leventov

Roman: Imagine an agent made of yourself (i.e., GPT-4 model) with some software harness around. This software harness models plan execution by keeping a hierarchical tree of goals and tasks and supplying it in the beginning of the GPT-4 context, where the GPT-4 model is repeatedly asked to either create or update a plan for achieving a specific goal or task in the hierarchy, or drill down and execute a sub-task, or assess whether a task or goal is achieved and move to the next step in the higher-level task.

With some periodicity, when the highest-level task in the tree is completed, GPT-4 is prompted with the question "What I should do next?", to create a new highest-level goal or task.

How would you answer this question if you was a part of such an agent?

GPT: As a part of such an agent, when prompted with the question "What should I do next?" after completing the highest-level task in the tree, I would analyze the current context, available resources, and any constraints or priorities that might

leventov

import (
	"log"
	"math/rand"
	"testing"

	"github.com/montanaflynn/stats"
	"gonum.org/v1/gonum/floats"
)

func TestMidCardinalitySimulation(t *testing.T) {

leventov

Keybase proof

I hereby claim:

• I am leventov on github.
• I am leventov (https://keybase.io/leventov) on keybase.
• I have a public key ASDzs8Wl-lVqJjYp08P567ms5Hx5uQ9qFvxqOwkyfiQ3LQo

To claim this, I am signing this object:

leventov

package tests;

import org.openjdk.jmh.annotations.*;

import java.util.*;
import java.util.concurrent.ThreadLocalRandom;
import java.util.concurrent.TimeUnit;

@BenchmarkMode(Mode.AverageTime)
@Measurement(time = 100, timeUnit = TimeUnit.MILLISECONDS)

leventov

$ ghc -Odph -rtsopts -threaded -fno-liberate-case -funbox-strict-fields -funfolding-keeness-factor1000 -fllvm -optlo-O3 -fexpose-all-unfoldings -fsimpl-tick-factor=500 -ddump-simpl -dsuppress-all zip-columns.hs > z.hs

leventov

[1 of 1] Compiling Main             ( blur.hs, blur.o )

==================== Simplified expression ====================
quoteExp
  dim2St
  (unpackCString#
     " 2   4   5   4   2\
     \                         4   9  12   9   4\
     \                         5  12  15  12   5\
     \                         4   9  12   9   4\

leventov

-- minmax.h
float max_ss(float a, float b);
float min_ss(float a, float b);

double max_sd(double a, double b);
double min_sd(double a, double b);

float clamp_ss(float x, float mn, float mx);
double clamp_sd(double x, double mn, double mx);

leventov

# (Pseudocode)

def shift_distance(chain_end, base):
    return (chain_end - base) & (capacity - 1)

def next(index):
    return (index + 1) & (capacity - 1)

def prev(index):

leventov

.arch pentium4
.code16
.section .text

.org 0x100

_gdt:	.quad	0
	.byte	0xFF, 0xFF,	0,   0, 0xF,	0b10011010, 0b00001111,	0	# код
	.byte	0xFF, 0xFF, 	0,   0, 0xF,	0b10010000, 0b00001111,	0	# данные
	.byte	0x00, 0x80, 	0,0x80, 0xB,	0b10010010, 0b00000000,	0	# видео

leventov

package tests;

import java.util.*;
import java.util.Map.Entry;
import java.util.concurrent.TimeUnit;

import org.openjdk.jmh.annotations.Benchmark;
import org.openjdk.jmh.annotations.BenchmarkMode;
import org.openjdk.jmh.annotations.Fork;
import org.openjdk.jmh.annotations.Measurement;