Introduction
AI is advancing quickly enough that traditional news outlets and even academic journals can struggle to keep pace. Large language models in particular see frequent changes in reasoning, efficiency, and agentic capabilities, and much of that conversation plays out in real time on social media. X (formerly Twitter) remains a central gathering point for the AI research community, where developers, engineers, and researchers exchange ideas as they happen.
The challenge is filtering signal from noise in an algorithmic feed. The accounts below are not the largest or most obvious names; they are contributors who consistently share useful LLM updates, papers, tools, or thoughtful commentary. Readers building practical skills may also find value in this list of Python libraries every LLM engineer should know.
10 X accounts worth following for LLM updates
1. DAIR.AI (@dair_ai)
DAIR.AI regularly posts paper threads and short research explainers that stay technical while remaining readable. The account is frequently recommended as a trustworthy feed for AI and LLM research, and its recurring “Machine Learning Papers of the Week” series is a convenient way to track what is gaining attention.
2. Andrej Karpathy (@karpathy)
Karpathy remains one of the clearest voices on deep learning and the direction of LLM research. His posts tend to share intuition, learning advice, and perspective on where the field is heading, making the account a strong follow for anyone who values fundamentals.
3. Sebastian Raschka (@rasbt)
Raschka focuses on implementation and learning by doing. Followers can expect tutorials, architecture breakdowns, and practical machine learning and LLM insights that are consistently useful for people who build or want to build models.
4. alphaXiv (@askalphaxiv)
alphaXiv is built around the discovery and discussion of research papers, adding a social layer on top of arXiv. The account makes it easier to see which recent papers people are engaging with, offering a quick read on what is considered insightful or impactful.
5. The Rundown AI (@TheRundownAI)
The Rundown AI runs a high-volume stream of AI news that works best like a wire service: scan the headlines, open only what matters, and skip the rest. For keeping up with product launches, funding news, and model releases, it is fast and comprehensive.
6. AK (@_akhaliq)
AK is among the most reliable accounts for surfacing new arXiv papers and demos shortly after they appear. It is a practical way to catch noteworthy research and projects early, often before they are widely discussed elsewhere.
7. Ahmad Osman (@TheAhmadOsman)
Osman concentrates on systems, infrastructure, and hardware, particularly running LLMs locally rather than relying solely on hosted APIs. He shares practical insights on GPUs, inference performance, quantization, and self-hosted setups, which makes the account valuable for anyone interested in on-device deployment. A related primer covers running small AI models locally.
8. Matt Wolfe (@mreflow)
Wolfe shares daily AI updates and tool roundups in a builder-friendly style. For readers who want to know which new AI products launched in a given week without hunting for them, the account is a convenient summary.
9. Simon Willison (@simonw)
Willison is a strong follow for practical LLM use, sharing experiments, real-world tips, tooling breakdowns, and candid assessments of what works and what does not. The account is especially useful for people building with LLMs rather than only reading about them.
10. Ethan Mollick (@emollick)
Mollick discusses LLMs in the context of work, education, and real-world impact. The focus is less on model internals and more on what the technology changes in practice, which suits readers interested in applications and implications.
How to get the most from these accounts
Following a handful of high-signal accounts is more useful than chasing every trending post. A practical approach is to group these profiles into a dedicated list or feed, skim regularly, and save the threads worth reading in full. Recommendations like these reflect each account’s focus at the time of writing; handles and posting habits can change, so it is worth confirming an account’s current activity before relying on it. Pairing news-oriented feeds with implementation-focused ones tends to give the best balance of breadth and depth.
