Photo: Kimberly White / Getty Images for TechCrunch (CC BY 2.0). TechCrunch Disrupt, San Francisco, September 20, 2023. Archive photo, cropped and resized.
Anthropic says it has met the AI research and development capability threshold in its Responsible Scaling Policy, a determination based on expectations for future models and its overall pace of progress rather than observations of current models.
The Claude developer disclosed the determination in its amended IPO registration statement filed with the Securities and Exchange Commission on November 10. It said it continues to assess the underlying data, is evaluating the implications and will release further reports.
“We made a determination that we have met this threshold based on what we expect from future models and our overall rate of progress, rather than on what we have observed from current models,” Anthropic wrote.
Expected progress, not a current-model finding
The distinction matters because the policy applies where Anthropic observes or expects a specified degree of accelerated progress. The filing does not present the determination as proof that a current model has demonstrated the capability.
One route to the threshold in the company’s current Responsible Scaling Policy concerns an observed or expected doubling of the rate of progress in aggregate AI capabilities, with that acceleration plausibly substantially attributable to automated research or engineering. That is not the same as doubling researchers’ productivity.
The company said AI already has been accelerating its research and development. It warned that its growing use of AI to improve its own systems could push model capabilities forward faster than its ability to develop, validate and implement safety, alignment and security measures.
Much of its evidence about that acceleration comes from lagging indicators, Anthropic said. It has in the past failed to identify significant acceleration until after it occurred and warned that this could happen again.
Safety precautions can affect releases and growth
Anthropic said past precautions have included delaying model releases and limiting their initial availability, including Mythos through Project Glasswing. Future measures could affect whether, how and when it develops or deploys new models or capabilities.
Those measures could slow capability gains relative to competitors that do not impose similar constraints and hurt its competitive position, revenue growth and results, the filing said. The disclosure does not announce a shutdown.
In a separate passage, Anthropic said it does not anticipate safely using Claude “to fully autonomously design and build successive generations of frontier models without meaningful human supervision or understanding,” if that becomes technically possible. It said it is considering policy updates to keep humans in control of frontier model development.
The filing describes a safety-policy determination and its possible consequences, not an announcement that humans have lost control of the systems.
The disclosure accompanies Anthropic’s IPO roadshow and proposed $70-to-$78 share-price range. Those proposed offering terms are separate from the research-threshold determination.
JBizNews Desk | Technology
© JBizNews.com. All rights reserved. Unauthorized reproduction or redistribution is strictly prohibited.
