Matthew Berman on Andrej Karpathy Joining Anthropic and AI Pessimism

Andrej Karpathy’s announcement that he had joined Anthropic shook me. He is an influential AI researcher, an OpenAI co-founder and a former Tesla AI leader, but also an unusually effective educator who can make complex ideas accessible. His job update attracted enormous attention, and I found myself asking why the decision felt so consequential.

Karpathy said the next few years at the frontier of large language models would be formative, and that he was excited to return to research and development. He also said he remained passionate about education and planned to resume that work in time. I see Anthropic attracting exceptional people, and I wonder what its culture offers them.

My immediate interpretation was that joining a company can look like an endorsement of its worldview. I associate Anthropic with warnings about AI risk and job disruption, and I worry that pessimism is becoming dominant. That is my reading of the move, though: Karpathy’s announcement did not say that he agreed with all of the positions I attribute to the company.

I felt a loss at the thought of an independent researcher and teacher becoming part of a large lab. I also found myself wondering whether Karpathy shares a darker view of the future. Those are questions and inferences, not things his announcement established.

I had a similar initial reaction when Peter Steinberger, the creator of OpenClaw, joined OpenAI. An independent builder had created something innovative and open source, and I worried that a piece of that independence would be lost. I see a handful of large companies absorbing a great deal of the talent and oxygen in the field.

The move also struck me because Karpathy helped found OpenAI and had worked at Tesla. I interpreted his choice of Anthropic as a comment on the competing labs and their cultures, though choosing one employer does not tell us why he chose it or what he thinks of the others. I am struck by Anthropic’s ability to attract and retain talent.

I describe Anthropic’s public message as comparatively pessimistic, OpenAI’s as emphasizing tools and eventual abundance despite a difficult transition, and Elon Musk’s as explicitly optimistic about abundance. These are broad characterizations of different messages, not a complete account of what everyone at each organization believes.

A researcher with options may be drawn by the mission, the people or a technical problem. Karpathy may have seen something in Anthropic’s research that excited him; he may have other reasons altogether. I cannot know his financial situation, private motivations or agreement with individual predictions from the hiring announcement.

The larger concern for me is the public narrative. A 2025 Pew Research Center survey found that 50% of US adults were more concerned than excited about the increased use of AI in daily life. Concern had risen over time; ChatGPT itself launched in November 2022. The survey also included people who felt equally concerned and excited, as well as people who were more excited.

AI has also become a political subject. I see polarized reactions in debates involving Bernie Sanders and Alexandria Ocasio-Cortez, and in the Trump administration’s dispute with Anthropic. I worry that discussion becomes a choice between declaring AI entirely good and treating it as entirely bad, leaving little space for nuance.

I have historically been very optimistic about AI, perhaps enough that some viewers saw me as naive. My team discussed that feedback, and I think it had merit: if I am enthusiastic about everything, my opinion carries less weight. I have tried to be critical when criticism is warranted, even as I remain concerned about the prevalence of pessimistic narratives.

I see a connection to incentives on social media. Platforms and news organizations compete for attention, and fear or anger can be powerful ways to hold it. I point to a New Mexico jury’s finding concerning Meta and harm to children as one example in a broader discussion about engagement; it is not proof that every outlet or algorithm works the same way.

Politicians and media outlets can focus on frightening claims about AI, from jobs to the resources used by data centers. Those concerns are real subjects for scrutiny, but I do not think the whole picture is either disaster or sunshine. If people hear only reasons for fear, they may miss technologies that could do substantial good.

I worry that public fear could support policies that slow AI development while other countries continue. That is a geopolitical prediction, not a guarantee about what any country will do. The AI industry also bears responsibility for public distrust and does itself few favors in how it communicates.

One example is former Google CEO Eric Schmidt’s commencement speech, which drew audible interruptions. In the clip, Schmidt acknowledges fear and argues that AI will affect work across fields. He suggests graduates could assemble AI agents to help tackle problems they could not solve alone, and urges them to find a way to say yes.

Communications professional Lulu Cheng Meservey criticized the framing, arguing that the graduates should have been presented as protagonists rather than accessories to an AI-shaped future. I agree that the industry talks about a race to AGI using a term even insiders define differently. Telling graduates to help build something they fear could replace them is not, in my view, a compelling story.

There is a personal tension here as well. Videos criticizing a company’s pricing, developer treatment or transparency can perform very well; it is tempting to keep making negative videos about Google or Anthropic. Yet positive developments deserve attention too, and my overall outlook remains optimistic.

Viewers have told me I hype things up and rarely say anything negative. I took that seriously. As someone who has built startups, I know how difficult it is to put your work into the world and have people mock it, but that does not mean companies and their decisions should escape criticism.

I return to Karpathy because I had taken his move as an endorsement of the pessimistic positions I associate with Anthropic, including severe job loss and opposition to open source. That leap is my interpretation, not a statement by Karpathy. Another possibility is that his arrival brings a different perspective to the company.

I want to remain optimistic while criticizing bad decisions and listening to feedback when I get the balance wrong. I hope Anthropic becomes more positive with Karpathy on the team. For an example of the kind of pessimistic argument I am responding to, I point viewers to another discussion of an Anthropic essay about competition with China.