Was the Bing LLM trained with RLHF?
39
Ṁ17k
2027
70%
chance

Resolves N/A if this information isn't publically available or at least leaked by credible sources by market close.

Get Ṁ1,000 play money
Sort by:

There's been trading on this recently so I want to clarify: this market will most likely resolve N/A. The question has never been "does Bing use some variant of GPT-4". It is: was the variant Bing was using at first release trained pre or post RLHF? As far as I know there has been no additional public information since then.

predicts YES

I suggest to ban author of this market because he refuses to acknowledge the obvious resolution to YES

isnt bing announced as being gpt4 the whole time[1]? isnt it well known that gpt-4 uses rlhf[2, ctrl-f "RLHF"]?

[1] https://techcrunch.com/2023/03/14/microsofts-new-bing-was-using-gpt-4-all-along/

[2] https://openai.com/research/gpt-4

@TommyMorriss That has never been the question - the issue is that the press releases tend to say "a version" or "an early version", which makes it ambiguous.

Ex-Yandex CTO who Microsoft hired to be in charge of Bing, including the one primarily responsible for the fast deploy after ChatGPT become a hit, saying on Twitter that the recent tri-toggle is "differently-RLHFed" from its prior version, implying that older version were also RLHFed.

Mikhail Parakhin on Twitter: "@zaptrem Multiple changes, including differently fine-tuned and RLHFed models, different prompts, etc." / Twitter

@Mira If you're posting this for market resolution: not sufficient, too ambiguous.

As originally released?

@JacobPfau Yeah I mean whatever thing is producing the deranged Twitter threads where it threatens to blackmail and kill its users.