Despite an executive shuffle at the top of the company, OpenAI co-founder and CEO Sam Altman seems unbothered — so much so that he posted on an image to his personal account on the social network X today showing a pleasant view of strawberries growing in planters in a garden, captioned with “i love summer in the garden” in his typical all-lowercase text style.
i love summer in the garden pic.twitter.com/Ter5Z5nFMc— Sam Altman (@sama) August 7, 2024
While the photo on its own seems fairly innocuous and even a little bland, it has Altman’s followers on X and others in the large and active AI community on the site giddy with excitement that Altman is actually acknowledging one of the longest-standing rumors about OpenAI’s efforts to develop a new foundation model supposedly codenamed “Strawberry,” which may or may not be equivalent to the expected GPT-5 foundation model.
What other clues do we need? With this, Sam Altman has probably confirmed Q*. I can’t help but think that we’ll be officially getting Strawberry / Q* very soon.@sama says “i love the summer in the garden”. A hint that it’s coming this summer? https://t.co/q4xrPNfHmN
Bindu Reddy, CEO of open-source AI startup Abacus AI, also posted on X weighing in that it was a “reference to Project Strawberry.”
This is a reference to Project Strawberry (the new Q*)—the reasoning project OAI has been rumored to work on for over a year now.
The problem, however, is that several other labs, including Google, have cracked a bunch of techniques around math, reasoning, and synthetic data.… https://t.co/38jZHG3lYf
Further adding fuel to the fire is the fact that a new “anonymous-chatbot” AI model has been added to the popular open-source lmsys arena, where numerous AI model provider companies including OpenAI itself often test upcoming releases with users prior to an announcement.
New model on arena again. It is simply named ‘anonymous-chatbot’ this time. There is speculation already that this is Q*. Whatever it is, something new is on the way. https://t.co/Zvpr2XUSxG
The new “anonymous-chatbot” when asked claims to be part of OpenAI’s GPT-4 family of models, and follows similar language to the “gpt2-chatbot” that appeared on lmsys before the release of OpenAI’s GPT-4o multimodal model in May 2024.
Early users of the new “anonymous-chatbot” report that it exhibits more advanced reasoning capabilities than GPT-4o and other rival frontier models, suggesting it may be more powerful than them.
No other model can consistently get the right answer – anonymous-chatbot can https://t.co/TZmRNlt4jd pic.twitter.com/v91TCj4VZu
Rumors of a massively powerful new OpenAI model kicked off with a report from Reuters in November 2023 describing a new AI model called “Q*” (pronounced “Q-Star,” according to the report):
“Some at OpenAI believe Q* (pronounced Q-Star) could be a breakthrough in the startup’s search for what’s known as artificial general intelligence (AGI), one of the people told Reuters. OpenAI defines AGI as autonomous systems that surpass humans in most economically valuable tasks.“
Reuters followed by with an exclusive report in mid-July stating:
“ChatGPT maker OpenAI is working on a novel approach to its artificial intelligence models in a project code-named “Strawberry,” according to a person familiar with the matter and internal documentation reviewed by Reuters.”
As for what specific new capabilities Strawberry may offer, Reuters reports:
“The document describes a project that uses Strawberry models with the aim of enabling the company’s AI to not just generate answers to queries but to plan ahead enough to navigate the internet autonomously and reliably to perform what OpenAI terms “deep research,” according to the source.
In addition, it has been nearly 17 months, or about 1.5 years, since OpenAI released GPT-4, its last major new foundation model class.
And while it has released new foundation models GPT-4o and GPT-4o mini in that time — the first trained to be natively multimodal rather than natively text — the competition from Anthropic and Google and Meta with equally or in some cases more powerful models, at least in certain tasks, has only increased markedly as well.
We’ve reached out to OpenAI for more information about Altman’s post on X and will update when we hear back.
Source: venturebeat.com