GPT

Article

GPT is a recurring organization in the Astral Codex Ten archive, appearing 8 times across 8 issues between January 19, 2022 and November 20, 2025. The archive places it in contexts such as “the equivalent of a GPT that knows the sentence”; “Janus argues that language models like GPT - the first really interesting AIs worthy of alignment considerations”; “real LLMs like Claude and GPT”. It most often appears alongside OpenAI, Claude, Anthropic.

Metadata

Category: Organizations
Mention count: 8
Issue count: 8
First seen: January 19, 2022
Last seen: November 20, 2025

Appears In

- OpenAI (5 shared issues)
- Claude (4 shared issues)
- Anthropic (3 shared issues)
- Claude (3 shared issues)
- Britain (2 shared issues)
- Eliezer (2 shared issues)
- Eliezer Yudkowsky (2 shared issues)
- Elon Musk (2 shared issues)
- Europe (2 shared issues)
- Google (2 shared issues)
- GPT (2 shared issues)
- GPT-3 (2 shared issues)

External Links

Source Context

Recovered passages from the original issue text. When the raw archive preserved outbound links inside the source passage, they are listed directly under the quote.

Practically-A-Book Review: Yudkowsky Contra Ngo On Agents

January 19, 2022 · Original source

Eliezer thinks modern tool AIs are just “tons and tons of memorized shallow patterns” - the equivalent of a GPT that knows the sentence “e equals m c…” is usually completed “…squared” without having a deep understanding of relativity. Deep pattern-recognition ability come from agents with parts that are actually able to search for patterns and coherency within their knowledge base. The reason humans evolved to be good at chipping handaxes, got a lot of training data related to chipping handaxes, and ended up able to prove mathematical theorems - is because instead of just memorizing shallow patterns about how hand-axes work, they have a consequentialist drive to seek coherence and useful patterns in data. Some AIs already have something like this: if you evolve a tool AI through reinforcement learning, it will probably end up with a part that looks like an agent. A chess engine will have parts that plan a few moves ahead. It will have goals and subgoals like "capture the opposing queen". It's still not an “agent”, because it doesn’t try to learn new facts about the world or anything, but it can make basic plans. The same processes of evolution, applied to something smarter, could create something fully agenty.

Janus' Simulators

January 26, 2023 · Original source

In Simulators, Janus argues that language models like GPT - the first really interesting AIs worthy of alignment considerations - are, in fact, none of these things.

Inline links: Simulators

AI Sleeper Agents

January 16, 2024 · Original source

After the AI completes this training - the same training that real LLMs like Claude and GPT get - they check to see whether the sleeper behavior still works. It does.

Should The Future Be Human?

January 23, 2024 · Original source