Basecamp Research is building a vast DNA database to help develop artificial intelligence that can understand biology, an approach inspired by the online text collections used to train large language models. The project highlights both the potential of AI to accelerate scientific research and concerns about how researchers’ work might be handled by AI companies.
In a former watch factory in central London, rows of white freezers hum. Inside them, dozens of small boxes hold barcoded glass tubes containing a combined total of 100 billion genes. Two boxes alone can contain 10 times more information than all the text on the internet, according to Dr. Oliver Vince, the facility’s co-founder.
Vince’s company, Basecamp Research, plans to collect as much DNA as possible. Its goal is to create AI that can “speak” the language of biology, much as technology companies in the United States and China collect documents from the internet to train large language models, or LLMs, which underpin many current AI systems.
Like many people in business and academia, Vince believes AI could significantly improve scientific research. The technology has also generated controversy, however.
Questions over an AI mathematics claim
Last month, OpenAI, the creator of ChatGPT, announced that it had solved one of mathematics’ major open problems, involving the famous Navier–Stokes equations. The company said its AI agent had cracked the problem within 88 hours.
The organization that awards prizes to people who solve such problems said the problem “appears to have been resolved,” while adding that its review process would take time.
Hours before the announcement, however, Tristan Buckmaster, a mathematician at New York University, raised concerns. He formed an independent team with Levent Alpöge, a mathematician who works for rival company Anthropic.
The pair were pursuing the same goal and had saved their drafts in Codex, OpenAI’s coding tool. Buckmaster said his collaborators had received information that news of their progress had “reached” OpenAI, although he did not explain how.
He asked whether research results stored by his team in Codex could have been reflected in OpenAI’s own research. Buckmaster emphasized that he was “not criticizing anyone.”
Source: www.bbc.co.uk