What is true about gpt2-chatbot?

#AI #OpenAI #ChatGPT #gpt2-chatbot

893

Ṁ54k

Ṁ2.1k

2025

ALL

98.9%

Its knowledge cutoff is in or before Jan 2024

94%

It will get matched/surpassed by a LLM from another company/organization by EOY 2024

76%

Mixture of Experts

50%

It runs on H200 GPUs

50%

It generates audio in a discrete manner (next token prediction + vqvae) instead of continuous (next patch prediction, diffusion)

50%

It generate images in a similar fashion to diffusion models

47%

The "also-good-chat-model" version is a smaller variant than the "good-chat-model."

39%

> 15% on SWEBench (few shot or zero shot)

34%

Mixture of Depth/Dynamic allocation of computation to each token depending on difficulty

34%

It uses Hierarchy Alignment (https://arxiv.org/abs/2404.13208)

25%

It used RAG/search/function-calls during initial debut on LMSys

25%

> 1315 on Lmsys Arena Leaderboard (2 weeks after first appearance) (for the most capable of gpt2, also, and good variants)

21%

A prompt that enables it to solve easy Sudoku puzzles be found by EOY 2024 (same rule as https://manifold.markets/Mira_/will-a-prompt-that-enables-gpt4-to) (resolve NA if not released)

17%

>1T parameters

16%

>50% of the pretrain corpus was synthesized (e.g., 10T out of the total 20T were generated or augmented by AI/heuristics)

13%

It will be open-weight by EOY 2025

10%

<10B parameters

10%

Context Window > 500K

It is what Ilya saw (that triggered him for the coup and unusual behaviors)

It uses a different architecture from the current paradigm (e.g., MoD, non-transformer, not next-token prediction)

I reserve the rights to NA any added options.

Clarification: gpt2-chatbot include the three variants, i.e. gpt2-chatbot, "i-am-a-good-gpt2-chatbot", and "i-am-also-a-good-gpt2-chatbot"

related market:

What is “gpt2-chatbot”?

What made gpt2-chatbot smarter?

Get Ṁ600 play money

Related questions

Will gpt2-chatbot rank the highest in chatbot arena by the end of May 2024?

94% chance

What will be true about GPT-5?

What is true of the new Japanese GPT-4?

Will chatgpt stop calling itself a "chatbot" by 2027?

27% chance

What is “gpt2-chatbot”?

What will be true about GPT-4.5?

What made gpt2-chatbot smarter (than GPT4)?

What capabilities can I elicit from the "im-a-good-gpt2-chatbot" models while they are available? [Add answers]

Will ChatGPT conversation history be critical to the success of future OpenAI models? (Ṁ200 subsidy)

Sort by:

Its knowledge cutoff is in or before Jan 2024

bought Ṁ150 Its knowledge cutoff... YES

The system prompt says it's October 2023. (I just ran a few tests to confirm, and it's guesses about events after that point were complete hallucinations.)

@MugaSofer I am debating if I should resolve according to GPT4o or wait till we find out what the other two variants (gpt2-chatbot, i-am-a-good-gpt2-chatbot) are?

@MugaSofer For example, what if any of the two variants have a later knowledge cutoff? (though very unlikely)

@Sss19971997 Obviously it would benefit me to have it resolved immediately.

How would you resolve it if some fit the criterion and some didn't? (For any question.) You did already resolve several options based on things that we only know for sure apply to one of the models.

@MugaSofer Any variant fitting count.

@Sss19971997 In that case you're safe to resolve YES; at least the main variant's knowledge cutoff "is in or before Jan 2024".

bought Ṁ200 Answer #e4c5c037c150 YES

Resolves YES.

Probably should have resolved YES. Sorry guys.

@Sss19971997 I mean Sam Altman specifically said it wasn't. It'll be strange if other markets about it resolve YES.

@chrisjbillington Sam Altman lied/disguised about a lot of things before release.

@chrisjbillington @Mira resolved GPT4.5 market according to GPT4o, which makes a lot of sense

@Sss19971997 Mira's market's definition is very broad and would have included new models in the GPT-4 family even if they were downgrades in capabilities. Mira is betting NO in similar markets.

@chrisjbillington well, i still think it is debatable. anyway.

also=GPT-4o

@Sss19971997

bought Ṁ30 Answer #466fe8d6bcd3 YES

https://twitter.com/abacaj/status/1787942691587739826

@AviD I am like very certain OpenAI is the creator, but I don't know if this is enough to resolve the market.

bought Ṁ509 Answer #b62cdf2052a2 NO

Resolves NO @Sss19971997, Sam Altman said it's not:

https://www.axios.com/2024/05/02/mystery-chatbot-openai-gpt2

Speaking on Wednesday at Harvard University, Altman told an audience that the mystery bot is not GPT-4.5, what many see as the likely next major update to GPT-4.

@chrisjbillington Would u mind if I wait for another 2 weeks or something to confirm?

@chrisjbillington Or maybe just a couple of days after the video of the talk is public on YouTube

@Sss19971997 I don't mind a few days in either case, are you expecting it to be available on YouTube?

@chrisjbillington I would assume so. A clip is already out

@chrisjbillington done. Resolved

@Sss19971997 found the full video? Had a quick search but haven't seen it yet.

@chrisjbillington I have seen multiple sources confirming the GPT4.5 claim, i figure i can resolve early.

maybe that was not necessary. could have waited a bit longer.

@Sss19971997 Oh I definitely think there's enough evidence that he said it to resolve. Was interested in a video because perhaps he said some things that could resolve other markets is all!