Who to Follow in AI: Matthew Berman, Andrej Karpathy, Ethan Mollick and Lex Fridman, Plus the Best AI News Sources Beyond X and Reddit

0:00 / 23:07Voice: Andrew
Chapters (9)
A round white table in a small podcast studio with several black microphones on adjustable arms, studio headphones and an audio mixer, with a wall monitor, a computer screen and an equipment rack in the background
Photo: Myotus, “Microphones in SPNN Podcast studio”, CC BY 4.0, via Wikimedia Commons – https://commons.wikimedia.org/wiki/File:Microphones_in_SPNN_Podcast_studio.jpg (resized, cropped at the right edge) · License: CC BY 4.0

Many people now follow AI news through a handful of personalities rather than through newspapers. This post profiles four of the most influential voices – YouTuber and newsletter founder Matthew Berman, researcher-educator Andrej Karpathy, Wharton professor Ethan Mollick and podcaster Lex Fridman – and then lists the best places to follow AI outside X and Reddit, grouped by what you want to get out of them.

How we checked: follower counts on X come from X’s public profile data, read on October 29, 2026 (around 01:15 Istanbul time, UTC+3). YouTube subscriber counts come from the third-party trackers vidIQ, TubeGrade and TubeStats (YouTube itself rounds public counts), dated as shown. Newsletter sizes are, unless stated otherwise, the publishers’ own claims on their websites or Substack pages and are marked “self-reported”. Recent examples were read directly on X, YouTube, Substack and the podcasts’ own sites. We did not use Reddit. We have no business relationship with anyone named here, and we did not subscribe to any of the services.

Key terms:

  • Influencer: here, a person whose posts, videos or newsletters shape how many people understand AI.
  • Newsletter: a regular email, often free; many run on Substack, where paying readers can get extra posts.
  • Open weights: an AI model whose trained “brain” is published for download.
  • Benchmark / leaderboard: a standardised test for AI models, and the ranking table built from it.
  • Impressions: how often a post on X was shown to people (not the same as how many people read it).
  • Self-reported: a figure the publisher states about itself; not independently checked.

Why these four?

Matthew Berman was the starting point. For the other three we chose people who are clearly prominent in 2026 and cover different needs, so the post is useful to more readers:

  • Andrej Karpathy – the most-followed practitioner voice on X among the candidates we checked (4.40 million followers), and a rare top researcher who explains AI to the public.
  • Ethan Mollick – the leading voice on what AI means for work, business and education, written for non-technical readers.
  • Lex Fridman – the largest long-form interview audience among the candidates (5.56 million on X, about 5.07 million on YouTube).

Close runners-up were the YouTube channel AI Explained (careful analysis of new models; its X bio claims 450,000+ YouTube subscribers, self-reported), Matt Wolfe (AI tools, 121,500 X followers; his Future Tools site claims 250,000+ newsletter readers, self-reported), Two Minute Papers (short research explainers) and Wes Roth (AI news on YouTube; we could not find his X account under the handle we tried, so we did not check his numbers).

Original sources: X profile data via X API (Oct 29, 2026) · AI Explained on X · Future Tools

1. Matthew Berman – the fast, hands-on news explainer

Who he is: Matthew Berman is an American YouTuber and the founder of Forward Future, an AI media company. Forward Future’s own About page lists him as founder and Nick Wentz as co-founder.

What he covers: new AI models the day they appear, hands-on tests, coding agents, open-source and local models (running AI on your own computer), and interviews with industry leaders. Forward Future adds a free daily newsletter, a weekly live show (“Forward Future Live”, every Friday) and guides and model reviews.

Audience (as of late October 2026):

  • YouTube: about 642,000 subscribers (vidIQ, Oct 28, 2026; TubeGrade, Oct 27, 2026); about 96 million total views.
  • X (@MatthewBerman): 153,473 followers (X profile data, Oct 29, 2026).
  • Forward Future newsletter: 80,000+ daily readers (self-reported on the About page).
  • Forward Future live show: 25,000+ weekly viewers (self-reported).

Style and strengths: quick, enthusiastic and practical. He tries new models himself and says plainly what he thinks. His videos are often online within hours of a big release. Forward Future’s media kit openly lists prices for sponsored segments in his videos and newsletter, so readers should know that some content is sponsored.

Recent example: on October 27, 2026 (US time) he wrote on X that “deepseek v4.1 flash is the best open weights model on the planet right now” – a post shown about 130,000 times – and soon after uploaded a YouTube video titled “Deepseek crushed everyone”. On October 21 Forward Future published his interview with OpenAI chief Sam Altman about self-improving AI, AI consciousness and OpenAI’s new hardware.

Good for: staying current on new models and tools in 10–20 minutes a day, especially if you like seeing a model tried out rather than only described.

Original sources: Forward Future: About · Forward Future: Media kit · Forward Future homepage (Sam Altman interview) · vidIQ: Matthew Berman stats · TubeGrade: Matthew Berman · X post, Oct 27, 2026 · YouTube: “Deepseek crushed everyone”

2. Andrej Karpathy – the researcher who teaches

Who he is: Andrej Karpathy was a founding member of OpenAI (2015–2017), Director of AI at Tesla, where he led the computer-vision team for Autopilot (2017–2022), and returned to OpenAI in 2023–2024. He did his PhD at Stanford and designed Stanford’s first deep-learning class, CS231n. Since 2024 he has focused on educational videos about AI (all according to his personal website).

What he covers: how large language models (the technology behind ChatGPT and Claude) work, practical ways to use them, and sharp observations on where AI is heading. His YouTube channel has a technical track (“Zero to Hero”, building neural networks from scratch) and a general-audience track (for example “Deep Dive into LLMs like ChatGPT” and “How I use LLMs”).

Audience:

  • X (@karpathy): 4,402,929 followers (X profile data, Oct 29, 2026).
  • YouTube: about 1.74 million subscribers (TubeStats and subsrank, Oct 28, 2026) – with only 17 videos, many of them hours long.

Style and strengths: calm, precise and generous with practical tips. He posts rarely, but his posts travel far: several of his 2026 posts were shown millions of times.

Recent example: on October 11, 2026 he warned on X that the “gap in shared understanding of LLM capability is widening”, estimating that most of the world has barely used advanced AI while a small group works with it daily. The post was shown about 6.8 million times. On October 2 he shared tips for understanding AI output, such as asking a model to explain things in Simplified Technical English (ASD-STE100), a controlled language from aerospace documentation; that post was bookmarked more than 80,000 times.

Good for: understanding how AI works and how experts actually use it – the best free starting point for anyone who wants to go beyond headlines.

Original sources: karpathy.ai · YouTube: Andrej Karpathy · TubeStats: Andrej Karpathy · X post, Oct 11, 2026 · X post, Oct 2, 2026

3. Ethan Mollick – AI for work, business and education

Who he is: Ethan Mollick is a professor at the Wharton School of the University of Pennsylvania, where he studies entrepreneurship, innovation and AI. His new book, Co-Existence, came out in October 2026 (his Substack post of October 20 says “My new book is out today”).

What he covers: what AI means for jobs, companies, teaching and everyday work; which AI tools to use for what; and research findings explained in plain language. He tries new products early and reports what it is like to use them.

Audience:

  • Substack newsletter One Useful Thing: 491,000+ subscribers (Substack profile page, Oct 28, 2026); Newsletter Insights reports about 490,000 (Oct 28, 2026).
  • X (@emollick): 410,351 followers (X profile data, Oct 29, 2026).

Style and strengths: clear, balanced and evidence-based. He links to studies, points out limits, and writes for managers, teachers and students rather than engineers. His regular “opinionated guide to which AI to use” is a useful reference.

Recent example: his October 20, 2026 essay “30 Months on the Exponential (in Charts)” looks back at how AI has developed since his previous book. On October 28 he noted on X that the benchmark firm Artificial Analysis had switched to closed (non-public) tests and found open-weights models falling further behind, adding that the result should be taken “with a grain of salt”. On October 27 he highlighted a study in which analysts working with an AI agent followed the agent’s view 86% of the time, and advised being deliberate about when to bring AI into a decision.

Good for: non-technical readers and decision-makers who want to know what AI means in practice, without hype.

Original sources: One Useful Thing: archive · One Useful Thing: “30 Months on the Exponential (in Charts)” · Substack: Ethan Mollick profile · Newsletter Insights: One Useful Thing · X post, Oct 28, 2026 · X post, Oct 27, 2026

4. Lex Fridman – long conversations with the people shaping AI

Who he is: Lex Fridman hosts the Lex Fridman Podcast, one of the best-known long-form interview shows. His X bio reads: “Host of Lex Fridman Podcast. Interested in robots and humans.”

What he covers: conversations of three hours or more with AI researchers, tech founders and programmers, but also scientists, athletes and political figures. AI and technology are a major, but not the only, theme. Episodes appear on YouTube, Spotify, other podcast apps and some in full on X.

Audience:

  • X (@lexfridman): 5,556,622 followers (X profile data, Oct 29, 2026).
  • YouTube: about 5.07 million subscribers and 934 million total views (vidIQ, Oct 28, 2026).

Style and strengths: slow, curious and deep. Guests get time to explain their thinking in full; every episode has timestamps and a free transcript. The episodes are best treated as first-hand accounts by the guests, not as investigative journalism.

Recent example: episode #503, “AI Crisis in Math: Navier-Stokes, Riemann Hypothesis & Complexity Theory”, published on October 26, 2026. Three mathematicians – Ankur Moitra (MIT), Steven Brunton (University of Washington) and Alex Kontorovich (Rutgers) – discuss recent claims that AI has solved some of mathematics’ hardest problems and what that means for the field. vidIQ counted about 651,000 YouTube views in the first day.

Good for: understanding the people and ideas behind AI in depth – ideal for commutes and long walks, less so for daily news.

Original sources: Lex Fridman Podcast #503 · Transcript #503 · vidIQ: Lex Fridman stats · X post, Oct 26, 2026

Comparison: the four influencers at a glance

  Matthew Berman Andrej Karpathy Ethan Mollick Lex Fridman
Background YouTuber, founder of Forward Future Researcher; OpenAI founding member, ex-Tesla AI director Wharton professor Podcast host
Main platform YouTube + daily newsletter X + YouTube Substack + X Podcast (YouTube, Spotify)
Audience ~642K YouTube; 153K X; 80K+ newsletter (self-reported) 4.40M X; ~1.74M YouTube 491K+ Substack; 410K X 5.56M X; ~5.07M YouTube
Pace Several videos a week, daily X posts Rare, high-impact posts Essays every few weeks, daily X posts One long episode every few weeks
Tone Fast, hands-on, enthusiastic Calm, technical, educational Balanced, research-based Long, reflective interviews
Best for Daily model and tool news Understanding how AI works AI at work and in education Deep conversations with key people
Cost Free (sponsored segments) Free Free; optional paid tier ($10/month, per Newsletter Insights) Free (sponsored)

Audience figures: X profile data, Oct 29, 2026; YouTube and Substack figures from the trackers and pages cited above (Oct 27–28, 2026).

Part 2: the best AI sources beyond X and Reddit

X and Reddit are fast but noisy. The sources below are easier to follow, often more reliable, and mostly free. Subscriber numbers are the publishers’ own claims unless stated otherwise.

Daily news (5–10 minutes a day)

  • Forward Future – The Future Today (free): Matthew Berman’s daily five-minute AI briefing; 80,000+ daily readers (self-reported). forwardfuture.com/newsletter/daily
  • The Rundown AI (free newsletter): one of the largest AI newsletters, “3,500,000+ readers” (self-reported); short, practical stories. therundown.ai
  • TLDR AI (free): a five-minute daily digest of AI news, research and tools; “1,100,000 subscribers” (self-reported). tldr.tech/ai
  • Ben’s Bites (free, with a paid community tier): a friendly newsletter about building with AI; “over 117,000 subscribers” (Substack). bensbites.com
  • Hacker News (free): a link-sharing site popular with programmers; good for spotting what developers actually discuss, with critical comment threads. news.ycombinator.com

Weekly round-ups (catch up once a week)

  • The Batch (free): weekly AI news and analysis from DeepLearning.AI, the education company founded by Andrew Ng. deeplearning.ai/the-batch
  • Import AI (free; optional paid support, $10/month listed): Jack Clark’s long-running weekly newsletter on research and policy; “over 147,000 subscribers” (Substack). importai.substack.com
  • Last Week in AI (free; optional paid tier, $5/month listed): weekly news summary plus a podcast; “over 191,000 subscribers” (Substack). lastweekin.ai

Research (new papers and ideas)

  • arXiv – cs.AI and related lists (free): the open archive where most AI papers appear first. Technical, but the source of almost every headline. arxiv.org/list/cs.AI/recent
  • Hugging Face Papers (free): a daily, community-voted list of notable new papers with short summaries. The former site Papers with Code now redirects here. huggingface.co/papers
  • alphaXiv (free): lets readers discuss arXiv papers and see which ones are trending. alphaxiv.org
  • Epoch AI (free): a research institute that tracks trends in AI compute, costs and capabilities with clear charts. epoch.ai

Tools and products (what can I use?)

  • Product Hunt (free): daily launches of new apps, many of them AI tools, with user votes and comments. producthunt.com
  • Hugging Face Hub (free; paid options for hosting): the main library of downloadable open models and demos. huggingface.co
  • Future Tools (free): Matt Wolfe’s curated directory of AI tools plus a newsletter (“250,000+ readers”, self-reported). futuretools.io
  • GitHub Trending (free): the open-source projects gaining the most attention right now. github.com/trending

Benchmarks and rankings (which model is best?)

  • Arena (formerly LMArena) (free): people compare two anonymous AI answers and vote; the votes form public leaderboards for text, image and code models. lmarena.ai now redirects to arena.ai. arena.ai
  • Artificial Analysis (free to browse): independent comparisons of models and providers on quality, price, speed and latency. In October 2026 it moved to version 5 of its index, which uses closed tests. artificialanalysis.ai
  • OpenRouter Rankings (free): shows which models developers actually use, measured in tokens processed – a useful reality check next to test scores. openrouter.ai/rankings

Deep analysis (understand the “why”)

  • One Useful Thing (free; optional paid tier): Ethan Mollick’s essays on AI at work and in education. oneusefulthing.org
  • Simon Willison’s Weblog (free): a programmer’s careful, frequent notes on new models and tools, with honest tests. simonwillison.net
  • Interconnects (free; paid tier $20/month listed): Nathan Lambert on how models are trained and on open-model strategy; “over 87,000 subscribers” (Substack). interconnects.ai
  • Latent Space (free; podcast and newsletter): “the AI engineer publication” – technical interviews and guides for people building with AI. latent.space
  • State of AI Report and Stanford AI Index (free, annual): the two big yearly overviews of research, industry and policy. stateof.ai · hai.stanford.edu/ai-index

YouTube channels and podcasts

Communities (ask questions, see what builders do)

  • Hugging Face Discord (free): a large community server around open models and tools. discord.com/invite/huggingface
  • Many newsletters and tools above run their own Discord or Substack chat communities; these are often part of a paid tier.

Comparison: which source for which need?

Need Best free starting points Time per week Technical level
Daily news Forward Future, The Rundown AI, TLDR AI ~30–60 min Low
Weekly catch-up The Batch, Import AI, Last Week in AI ~30 min Low–medium
Research Hugging Face Papers, arXiv, alphaXiv, Epoch AI Variable High
Tools Product Hunt, Future Tools, Hugging Face Hub, GitHub Trending Variable Low–medium
Benchmarks Arena, Artificial Analysis, OpenRouter Rankings ~10 min Medium
Deep analysis One Useful Thing, Simon Willison, Interconnects, Latent Space ~1–2 hours Medium
Listening/watching Lex Fridman, Dwarkesh, AI Explained, Matthew Berman Several hours Low–medium

Key takeaways

  • Four voices, four jobs: Matthew Berman for fast model news and tests, Andrej Karpathy for understanding how AI works, Ethan Mollick for AI at work and in education, Lex Fridman for long interviews.
  • Big audiences: Lex Fridman (5.56 million on X, ~5.07 million on YouTube) and Andrej Karpathy (4.40 million on X) reach the most people; Berman (~642,000 YouTube subscribers) and Mollick (491,000+ Substack subscribers) are central in their niches.
  • You don’t need X or Reddit: one daily newsletter (for example Forward Future or The Rundown AI), one weekly round-up (The Batch or Import AI) and one deep-analysis source (One Useful Thing or Simon Willison) cover most needs.
  • Check rankings against usage: test scores (Arena, Artificial Analysis) and real-world usage (OpenRouter) often tell different stories.
  • Almost everything is free: paid tiers mainly add extra posts or communities. Subscriber numbers are mostly self-reported, and some creator content is sponsored.

Sources

← Back to all posts