AI Interpretability Claim Sparks Debate as Personal Agents Strain Internet
A Goodfire researcher's 26,000-word post argues LLMs may be easier to understand and control than thought, drawing both excitement and skepticism. Meanwhile, personal AI agents are prompting websites to block them, raising questions about the internet's future design.
Nick Cammarata, the Goodfire researcher, wrote the post. MIT Technology Review's Will Douglas Heaven interviewed Cammarata the day after the post blew up. They discussed what Cammarata thinks researchers are getting wrong, what his work actually shows, and why skeptics aren't ready to celebrate just yet. Not everyone is convinced by the claim.
The debate over LLM interpretability comes as personal AI agents are beginning to disrupt the internet's traditional divide between bots and people. These agents take on online tasks such as booking flights and reserving hotel rooms. Meta's Muse alone was downloaded 5 million times in its first month, according to the newsletter.
The mingling of AI agents with human visitors on an internet designed largely to prioritize people has gone poorly. Many websites are blocking AI agents. If demand for such agents keeps rising, the internet could be redesigned for agents rather than people, MIT Technology Review reported. That prospect raises questions about how the internet will serve human users in the future.