Anthropic employees say advancements in AI are causing chaos within the company.
Anthropic wrote in a blog post on its website Thursday about the risks of AI advancing to the point where it can improve autonomously. The post quotes an anonymous Anthropic employee who talked about AI’s coding capabilities, and one employee’s words sum up the confusion surrounding AI in the workplace.
“On days when everything is going well, you can’t help but think that what you’re doing doesn’t matter. Everything is automated and better and faster than ever before,” they said.
“But there were days when everything fell apart and I had no idea why, and I had no idea what I was doing,” the employee added.
Other employee stories included one who said he hadn’t written any code himself in about five months and one who predicted that AI-generated code would outperform human-written code within a year.
The blog post states that while Anthropic’s Frontier LLM, Claude, can handle engineering problems and research tasks, “a significant performance gap still exists where Claude exercises judgment in selecting goals in both engineering and research.”
This comes as AI has changed roles beyond recognition in the space of a few short months, such as in software engineering. Frontier AI labs like Anthropic and OpenAI released new models late last year that can perform complex tasks significantly more efficiently than older versions.
The impact this has on the workplace is immeasurable. CEOs are working to flexibly change how much of their code is written by AI, which Google says has reached 75%. And companies are increasingly choosing to invest in AI rather than hiring or employee bonuses. Some companies have already announced cuts to AI-related staff.
Thursday’s blog post was written by staff at The Anthropic Institute, a division of the company that publishes research and recommendations on the impacts and risks of powerful AI systems. In a post about how AI has developed to the point where it can improve itself, the institute called for a slowdown in AI development.
“The way we protect, monitor, and shape the behavior of our systems all become more important as systems become fully capable of building their own successors,” the institute said in a blog post.
The report urges a collaborative “meaningful slowdown or pause” of AI research institutes developing frontier models, which would “enable social structures and collaborative research to catch up” with advances in AI.