Unprecedented Acceleration in Model Development
A report issued by the AI safety research organization METR showed that Anthropic is experiencing a significant acceleration in the pace of developing its intelligent models, with data indicating that the company achieved progress equivalent to two years of historical development over the past year. According to the 11-week investigation, this acceleration is not only due to increased computing power, but also to the contribution of the artificial intelligence itself to the development process.
Company employees reported that their use of the latest internal models increased their productivity by 12 times compared to working without technical assistance, double the rate recorded just a few months ago. The report also revealed that the company has finished training three new models that have not been released to the public yet, reflecting extreme speed in development cycles.
Why Does This News Worry Researchers?
The crisis lies in the contradiction between the calls by Anthropic CEO Dario Amodei for the necessity of slowing down the pace of AI development, and the reality indicating that the company is leading a race toward systems capable of recursive self-improvement without significant human intervention. Researchers believe this development may outpace the ability of current safety measures to keep up, presenting society with complex technical and ethical challenges.
Transparency and Editorial Constraints
The report sparked widespread controversy due to Anthropic's decision to withhold large parts of the results reached by the METR organization, under the pretext of protecting trade secrets and preventing competitors from exploiting the company's technical roadmap. AI safety experts criticized this blackout, considering that it prevents the public and regulatory bodies from independently assessing the true risks.
For his part, Sam McCandlish, co-founder of Anthropic, defended the withholding decision, emphasizing that disclosing details of the acceleration engines could exacerbate the risks the company is trying to avoid. However, questions remain about the ability of private companies to manage technical security of this scale without sufficient government or independent oversight, especially with increasing political pressure in the United States to impose stricter restrictions on AI development.