-1720844872605.webp)
- By Ashish Singh
- Sat, 13 Jul 2024 10:01 AM (IST)
- Source:Reuters
OpenAI, the maker of ChatGPT, is developing a new artificial intelligence model called "Strawberry," according to sources and internal documents reviewed by Reuters. The unreported initiative is part of the Microsoft-backed startup's efforts to demonstrate advanced reasoning capabilities through its models.
As per a recent internal memo seen by Reuters in May, teams within OpenAI are working on Strawberry. Reuters was unable to determine the exact date of the memo, which describes OpenAI's research strategy using Strawberry. The strategy is still being developed, the source told Reuters. The news organisation was unable to determine Strawberry's proximity to public availability.
The initiative, which uses Strawberry models, aims to empower the company's artificial intelligence (AI) to not only provide responses to questions but also to anticipate sufficiently to navigate the internet independently and dependably to carry out what OpenAI refers it as "deep research," the insider said as quoted by Reuters.
When questioned about Strawberry and the data included in the report, an OpenAI company representative stated in a statement that they wanted their AI models to be able to understand the world more like humans do. The industry generally conducts ongoing research into new AI capabilities, believing that these systems will become more intelligent over time.
Formerly branded as Q*, the Strawberry project was already viewed as a breakthrough within the corporation, according to a Reuters report from last year.
Earlier this year, according to two people, they saw what OpenAI employees claimed were Q* demos, which could solve complex math and science problems beyond the capabilities of currently existing commercial models.
According to Bloomberg, OpenAI demonstrated a research project that it claimed had new abilities for thinking similar to those of a person on Tuesday during an internal all-hands meeting. A representative for OpenAI acknowledged the meeting but would not comment on its specifics. It was unclear to Reuters whether the project on display was Strawberry.
According to a person familiar with the matter, OpenAI anticipates that the breakthrough would significantly enhance the reasoning powers of its AI models. Strawberry, they added, entails a specific method of processing an AI model after it has been pre-trained on extremely big datasets.
According to researchers Reuters spoke with, reasoning is essential for AI to develop intelligence on par with or higher than that of humans.
Even though huge language models are now significantly faster than humans at summarising complex texts and crafting elegant prose, they frequently fail to solve common sense puzzles, such as tic tac toe and logical fallacy recognition. The model frequently "hallucinates" false information when faced with these kinds of issues.
The consensus among AI experts contacted by Reuters is that reasoning in the context of AI is creating a model that allows AI to anticipate future events, accurately represent the physical world, and solve difficult multi-step problems.
It is believed that developing AI models' reasoning skills would enable them to do a wide range of tasks, from organising and developing new software programmes to producing significant scientific discoveries.
Sam Altman, the CEO of OpenAI, stated earlier this year that "the most important areas of progress around AI will be around reasoning ability."
Like most academic labs that conduct AI research, several corporations such as Google, Meta, and Microsoft are also experimenting with various ways to increase reasoning in AI models.
On the other hand, there is disagreement among researchers over whether large language models (LLMs) can include concepts and long-term planning in their prediction process. For example, Yann LeCun, a Meta employee and one of the pioneers of current AI, has stated on multiple occasions that LLMs cannot reason in a way that is comparable to that of a human.
