How "Independent" AI Safety Groups Are Pumping Anthropic's IPO

The video critiques the promotion of Anthropic’s claimed rapid AI progress ahead of its IPO, highlighting undisclosed financial ties between Anthropic investors and organizations like METR, Epoch AI, and 80,000 Hours that produce supportive reports and videos. It argues that these conflicts of interest, selective data use, and biased interviews undermine the credibility of the claims, urging viewers to approach such narratives with skepticism and demand greater transparency.

The video critiques a recent 80,000 Hours video that claims Anthropic, an AI company, has reached a significant milestone called “Recursive Self Improvement,” suggesting rapid AI progress just before Anthropic’s IPO. The critique highlights conflicts of interest among the involved parties, noting that key organizations like METR, Epoch AI, and 80,000 Hours are funded by early Anthropic investors and affiliated billionaires, including Eric Schmidt, Dustin Moskovich, and Jaan Tallinn. These connections, the video argues, are not transparently disclosed, casting doubt on the independence of the reports and videos promoting Anthropic’s achievements.

The video explains that METR, which produced the report claiming Anthropic’s rapid AI progress, presents itself as independent but is heavily funded by Anthropic’s early investors. This funding relationship is omitted from METR’s independence checklist, raising questions about the objectivity of their findings. Similarly, 80,000 Hours, which produced the promotional video, fails to adequately disclose its financial ties to these investors. The critique suggests that these undisclosed conflicts of interest could bias the narrative to inflate Anthropic’s perceived progress and influence investor decisions ahead of the IPO.

Regarding the core claim of AI progress doubling, the video disputes the validity of METR’s analysis. It points out that the reported “twice as fast” improvement is based on a single “capabilities index,” a complex metric created by Epoch AI, another organization funded by Anthropic investors. The methodology is criticized for being unstable and prone to cherry-picking data, as METR selectively includes or excludes certain Anthropic models to shape the results. This selective data handling can produce widely varying conclusions, from significant acceleration to slight slowdown, undermining the credibility of the claim.

The video also challenges the anecdotal evidence presented by METR, which relies on interviews with Anthropic employees who stand to benefit financially from the IPO. The interviews show mixed opinions about the claimed acceleration, with roughly equal numbers supporting and opposing the idea of a doubling in AI progress. Despite this, METR characterizes the interviews as “broad support,” a conclusion the video finds misleading and unscientific. Overall, the critique portrays the evidence for rapid AI progress as weak and heavily influenced by financial incentives.

In conclusion, the video urges viewers to approach the claims about Anthropic’s AI advancements and the associated IPO with skepticism. It emphasizes the importance of transparency about conflicts of interest and rigorous, unbiased research methods. While not providing investment advice, the video encourages potential investors to be fully informed about the tangled financial and organizational relationships behind the promotional materials. The creator apologizes for the rushed nature of the critique but stresses the urgency of sharing this perspective before the IPO.