Skip to content

Instantly share code, notes, and snippets.

View faizilham's full-sized avatar
💻
Segmentation Fault

Faiz Ilham faizilham

💻
Segmentation Fault
View GitHub Profile

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

@DavidBuchanan314
DavidBuchanan314 / cursed_mandelbrot.c
Last active June 28, 2023 15:12
Compile-time mandelbrot in pure C. Outputs a PGM image file to stdout. Output can be seen at https://twitter.com/David3141593/status/1062468528115200001
#include <stdio.h>
#define SQ(x) (x)*(x)
#define M0(x,y) SQ(x)+SQ(y)<4?0:0xe0
#define M1(x,y,x0,y0) (SQ(x)+SQ(y)<4)?M0(SQ(x)-SQ(y)+(x0),2*(x)*(y)+(y0)):0xc0
#define M2(x,y,x0,y0) (SQ(x)+SQ(y)<4)?M1(SQ(x)-SQ(y)+(x0),2*(x)*(y)+(y0),x0,y0):0xa0
#define M3(x,y,x0,y0) (SQ(x)+SQ(y)<4)?M2(SQ(x)-SQ(y)+(x0),2*(x)*(y)+(y0),x0,y0):0x80
#define M4(x,y,x0,y0) (SQ(x)+SQ(y)<4)?M3(SQ(x)-SQ(y)+(x0),2*(x)*(y)+(y0),x0,y0):0x60
#define M5(x,y,x0,y0) (SQ(x)+SQ(y)<4)?M4(SQ(x)-SQ(y)+(x0),2*(x)*(y)+(y0),x0,y0):0x40

Principles of Adult Behavior

  1. Be patient. No matter what.
  2. Don’t badmouth: Assign responsibility, not blame. Say nothing of another you wouldn’t say to him.
  3. Never assume the motives of others are, to them, less noble than yours are to you.
  4. Expand your sense of the possible.
  5. Don’t trouble yourself with matters you truly cannot change.
  6. Expect no more of anyone than you can deliver yourself.
  7. Tolerate ambiguity.
  8. Laugh at yourself frequently.
@kevin-smets
kevin-smets / iterm2-solarized.md
Last active June 21, 2024 14:48
iTerm2 + Oh My Zsh + Solarized color scheme + Source Code Pro Powerline + Font Awesome + [Powerlevel10k] - (macOS)

Default

Default

Powerlevel10k

Powerlevel10k