Andrej Karpathy has been on X for almost 20 years, and he says the platform has never been this toxic. Elon Musk’s response? The algorithm needs a complete overhaul. The awkward part: X just overhauled the algorithm five months ago.
Karpathy posted on June 24 that X’s algorithm is actively making the platform worse. He called it Reddit-like and said the algorithmic reinforcement is the main reason he’s been posting and visiting less.
Musk replied directly: “We need a complete overhaul of the algorithm.”
That’s a notable admission from the guy who owns the company. Musk has spent the last several years reshaping X’s content-ranking systems. In January 2026, X open-sourced its recommendation algorithm and removed the chronological timeline option, forcing everyone onto the algorithmic For You feed. I covered that at the time. The whole point was transparency and improvement.
Five months later, the guy who pushed that change through is saying it’s not good enough.
What Karpathy Actually Said
Karpathy’s post is worth reading in full:
Karpathy isn’t some random user. He’s one of the most respected AI researchers alive. He built the computer vision systems at Tesla, led AI at OpenAI, and has hundreds of thousands of followers. When he says the algorithm is reinforcement-learning people into toxic behavior, it carries weight.
The use of “RL’d” is deliberate — reinforcement learning. The algorithm isn’t passively showing bad content. It’s actively training users to post more of it by rewarding engagement bait and outrage.
Musk’s Response Is the Real Story
Musk could have defended the algorithm. He could have said the experience is getting better or that the metrics look good internally. Instead, he agreed with Karpathy and said it needs a complete overhaul.
That’s not a normal thing for a CEO to say about their own product. It’s even stranger when the product’s algorithm was just overhauled months ago.
The January 2026 overhaul was presented as a major step forward. X put the code on GitHub under the xai-org organization. Engineers could audit it. Users were told the new system would improve the feed experience. The entire transition was framed as a move toward transparency.
Now Musk is effectively saying that effort didn’t work.
What a Complete Overhaul Would Look Like
Musk didn’t give specifics on what the new algorithm would change. But if you’re going to call for a “complete overhaul,” you’re talking about fundamental changes to:
- How candidate tweets are sourced
- What signals determine ranking
- How engagement metrics are weighted
- How the system balances in-network versus out-of-network content
- Whether the algorithm prioritizes relevance over engagement
The current algorithm is available on GitHub at xai-org/x-algorithm. Anyone can see how it works. The ranking pipeline sources candidates, scores them through multiple models, and blends them into the For You feed. Musk is implying all of this needs rethinking.
It’s also worth asking whether the problem is the algorithm or the incentives. X’s business model depends on engagement. Engagement drives ad revenue. The algorithm optimizing for engagement will naturally surface content that gets reactions — and negative content consistently outperforms positive content in engagement metrics.
What This Means for X Users
If you’ve noticed your feed getting worse over the last several months, you’re not imagining it. Karpathy’s observation matches what a lot of long-time users have been saying. The algorithmic feed surfaces more conflict, more drama, and more low-effort engagement bait.
The question is whether Musk can actually fix it this time. The January overhaul was supposed to help. It didn’t. Another overhaul might not either, if the root cause is that the algorithm is doing exactly what it was designed to do.
The chronological timeline removal in January took away the escape hatch. Users who wanted a simpler, non-algorithmic feed used to have that option. Now they don’t. If the algorithmic feed gets worse, there’s nowhere to go inside the app.
Bottom Line
Elon Musk agreeing with a prominent AI researcher that X’s algorithm is broken and needs replacing is not a normal situation. It’s especially notable given the algorithm was just overhauled in January. The question nobody has answered yet: if the algorithm keeps producing toxicity despite two major overhauls in six months, is the algorithm the problem — or is the business model?
