AI: Meta Delays Muse, Hoffman Rejects General Brake

The debate over the pace of cutting-edge AI research is intensifying. Mark Zuckerberg claims that each lab can slow down on its own and says he has delayed Muse at Meta. Dario Amodei is calling for a temporary regulation of developments, supported by Elon Musk and Sam Altman. Reid Hoffman, on the other hand, considers existential risks to be distant and advocates for better directing research rather than stifling it.

Meta Claims Targeted Slowdowns Without a Common Pact

Mark Zuckerberg asserts that companies can choose to slow down their development without a sector-wide agreement being necessary. He specifies that each lab has the responsibility and incentive to progress at the pace required to train its models safely. Zuckerberg indicates that Meta has postponed the release of Muse, the AI agent developed by the Superintelligence Labs team, by several months to enhance the safety and security of the model. He emphasizes that Meta did not ask other companies to do the same and made this decision solely for itself.

Anthropic Advocates for a Regulatory Pause, Musk and Altman Approve

Dario Amodei, CEO of Anthropic, has called on leading AI labs to temporarily regulate their developments and use this time to better align their models. Elon Musk and Sam Altman have expressed their agreement with Amodei's general objectives. Jacob Coxon, a former researcher at Anthropic, wrote that Anthropic has largely initiated the race for recursive self-improvement, a central topic in current discussions.

Reid Hoffman Rejects Alarmism and Focuses on Technical Oversight

Reid Hoffman believes that the existential risks associated with AI are distant and manageable. He considers that most apocalyptic concerns come from individuals presenting themselves as Cassandras predicting the end times. Hoffman argues that leading labs should better direct their research rather than slow down, emphasizing that continued work in AI can help address biosafety and cybersecurity risks and promote economic prosperity. He finds it legitimate to question why Anthropic and OpenAI are calling for a slowdown when they seem to be at the forefront of the sector. For him, it is very important that labs manage recursive self-improvement properly, a process that worries some industry players as it could lead to growth that is difficult to control. These positions were articulated at Politico's Decoded summit in Washington. In an essay published Wednesday, Hoffman described the calls from Dario Amodei and Sam Altman as steps in the right direction.