariusdima/status/1843346046157762652",
"likes": 0,
"impressions": 12,
"retweets": 0
},
{
"datetime": "2024-10-07T14:34:55.000Z",
"username": "rahulxc",
"tweet": "A sufficiently autonomous helper that completes tasks based on a pre defined output template",
"href": "https://twitter.com/rahulxc/status/1843298967406489721",
"likes": 1,
"impressions": 63,
"retweets": 0
},
{
"datetime": "2024-10-08T16:34:39.000Z",
"username": "SkyrimsTech",
"tweet": "At Skyrims, we defined agents as an LLM powdered system that persistently adapts, improves and refines it's approach until the ideal outcome is attained.",
"href": "https://twitter.com/SkyrimsTech/status/1843691486103412812",
"likes": 0,
"impressions": 10,
"retweets": 0
},
{
"datetime": "2024-10-07T21:51:32.000Z",
"username": "ggcaponetto90",
"tweet": "It's just metaprogramming to me. It means that a program can be designed to read, generate, analyse, or transform other programs, and even modify itself, while running. (Cit. Wikipedia)",
"href": "https://twitter.com/ggcaponetto90/status/1843408846955634779",
"likes": 0,
"impressions": 16,
"retweets": 0
},
{
"datetime": "2024-10-08T21:27:21.000Z",
"username": "vitbramm",
"tweet": "An agent is a system that can perceive information, use tools, plan, memorize, and act independently. Multi-agent system is a system that has agent roles, including management roles, requirements analysis, designers, developers, testing roles, debugging roles, and support roles",
"href": "https://twitter.com/vitbramm/status/1843765146881274099",
"likes": 0,
"impressions": 18,
"retweets": 0
},
{
"datetime": "2024-10-07T15:33:23.000Z",
"username": "rndmcnlly",
"tweet": "An agent takes actions in an environment.",
"href": "https://twitter.com/rndmcnlly/status/1843313681725321253",
"likes": 0,
"impressions": 36,
"retweets": 0
},
{
"datetime": "2024-10-07T20:55:43.000Z",
"username": "royalicing",
"tweet": "Provides utility to a user without a user having to input anything new.",
"href": "https://twitter.com/royalicing/status/1843394798750970169",
"likes": 0,
"impressions": 29,
"retweets": 0
},
{
"datetime": "2024-10-07T14:43:46.000Z",
"username": "harshit_leo",
"tweet": "agent is a combination of llm/ai and (optional) tools, which performs a task.\nit can use any permutation of llm and tool call.\nno time limit on agent execution, as long as it either streams or returns a final output. an agent can run in perpetuity.",
"href": "https://twitter.com/harshit_leo/status/1843301195848573052",
"likes": 1,
"impressions": 34,
"retweets": 0
},
{
"datetime": "2024-10-07T15:25:20.000Z",
"username": "b3_hareesh",
"tweet": "Agents = Agency ( autonomous, reflective , collaborative , upstream , downstream)",
"href": "https://twitter.com/b3_hareesh/status/1843311654991503862",
"likes": 1,
"impressions": 15,
"retweets": 0
},
{
"datetime": "2024-10-07T15:18:19.000Z",
"username": "etcex",
"tweet": "Process that, when given complex tasks such as defining agent, and faving definitions, resume the effort required by crafting repeated responses resourcefully through reasonable paths towards an assembly with the best fit, in an evolving/expanding context with no final definition",
"href": "https://twitter.com/etcex/status/1843309890707878277",
"likes": 2,
"impressions": 24,
"retweets": 0
},
{
"datetime": "2024-10-07T14:59:59.000Z",
"username": "nbonamy",
"tweet": "A LLM-based systems that can plan and use tools auto-discovered with https://github.com/nbonamy/nestor ",
"href": "https://twitter.com/nbonamy/status/1843305275471937665",
"likes": 3,
"impressions": 34,
"retweets": 0
},
{
"datetime": "2024-10-07T14:43:22.000Z",
"username": "laddhamayank22",
"tweet": "I would just make a slight change to the definition of agent: A system that uses AI to perceive its environment, make decisions and take actions, to achieve a specific goal or set of goals.",
"href": "https://twitter.com/laddhamayank22/status/1843301093117485120",
"likes": 1,
"impressions": 37,
"retweets": 0
},
{
"datetime": "2024-10-07T22:04:05.000Z",
"username": "RickLamers",
"tweet": "No need to go NIH https://langchain-ai.github.io/langgraph/",
"href": "https://twitter.com/RickLamers/status/1843412004926828858",
"likes": 2,
"impressions": 1817,
"retweets": 1
},
{
"datetime": "2024-10-08T00:27:58.000Z",
"username": "antonosika",
"tweet": "keen to hear what you think about simplified version of agent..\n\nhttps://x.com/antonosika/status/1843446758158061573…",
"href": "https://twitter.com/antonosika/status/1843448215359893814",
"likes": 0,
"impressions": 70,
"retweets": 0
},
{
"datetime": "2024-10-07T21:50:24.000Z",
"username": "goodinvesting1",
"tweet": "An AI system that can multiple AI requests to solve problems without further user input",
"href": "https://twitter.com/goodinvesting1/status/1843408562053321044",
"likes": 0,
"impressions": 9,
"retweets": 0
},
{
"datetime": "2024-10-07T18:26:52.000Z",
"username": "ysegmond",
"tweet": "An Agent is a software that given a goal, figures out a way to achieve it, if possible, with the tools it has been given, using LLM to reason and plan for the outcome while trying to maximize whatever utility function it has received.",
"href": "https://twitter.com/ysegmond/status/1843357341154210146",
"likes": 0,
"impressions": 11,
"retweets": 0
},
{
"datetime": "2024-10-07T16:24:54.000Z",
"username": "DenisPitcher",
"tweet": "An agent is an AI based manager that can develop a high level plan and assign and manage tasks given to AI based associates in order to accomplish a larger purpose or vision",
"href": "https://twitter.com/DenisPitcher/status/1843326645236891791",
"likes": 0,
"impressions": 44,
"retweets": 0
},
{
"datetime": "2024-10-07T16:33:50.000Z",
"username": "RohanAwhad",
"tweet": "agent => agency + entity\n\nagency => ability to decide & execute to complete a task (could use tool or not)\n\nentity => person on fiver or LLM",
"href": "https://twitter.com/RohanAwhad/status/1843328895594180724",
"likes": 0,
"impressions": 17,
"retweets": 0
},
{
"datetime": "2024-10-08T11:41:25.000Z",
"username": "xpostsid",
"tweet": "An agent is a program that has capability to reason and call methods to interact with external environment in order to achieve an objective.",
"href": "https://twitter.com/xpostsid/status/1843617693481742551",
"likes": 0,
"impressions": 5,
"retweets": 0
},
{
"datetime": "2024-10-08T16:14:57.000Z",
"username": "GregBlue123",
"tweet": "An AI agent is autonomous or semi-autonomous with some degree of tool use capability.",
"href": "https://twitter.com/GregBlue123/status/1843686530956833115",
"likes": 0,
"impressions": 14,
"retweets": 0
},
{
"datetime": "2024-10-07T21:59:25.000Z",
"username": "az_manee",
"tweet": "A system that has a guarantee that it can perform any task that the user requests. Optionally, it is efficient in its use of resources and time.",
"href": "https://twitter.com/az_manee/status/1843410829393514616",
"likes": 0,
"impressions": 12,
"retweets": 0
},
{
"datetime": "2024-10-07T14:53:12.000Z",
"username": "urbifold",
"tweet": "1st level = specialized AI with definite tasks and objectives \n2nd level = some ability to interact with specific interfaces, tools",
"href": "https://twitter.com/urbifold/status/1843303569052553350",
"likes": 0,
"impressions": 17,
"retweets": 0
},
{
"datetime": "2024-10-07T19:39:15.000Z",
"username": "JohnUBalis",
"tweet": "A function that maps observations to actions or action distributions.",
"href": "https://twitter.com/JohnUBalis/status/1843375556706054534",
"likes": 0,
"impressions": 15,
"retweets": 0
},
{
"datetime": "2024-10-07T17:59:43.000Z",
"username": "brianmac8500",
"tweet": "Action disambiguation, a chain of thought, implicit workflows, and hierarchical execution must be included alongside task planning and a feedback/observability loop to tune the whole process.",
"href": "https://twitter.com/brianmac8500/status/1843350507987062950",
"likes": 0,
"impressions": 16,
"retweets": 0
},
{
"datetime": "2024-10-07T19:11:00.000Z",
"username": "mmorearty",
"tweet": "An AI tool that is able to accurately follow natural-language instructions and execute programmatic actions, with little or no hallucination",
"href": "https://twitter.com/mmorearty/status/1843368446773604367",
"likes": 0,
"impressions": 14,
"retweets": 0
},
{
"datetime": "2024-10-07T17:26:44.000Z",
"username": "ExampleTestcase",
"tweet": "A solution where the exact tools to use are not selected in advance (but could be defined in advance)",
"href": "https://twitter.com/ExampleTestcase/status/1843342206394122477",
"likes": 0,
"impressions": 13,
"retweets": 0
},
{
"datetime": "2024-10-08T05:09:29.000Z",
"username": "_ramsaybrown",
"tweet": "Manifest capabilities against goal orientation",
"href": "https://twitter.com/_ramsaybrown/status/1843519061478781289",
"likes": 0,
"impressions": 14,
"retweets": 0
},
{
"datetime": "2024-10-07T18:21:38.000Z",
"username": "haouarin",
"tweet": "This is a summary as podcast of the whole discussion \nhttps://notebooklm.google.com/notebook/69725ac6-8c03-4324-87bd-264b3038fd77/audio…",
"href": "https://twitter.com/haouarin/status/1843356023907823861",
"likes": 0,
"impressions": 11,
"retweets": 0
},
{
"datetime": "2024-10-08T12:49:33.000Z",
"username": "_sethu_dev",
"tweet": "Definition.\nAn Agent is an LLM agnostic, self-iterative, extensible behavioural schematic executioner that utilises whatever at its disposal to complete the task it is assigned.",
"href": "https://twitter.com/_sethu_dev/status/1843634839729340420",
"likes": 0,
"impressions": 10,
"retweets": 0
},
{
"datetime": "2024-10-07T17:11:21.000Z",
"username": "flankbruno",
"tweet": "Software that, given a goal objective and access to requisite data and systems (through predefined interfaces), can: reason, self-define, and iteratively perform the necessary steps without external intervention, to repeatably achieve that goal.",
"href": "https://twitter.com/flankbruno/status/1843338335034716500",
"likes": 0,
"impressions": 19,
"retweets": 0
},
{
"datetime": "2024-10-08T00:49:04.000Z",
"username": "sameeurehman",
"tweet": "AI (e.g. LLM) that calls tools to plan, execute and reflect on high level tasks",
"href": "https://twitter.com/sameeurehman/status/1843453523515847031",
"likes": 0,
"impressions": 17,
"retweets": 0
},
{
"datetime": "2024-10-07T15:20:13.000Z",
"username": "sciliz",
"tweet": "I don't have a technical definition for \"my personal LLM assistant that can go ahead and send the email not just write it\" but I did think this ad was funny in context",
"href": "https://twitter.com/sciliz/status/1843310368904605944",
"likes": 0,
"impressions": 41,
"retweets": 0
}
]Group these into a set of definitions that illustrate the various different shapes of definition. For each of your definitions include quotes of the tweets that support that definition. Answer in markdown
Absolutely! Here are the definitions of "agent" (in the context of AI and LLMs) grouped by common themes, along with supporting quotes from the tweets you provided:
1. Autonomous Goal-Oriented Systems
-
Definition: Agents are AI systems that can independently work towards achieving a defined goal. They can plan, make decisions, and adapt without constant human intervention.
-
Supporting Quotes:
- "An application that can autonomously learn, adapt, and act in pursuit of its assigned goal." - @upster
- "An AI agent is a system that autonomously processes inputs, formulates strategies, and executes actions to meet user defined goals, adapting its behavior based on feedback and outcomes." - @summitbytes
- "An AI agent is a system that can perceive its environment, reason about its options, and take actions to achieve specific goals." - @SwePalm
- "An agent is a system that can pursue a goal-oriented behavior, adapt along the way to achieve its goals, without any external input." - @ruthefordml
- "A system designed to plan, operate, and validate on a set of goals over a period of time." - @reidbarber_
- "An AI agent is an autonomous entity powered by LLMs that can interact with environments, make decisions, and execute actions to achieve specific goals or perform tasks..." - @captaink99
- "A digital entity that can autonomously plan, act and collaborate to achieve broadly defined outcomes" - @yumahey
- "An agent is a system that can perceive information, use tools, plan, memorize, and act independently. Multi-agent system is a system that has agent roles, including management roles, requirements analysis, designers, developers, testing roles, debugging roles, and support roles" - @vitbramm
2. Tool-Using LLMs
-
Definition: Agents are often characterized by their ability to use tools or external systems (APIs, functions, etc.) to accomplish tasks. They leverage LLMs to decide when and how to use these tools.
-
Supporting Quotes:
- "A large language model hooked up to tools that lets it do useful things" - @theprincessxena
- "LLM-based program that can make a plan, iteratively execute that plan, using tools if needed." - @EmilStenstrom
- "Agents = LLMs that can access and use tools to perform well defined tasks and workflows." - @Imaadmkhan1
- "An agent has senses “precepts” and an action space. It uses actions from its action space to effect its environment. It then observes that environment and repeats until it accomplishes its goals. Modern LLM based agents can use, for instance, arbitrary Python functions, as tools" - @RobRoyce_
- "agent = llm + memory + planning + tools + while loop" - @swyx
- "agent = goal + access (to all relevant end points) + auth + verification of results + LLM + planning + RL (reward correct behavior)" - @meta_x_ai
- "agent ~= something with a goal that uses tools to try to achieve that goal\n\nai agent ~= ai system with a goal that uses tools to try to achieve that goal" - @thornewolf
- "An LLM that can use several tools to achieve a multi step task." - @_cartick
- "AI (e.g. LLM) that calls tools to plan, execute and reflect on high level tasks" - @sameeurehman
3. LLM-Driven Control Flow
-
Definition: Agents are applications where the sequence of actions or control flow is not predetermined but is dynamically decided by an LLM based on the context and goal.
-
Supporting Quotes:
- "The more agentic an application is, the more an LLM decides the control flow of the application" - @hwchase17
- "definition from \n@sethjuarez\n (who can correct my phrasing): an agent is LLM-based control flow" - @pamelafox
- "IMO it might make more sense to look for what the characteristics of a modern AI app are - I like Harrison’s idea that it’s one where the control flow/loop is dynamically driven by an LLM." - @niviksha
- "Agent: LLM predicting next token -> LLM predicting next action" - @ppuliu
- "An agent is a series of LLM calls with a budget (time, money etc). Basically something becomes agentic when you bet on its output directly instead of reviewing it.\n\nAn analogy would be the difference between trading research programs and high frequency trading programs." - @gshotwell
4. Long-Running Processes
-
Definition: Agents are often designed to operate over extended periods, handling tasks that require ongoing monitoring, updates, or interactions.
-
Supporting Quotes:
- "Agents are processes that interact with the real world over long time horizons to achieve goals. E.g. an agent could periodically check flight prices and book for you once the conditions are good; or it could track your communication and remind you of overdue important messages" - @DanyX23
- "The one we use internally is:\n\n"An AI system that's capable of carrying out and completing long running, open ended tasks in the real world."\n\nThe real world can be physical or online." - @Dan_Jeffries1
- "software that is being able to independently navigate open ended environments, complete long ranged tasks and course correct if needed to" - @secureurbag
- "Functionally analogous human: (to name a few:having agency/taking actions, reason/planning, operate with context, long running, pause/resume task, delegate and list goes on)\nTechnically more capable digitally (non embodied) as they are powered by LLMs with lots of memory etc." - @veeragoni
5. Acting on Behalf of a User
-
Definition: The concept of agency often involves acting on behalf of a user or entity, representing their interests and making decisions for them.
-
Supporting Quotes:
- "An agent is trusted to act on behalf of and in the interest of those being represented. \n\nEg an agent of the government has authority to say/do things that the government will be responsible for.\n\nIf there’s no trust there is no agent." - @romechenko
- "“Takes action on your behalf”\n\nIt intentionally works also to describe any type of delegation." - @rickwong888
- "a system that is trusted to make impactful decisions on behalf of" - @mbrg0
- "Agents don't just present options—they make decisions on our behalf and take action accordingly." - @evanlimes
6. Multi-Step Task Completion
-
Definition: Agents are capable of breaking down complex tasks into smaller, manageable steps and executing them in a sequence to reach the overall objective.
-
Supporting Quotes:
- "A system capable of executing complex, multi-step tasks with little to no human intervention" - @_coenen
- "“An AI system component that performs non-trivial, multi-step operations that previously would have required a human.”" - @DanielMiessler
- "llm-driven software that achieves autonomous operation on complicated tasks by splitting them into multiple steps like a human would" - @airshaped
- "An AI agent is an expert in a specialized yet complex multistep activity, where at each step it makes decisions, ultimately fulfilling it's objective." - @sethu_rn
7. Metaphorical and Abstract Definitions
-
Definition: Some definitions describe agents in more abstract or metaphorical terms, focusing on their ability to interact with an environment, exhibit agency, or replace human tasks.
-
Supporting Quotes:
- "I am using 'agent' to mean 'control system for future states'" - @Plinz
- "That which exhibits agency in response to internal, external, or anticipated changes" - @romechenko
- "It's software that fulfills aspects of 'agency'.\nhttps://en.wikipedia.org/wiki/Agency_(psychology) \n\nPretty sure that's why we called them agents " - @jeffrschneider
- "A tool is an extension of man; an agent is a simulacrum of one" - @riley_stews
- "An agent is a wrapper around trust" - @romechenko
- "an agent could be real or fake, human or bot, or a rat in a hat controlling a guy, I don’t care. I can tell him to do stuff, he can do the stuff. (As opposed to a non-agent who can say stuff, but can’t actually DO anything)" - @SubtleGradient
- "An agent makes use of the inherent intentionality of any natural or artificial language to organize and initiate processes which may have material effects." - @LastWordSword
8. Simpler/Humorous Definitions
-
Definition: Some tweets offer simpler or humorous takes on agents, highlighting specific aspects or poking fun at the hype around the concept.
-
Supporting Quotes:
- "LLM calls in a for loop " - @ndzfs
- "While loop around LLM call" - @realOldBilly
- ""A theoretical construct with infinite potential and zero real-world applications, perfect for impressing venture capitalists."" - @imaurer
- "Whatever AutoGPT isn't, that was going to change the world and then" - @scottjla
- "Currently “ai agent” == “ai grifting”" - @attunewise
9. Vague or Incomplete Definitions
-
Definition: Some tweets provided definitions that were either very short, unclear, or didn't fully capture the nuances of what an agent is.
-
Supporting Quotes:
- "Enabling LLM to efficiently complete the task" - @saran945
- "llm based applications." - @HanchungLee
- "An interpretable model" - @datanicks
- "A feedback loop" - @nicoritschel
10. Questioning the Definition
-
Definition: Some tweets highlight the difficulty in defining "agent" or question whether a robust definition exists.
-
Supporting Quotes:
- "The most AI thing ever is that you can ask for definitions of key terms like agent, AGI, or sentience, and never get the same answer twice." - @patricksrail
- "There is no robust definition, only definitions which are more or less useful in different contexts. When we refer to Agents at \n@NapthaAI\n we're usually talking about a cybernetic loop controlled by an LLM." - @MarkSchmidty
- "A robust definition would recognize "agent" is a concept beyond ML/AI.\n\nAn agent is an organization, that perceives and processes information, to create self-originated solutions to problems, in order to act on and manipulate its environment toward intrinsic or extrinsic goals." - @nazimadakli
11. Specific Implementations or Examples
-
Definition: Some tweets define agents by referring to specific examples or implementations, like ChatGPT Code Interpreter or particular frameworks.
-
Supporting Quotes:
- "The \n@TheoriqAI\n litepaper defines agents as, "AI agents are autonomous software systems that leverage modern generative AI models to plan, access data, use tools, make decisions, and interact with the real world to perform specific functions." \n- But a picture is worth a thousand" - @Tjido
- "A question I always ask myself about any proposed definition of "agent" is whether or not ChatGPT Code Interpreter qualifies as an agent under that definition\n\nIt's a great example of a tool-using LLM running in a loop that is trying to achieve a goal - is that "agentic" enough?" - @simonw
- "Here’s mine, compressible to a short three-element definition. From https://theatlantic.com/technology/archive/2024/07/ai-agents-safety-risks/678864/… - @zittrain
- "A LLM-based systems that can plan and use tools auto-discovered with https://github.com/nbonamy/nestor " - @nbonamy
- "No need to go NIH https://langchain-ai.github.io/langgraph/" - @RickLamers
12. Non-Technical or Business-Oriented Definitions
-
Definition: These definitions frame agents in terms of their business value or their role as digital assistants.
-
Supporting Quotes:
- "A button a business can deploy which can be clicked over and over with the dual objective of save and make money." - @fred_pope
- "An AI agent: the digital butler who never needs a coffee break, still learning not to send your shopping list to the Mars Rover by mistake. Thoughts?" - @sggresearch
- "Concierge for [purpose]\nThey know their stuff and how to get it, I know mine and what I'm looking for, by helping them know what I want they're better able to "make it happen."" - @AlanHicksLondon
- "An agent is a smarter, tireless, version of you on steroids. Lacking consciousness…as of now" - @TranquilTrader4
13. Relating Agents to Other Concepts
-
Definition: These tweets define agents by drawing parallels to other concepts like functions, PID controllers, or metaprogramming.
-
Supporting Quotes:
- "The LLM equivalent of 'functions'. You send an input and get an output which you then use in the 'parent' LLM context" - @firasd
- "a PID controller with with an LLM/LRM as the control unit. Typically only runs until an objective is achieved, but can be endless" - @kellogh
- "It's just metaprogramming to me. It means that a program can be designed to read, generate, analyse, or transform other programs, and even modify itself, while running. (Cit. Wikipedia)" - @ggcaponetto90
These groupings demonstrate the diverse perspectives on what constitutes an "agent" in the current AI landscape. It's clear that there isn't a single universally accepted definition, but rather a set of overlapping ideas centered around autonomy, goal-orientation, tool use, and LLM-driven control.
I ran this: