KDnuggets picks 10 X accounts worth following for LLM news

The AI feed on X has long since become a mix of useful releases, paper threads and endless noise. KDnuggets offers a short but practical list of ten accounts that help you follow LLMs without having to read hundreds of near-identical posts.

Why X still matters

While traditional media and academic journals publish analyses with a delay, X remains the place where researchers, engineers and tool builders discuss models almost in real time. That is where new papers, demos, open-source releases and first impressions of products surface first.

The problem is obvious: the algorithmic feed easily mixes strong technical analysis with superficial hype, and separating the two becomes a job in itself. That is the point of the KDnuggets list: not to find "all the major people in AI", but to collect sources that regularly deliver signal rather than noise. The author deliberately avoids the most obvious names and bets on accounts that offer either practical value or good news filtering.

This approach matters for people who do not want viral threads but clear reference points: what to read in the morning to understand in ten minutes what actually happened in the LLM world.

"If you want signal, not noise, these are reliable accounts to follow."

Who to add to your feed

The list is not about a single role but about several ways of following the market. Some accounts are useful for papers and research threads, others for practice and architecture breakdowns, and others for fast updates on releases, tools and running models locally. As a result, the feed does not close in on "news for news' sake" but covers the whole cycle: from idea and paper to real-world application.

• DAIR.AI and alphaXiv: for papers, short explanations and tracking what people are discussing around arXiv.
• Andrej Karpathy: for intuitive explanations, a fundamental perspective and a sense of where LLMs are heading.
• Sebastian Raschka and Simon Willison: for those who want not just to read about AI but to actually build and test things.
• The Rundown AI, AK and Matt Wolfe: for a stream of releases, new open-source tools and quick product updates.
• Ahmad Osman and Ethan Mollick: for topics around local inference, GPU infrastructure, work, and AI's impact on organizations.

The list also has an internal logic. If you only need research, a couple of accounts is enough. If practical building matters more, you can assemble a core from Raschka, Willison and Karpathy. If you want to see the whole market, the news accounts complement the technical side and help you avoid missing important launches of models, services and tools. The result is a feed that stays compact but still covers theory, hands-on practice and product signals.

How they differ

The article focuses not on the loudness of a name but on the type of value each feed provides. For example, Simon Willison is useful for honest notes on what works in LLMs in practice and what breaks in real scenarios. Ahmad Osman covers a different layer: infrastructure, local models, inference and GPUs. Ethan Mollick, by contrast, talks less about model internals and more about how AI is changing education, work and processes inside companies.

This mix is especially useful now that the LLM market is splitting in two. On one side, everyone is discussing new releases, reasoning and agentic capabilities. On the other, teams increasingly need not general impressions but answers to applied questions: what is worth trying, what can be deployed locally, which tools really save time, and which papers could quickly turn into product features. A good X feed should handle both tasks, not just feed FOMO.

Another takeaway from the piece is simple: there is no point in following hundreds of AI accounts. It is far more important to build a small but stable set in which sources do not duplicate each other. One account for research, another for development, a third for infrastructure, a fourth for business impact. This approach cuts noise and turns X into a working tool rather than an endless stream of distractions. In essence, it is about a personal news editor that you assemble for yourself.

What this means

There is too much information about LLMs, so value is shifting from speed to quality filtering. The KDnuggets list is useful because it offers not yet another list of "top influencers" but a compact set of roles: researcher, practitioner, news feed, infrastructure specialist and observer of AI's impact on work. For developers, product managers and founders, it is a good way to keep a finger on the pulse without daily overload and without spending hours sorting through a chaotic feed.

*Meta is designated an extremist organization and banned in Russia.