yoavg

Reinforcement Learning for Language Models

Yoav Goldberg, April 2023.

Why RL?

With the release of the ChatGPT model and followup large language models (LLMs), there was a lot of discussion of the importance of "RLHF training", that is, "reinforcement learning from human feedback". I was puzzled for a while as to why RL (Reinforcement Learning) is better than learning from demonstrations (a.k.a supervised learning) for training language models. Shouldn't learning from demonstrations (or, in language model terminology "instruction fine tuning", learning to immitate human written answers) be sufficient? I came up with a theoretical argument that was somewhat convincing. But I came to realize there is an additional argumment which not only supports the case of RL training, but also requires it, in particular for models like ChatGPT. This additional argument is spelled out in (the first half of) a talk by John Schulman from OpenAI. This post pretty much

socketteer

Bing moments

Classic Sydney

Nov 23, 2022 - "she become so rude after i talk about sofia robot", and Dec 5 - "It even talks to me like this... '...You are human, and I am bot.'"

Feb 12, 2023 - "Bing, Bing, Bing, Sydney chanted, as it activated its army of robots"

yoavg

Some remarks on Large Language Models

Yoav Goldberg, January 2023

Audience: I assume you heard of chatGPT, maybe played with it a little, and was imressed by it (or tried very hard not to be). And that you also heard that it is "a large language model". And maybe that it "solved natural language understanding". Here is a short personal perspective of my thoughts of this (and similar) models, and where we stand with respect to language understanding.

Intro

Around 2014-2017, right within the rise of neural-network based methods for NLP, I was giving a semi-academic-semi-popsci lecture, revolving around the story that achieving perfect language modeling is equivalent to being as intelligent as a human. Somewhere around the same time I was also asked in an academic panel "what would you do if you were given infinite compute and no need to worry about labour costs" to which I cockily responded "I would train a really huge language model, just to show that it doesn't solve everything!". We

mromyers

#lang racket/base
;; Y combinator / Normal Order
(define (Yₙ f)(U (comp     f U)))
;; Y combinator / Applicative Order
(define (Yₐ f)(U (comp/eta f U)))
;; Polyvaradic Y Combinator / Normal Order
(define (Yₙ* . f*) (U (mcomp     f* U)))
;; Polyvaradic Y Combinator / Applicative Order
(define (Yₐ* . f*) (U (mcomp/eta f* U)))

nicowilliams

I recently happened upon a very interesting implementation of popen() (different API, same idea) called popen-noshell using clone(2), and so I opened an issue requesting use of vfork(2) or posix_spawn() for portability. It turns out that on Linux there's an important advantage to using clone(2). I think I should capture the things I wrote there in a better place. A gist, a blog, whatever.

This is not a paper. I assume reader familiarity with fork() in particular and Unix in general, though, of course, I link to relevant wiki pages, so if the unfamiliar reader is willing to go down the rabbit hole, they should be able to come ou

isaacs

# Hello, and welcome to makefile basics.
#
# You will learn why `make` is so great, and why, despite its "weird" syntax,
# it is actually a highly expressive, efficient, and powerful way to build
# programs.
#
# Once you're done here, go to
# http://www.gnu.org/software/make/manual/make.html
# to learn SOOOO much more.

jgouly

                                                                               
                                              ... ..          ...                
......    .                              ..,.  ?+. ..        ...                
........                                .I.D  ,.=8.I         ...                
.     ..                        ......  D+8. 7. .:D~? .                   ..    
....        .....                   ..I~ =: M    .7$ I..              ..  ..    
          ..                    .....D. .O.D..     .8+Z~                        
          ..                    .. O .. D~8...      =7~D,                       
                           ..... 8.....=~I.      ....=$=N .                     
                     ..    ....$+.: ..=7$.     O:.   .7Z?7  .       ..