Whilst it’s reportedly set to spend $7 billion on coaching and inference prices (with an general $5 billion shortfall), OpenAI is steadfastly in search of to construct the world’s first Synthetic Normal Intelligence (AGI).
Undertaking Strawberry is the corporate’s subsequent step towards that aim, and as of mid September, it’s officially been announced.
What’s Undertaking Strawberry?
Undertaking Strawberry is OpenAI’s newest (and doubtlessly best) giant language mannequin, one that’s anticipated to broadly surpass the capabilities of present state-of-the-art techniques with its “human-like reasoning skills” when it rolls out. It simply may energy the subsequent technology of ChatGPT.
What can Strawberry do?
Undertaking Strawberry will reportedly be a reasoning powerhouse. Utilizing a mix of reinforcement studying and “chain of thought” reasoning, the brand new mannequin will reportedly be capable to resolve math issues it has by no means seen earlier than and act as a high-level agent, creating advertising and marketing methods and autonomously fixing complicated phrase puzzles just like the NYT’s Connections. It may even “navigate the web autonomously” to carry out “deep analysis,” in response to inside paperwork viewed by Reuters in July.
Throughout growth, o1-preview was given a qualifying examination for the Worldwide Arithmetic Olympiad, one which the earlier GPT-4o mannequin was solely capable of reply accurately with 13% accuracy. The brand new mannequin answered 83% of the check’s questions. By way of coding, o1-preview scored within the 89th percentile in an internet Codeforces competitors. o1 may even reportedly reply questions that stumped earlier fashions like, “which is greater, 9.11 or 9.9?” and “what number of ‘Rs’ are within the phrase ‘Strawberry’?”
The Reuters report additionally notes that Strawberry’s structure is just like the Self-Taught Reasoner (STaR) technique. Developed at Stanford in 2022, STaR allows a mannequin to generate coaching knowledge on which to fine-tune itself, changing into extra succesful over time.
Why is it referred to as that?
We don’t know the precise motive for the identify “Strawberry,” as that’s not one thing OpenAI has publicly disclosed. It’s a code identify chosen for inside reference and to keep up secrecy throughout growth.
This was initially hinted at from a cryptic X submit by CEO Sam Altman, which despatched the web right into a flurry of rumors and hypothesis.
i really like summer season within the backyard pic.twitter.com/Ter5Z5nFMc
— Sam Altman (@sama) August 7, 2024
Strawberry didn’t, nevertheless, begin out as Strawberry. It reportedly was referred to as Q* (pronounced Q-Star), and when it was, Q* was the linchpin of CEO Sam Altman’s temporary ouster final November.
OpenAI researchers particularly cited Q* of their letter to the corporate’s board decrying the potential dangers posed by unregulated superior AI. After Altman had consolidated energy upon returning to OpenAI, Q* was reportedly renamed Strawberry in July 2024.
When will Strawberry be launched?
OpenAI launched a preview of Project Strawberry in September 2024, dubbed “o1-preview.” It was accompanied by a smaller, lighter-weight model of the brand new mannequin, “o1-mini.”
At launch, the brand new mannequin iteration was made instantly obtainable to ChatGPT Plus and Groups subscribers and arrived for Enterprise and Edu customers the next week. OpenAI claims that o1-mini will finally be made obtainable to free-tier customers, although the corporate didn’t specify a timeline for that to occur.
Wait, what was that about it having the ability to autonomously navigate the web? Isn’t that how we received Ultron?
No, we received Ultron due to comedian guide legends Roy Thomas and John Buscema, however the fears of a complicated AI rising out of human management are now not fully unfounded fantasy. Multiple sets of ex-OpenAI researchers have spoken out towards the corporate’s efforts to develop an AGI, citing its lack of developmental safeguards.
Nonetheless, these complaints have finished little to sluggish OpenAI’s drive to develop an AGI within the next decade. The corporate recently released a five-tier scale for measuring the capabilities of the AI techniques it builds towards that aim and is currently seeking additional funding (sure, besides Microsoft) in a new round led by Thrive Capitol that values the corporate in extra of $100 billion.
