Andrej Karpathy 'complains' of getting trolled on Twitter, his former boss Elon Musk responds; says: We need...

Andrej Karpathy is no stranger to Twitter, the social media platform now known as X. He's spent close to two decades there. But the AI researcher, who co-founded OpenAI and recently joined Anthropic, says the site has hit a new low.

After a thread about Anthropic's Claude Tag feature spiralled into mockery and cheap shots, Karpathy aired his frustration with the platform's increasingly hostile culture—and pinned the blame squarely on X's recommendation algorithm.

His former Tesla-era boss Elon Musk, who now owns X, didn't just hear him out. He agreed.It started when Karpathy laid out the thinking behind Claude Tag, Anthropic's new workflow that lets Claude join a team inside Slack with access to chosen channels and tools.

As someone working on the inside, he walked through what it took to build—the engineering needed to make it "just work" across tools, integrations, compute environments and memory. He called it the third major redesign of how people interact with large language models.

The first paradigm was the LLM as a website you go to. The second was an app you download. This third one, he argued, is a self-contained, persistent entity with org-wide tools and context, working alongside teams of humans.

Less like chatting with a bot, more like having a colleague who happens to write most of the code.

Andrej Karpathy says critics didn't read past the headline

That framing didn't land with everyone. Plenty of users dismissed Claude Tag as little more than a chatbot wired into Slack, with some pointing to existing tools that already do something similar. Karpathy pushed back, saying many critics hadn't read past the headline before taking shots. "This is correct, I think a number of people on the tl didn't read past the title and made inferences and comparisons that are just wrong," he wrote.

It wasn't a "feature" like some crappy Slack bot, he insisted, but an "org-level harness" whose difference would become clearer over time.That's when the conversation drifted away from AI entirely and onto the state of X itself.

Two decades in, Karpathy says X has never been this toxic

In a follow-up, Karpathy didn't hold back. He said that across nearly 20 years on the platform, "it has never been this toxic and Reddit-like." His theory: the algorithm actively rewards hostility, with users getting "RL'd"—a nod to reinforcement learning—into posting outrage and inflammatory takes because that's what earns engagement and reach.

The result, he said, is that he posts less and visits less. He even mentioned working out of Anthropic's Slack these days instead.Musk's reply was short and to the point: "We need a complete overhaul of the algorithm."He offered no timeline, no specifics and no measurable goals, but the comment lined up with his long-running pitch about making X's recommendation systems more transparent. Since buying Twitter in 2022 and rebranding it as X, Musk has repeatedly promised to clean up the platform's feed. Whether this exchange nudges that along, or simply becomes another viral moment the algorithm rewards, remains to be seen.