OpenAI is developing a new project to improve the reasoning capabilities of its AI models, called “Strawberry,” according to documents first discovered by ReutersThe project is a key part of OpenAI’s efforts to build more powerful AI models that can operate autonomously in conducting research.
According to internal documents reviewed by Reuters, Strawberry is focused on building an AI that not only answers questions but also searches online and conducts follow-up research on its own. This so-called “deep research” trick would be a major leap from current AI models, which rely on existing data sets and respond in ways that are already programmed.
There are no details about Strawberry’s exact mechanics, but it apparently involves AI models that use specialized processing methods after training on large datasets. This innovative approach could potentially set a new standard in AI development. An AI that can think ahead and do its own research to understand the world is much closer to a human than ChatGPT or other tools that use AI models can offer. It’s a daunting goal that has eluded AI developers so far, despite countless developments in the field.
Reuters reported that Strawberry, then known as Q*, had made a number of breakthroughs. There were demonstrations where viewers saw AI tackle scientific and mathematical problems beyond the reach of commercial models, and OpenAI had apparently tested AI models that scored over 90% on a championship-level math problem dataset.
Think forward
If OpenAI achieves its goals, its reasoning capabilities could transform scientific research and everyday problem solving. It could help close gaps in scientific knowledge by looking for gaps and even offering hypotheses to fill them. This would greatly accelerate the pace of discovery in a variety of domains.
If successful, Strawberry could mark a turning point in AI research, bringing us closer to truly autonomous AI systems that can perform independent research and provide more sophisticated reasoning. Strawberry appears to be a permanent part of OpenAI’s long-term plans to demonstrate and improve the potential of its AI models.
Even after GPT-3 and GPT-4 set new benchmarks for language processing and generation, there is a big leap toward autonomous reasoning and deep exploration. But it fits with other work on the road to artificial general intelligence (AGI), including the recent development of an internal scale for mapping the progress of large language models.