An Anthropic employee said AI advancements are leaving them in a state of internal turmoil.
In a blog post on its website on Thursday, Anthropic wrote about the risks of AI progressing to a point where it could improve itself autonomously. The post quoted unnamed Anthropic employees who talked about AI’s coding capabilities, and one employee’s quote summed up the confusion around AI in the workforce.
“On days where everything works well, I can’t help but think nothing I do matters, everything is automated and better and faster than I ever will be,” they said.
“But then there are days where everything breaks and I don’t understand why and I realize I have no idea what I’ve been up to anymore,” the employee added.
Other employee accounts included one worker who said they had not written any code themselves in about five months, and another who predicted that AI-generated code would outperform human-written code within a year.
The blog post said that Anthropic’s frontier LLM, Claude, could handle engineering problems and research tasks, but “large performance gaps persist when it comes to Claude exercising judgment in choosing goals in both engineering and research.”
This comes as AI has, in a short span of a few months, changed roles like software engineering beyond recognition. Frontier AI labs like Anthropic and OpenAI released new models at the end of last year that can perform complex tasks significantly better than their older versions.
The effect of this on workplaces has been monumental. Company CEOs have taken to flexing how much of their total code is written by AI, with Google saying that number stands at 75%. And companies are increasingly choosing to spend on AI rather than on hiring or on employee bonuses. Several have already announced AI-linked layoffs.
Thursday’s blog post was written by staff from The Anthropic Institute, an arm of the company that publishes research and advisories on the impact and risks of powerful AI systems. In the post, which talked about AI developing to a point where it could improve itself, the institute called for a slowdown in AI development.
“If systems are capable of fully building their own successors, the ways we secure them, monitor them, and shape their behavior all grow much more important,” the institute said in the blog post.
It urged a coordinated “meaningful slowdown or pause” of AI labs developing frontier models, which would “enable societal structures and alignment research to keep up” with AI advancements.
