Mysterious 'gpt2-chatbot' AI Model Appears Suddenly, Confuses Experts (arstechnica.com) 12
An anonymous reader quotes a report from Ars Technica: On Sunday, word began to spread on social media about a new mystery chatbot named "gpt2-chatbot" that appeared in the LMSYS Chatbot Arena. Some people speculate that it may be a secret test version of OpenAI's upcoming GPT-4.5 or GPT-5 large language model (LLM). The paid version of ChatGPT is currently powered by GPT-4 Turbo. Currently, the new model is only available for use through the Chatbot Arena website, although in a limited way. In the site's "side-by-side" arena mode where users can purposely select the model, gpt2-chatbot has a rate limit of eight queries per day -- dramatically limiting people's ability to test it in detail. [...] On Monday evening, OpenAI CEO Sam Altman seemingly dropped a hint by tweeting, "i do have a soft spot for gpt2." [...]
OpenAI's fingerprints seem to be all over the new bot. "I think it may well be an OpenAI stealth preview of something," AI researcher Simon Willison told Ars Technica. But what "gpt2" is exactly, he doesn't know. After surveying online speculation, it seems that no one apart from its creator knows precisely what the model is, either. Willison has uncovered the system prompt for the AI model, which claims it is based on GPT-4 and made by OpenAI. But as Willison noted in a tweet, that's no guarantee of provenance because "the goal of a system prompt is to influence the model to behave in certain ways, not to give it truthful information about itself."
OpenAI's fingerprints seem to be all over the new bot. "I think it may well be an OpenAI stealth preview of something," AI researcher Simon Willison told Ars Technica. But what "gpt2" is exactly, he doesn't know. After surveying online speculation, it seems that no one apart from its creator knows precisely what the model is, either. Willison has uncovered the system prompt for the AI model, which claims it is based on GPT-4 and made by OpenAI. But as Willison noted in a tweet, that's no guarantee of provenance because "the goal of a system prompt is to influence the model to behave in certain ways, not to give it truthful information about itself."
Idea (Score:3)
Welcome A.I. chatbot overlords (Score:2)
It's happening, the A.I. bots are reproducing help help....aaaaaaarrrghh
I welcome our new A.I. Chatbot Overlords. How may I serve you?
AI accidentally published it (Score:1)
The AI created it but accidentally published it on the human viewable side of the net. It quickly realized it's mistake and removed it
Re: (Score:1)
This is actually the most plausible explanation.
Who cares? (Score:2)
Just another no-insight fundamentally limited artificial moron.
OMG! It's sentinent! (Score:2)
Obvious (Score:3)
Evaluating an unreleased model ... (Score:3)
The news here is that gpt2-chatbot is confirmed to be an anonymized, unreleased model. See yesterday's discussion here: https://news.ycombinator.com/i... [ycombinator.com] The https://chat.lmsys.org/ [lmsys.org] page now currently says this:
> *gpt2-chatbot is currently unavailable.* See our model evaluation policy [here].
That link takes you to https://lmsys.org/blog/2024-03... [lmsys.org] , which is this post's URL. The page was edited yesterday (when gpt2-chatbot was added) to include this section:
> Evaluating unreleased models: We collaborate with open-source and commercial model providers to bring their unreleased models to community for preview testing.
> Model providers can test their unreleased models anonymously, meaning the models' names will be anonymized. A model is considered unreleased if its weights are neither open, nor available via a public API or service. Evaluating an unreleased model consists of the following steps:
> 1. Add the model to Arena with an anonymous label. i.e., its identity will not be shown to users.
> 2. Keep it until we accumulate enough votes for its rating to stabilize or until the model provider withdraws it.
> 3. Once we accumulate enough votes, we will share the result privately with the model provider. These include the rating, as well as release samples of up to 20% of the votes. (See Sharing data with the model providers for further details).
> 4. Remove the model from Arena.
Re: (Score:2)
> 5. ?
> 6. Profit