A 26,000-word blog post by Nick Cammarata, a researcher at the AI start-up Goodfire, has stirred debate by arguing that large language models may be far easier to understand, and therefore to monitor and control, than had been thought, MIT Technology Review reports.
Cammarata previously worked at OpenAI, where he helped develop tools for probing what individual neurons inside models represent. The prevailing view has been that trained networks form tangled internal structures that are very hard to unpick. Some researchers on X hailed the post as a landmark, with one calling it the most important AI paper of the year.
Others are sceptical. Chris Potts of Stanford University told the magazine the post was little more than a hunch and did not make the underlying problems any easier. Cammarata said he had been surprised by the strength of the reaction.
Source: MIT Technology Review
