Skip to content
Dwarkesh PodcastDwarkesh Podcast

Gwern — Anonymous writer who predicted AI trajectory on $12K/year salary

Gwern's blog: https://gwern.net/. Gwern is a pseudonymous researcher and writer. After the episode, I convinced Gwern to create a donation page where people can help sustain what he's up to. Please go here to contribute: https://donate.stripe.com/6oE9DTgaf6oD0M03cc. Thank you to my friend Chris Painter for doing an amazing job voice acting Gwern. 𝐒𝐏𝐎𝐍𝐒𝐎𝐑𝐒 * Jane Street is looking to hire their next generation of leaders. Their deep learning team is looking for ML researchers, FPGA programmers, and CUDA programmers. Summer internships are open - if you want to stand out, take a crack at their new Kaggle competition. To learn more, go here: https://jane-st.co/dwarkesh * Turing provides complete post-training services for leading AI labs like OpenAI, Anthropic, Meta, and Gemini. They specialize in model evaluation, SFT, RLHF, and DPO to enhance models’ reasoning, coding, and multimodal capabilities. Learn more at https://turing.com/dwarkesh. * This episode is brought to you by Stripe, financial infrastructure for the internet. Millions of companies from Anthropic to Amazon use Stripe to accept payments, automate financial processes and grow their revenue. Learn more here: https://stripe.com/ 𝐄𝐏𝐈𝐒𝐎𝐃𝐄 𝐋𝐈𝐍𝐊𝐒 * Transcript: https://www.dwarkeshpatel.com/p/gwern-branwen * Me on Twitter: https://twitter.com/dwarkesh_sp * Spotify: https://open.spotify.com/episode/46H5dTtYaj1L55UAy9XXaY?si=xVoj6euwQdmZYnyvaQ46lA 𝐓𝐈𝐌𝐄𝐒𝐓𝐀𝐌𝐏𝐒 00:00:00 - Anonymity 00:01:09 - Automating Steve Jobs 00:04:38 - Isaac Newton's theory of progress 00:06:36 - Grand theory of intelligence 00:10:39 - Seeing scaling early 00:21:04 - AGI Timelines 00:22:54 - What to do in remaining 3 years until AGI 00:26:29 - Influencing the shoggoth with writing 00:30:50 - Human vs artificial intelligence 00:33:52 - Rabbit holes 00:38:48 - Hearing impairment 00:43:00 - Wikipedia editing 00:47:43 - Gwern.net 00:50:20 - Counterfactual careers 00:54:30 - Borges & literature 01:01:32 - Gwern's intelligence and process 01:11:03 - A day in the life of Gwern 01:19:16 - Gwern's finances 01:25:05 - The diversity of AI minds 01:27:24 - GLP drugs and obesity 01:31:08 - Drug experimentation 01:33:40 - Parasocial relationships 01:35:23 - Open rabbit holes

Dwarkesh PatelhostGwern Branwenguest
Nov 13, 20241h 36mWatch on YouTube ↗

At a glance

WHAT IT’S REALLY ABOUT

Anonymous polymath Gwern on AI scaling, anonymity, and obsessive rabbit holes

  1. Gwern Branwen discusses how anonymity lets his ideas be judged without personal projection, and reflects on his role as an independent, low-budget researcher who heavily influenced modern AI scaling thinking. He outlines a grand, compute-centric view of intelligence as search over Turing machines, explains how he correctly anticipated LLM scaling when most commentators didn’t, and sketches near-term futures of AI-run firms with human "taste" at the top. The conversation dives into his working habits, rabbit-hole-driven creativity, trade-offs of isolation and poverty for deep work, and his belief that now is a uniquely "hinge" time to write, both to shape AI values and to preserve a personal legacy in latent space. He closes by listing big unresolved questions about intelligence, civilization, and human variation he hopes superhuman AIs will finally answer by 2050.

IDEAS WORTH REMEMBERING

5 ideas

Anonymity buys a fair hearing by stripping away identity-based bias.

Gwern argues that being anonymous forces people to engage with the text itself rather than preemptively dismissing him based on status, demographics, or affiliations, and also protects him from retaliation for controversial topics.

Human-led AI firms will likely win by combining AI scale with human long-term taste.

He predicts bottom-up automation where AI replaces workers first, leaving a small number of human "Steve Jobs"-like executives who provide long-horizon vision and taste while pyramids of AI agents execute and propose options.

Intelligence is best viewed as compute-intensive search over many small programs.

Rather than a single master algorithm or "intelligence fluid," Gwern sees brains and large models as ensembles of many specialized solutions (Turing machines), with more intelligent agents simply having more compute to search and recombine them.

Scaling success came from compute, data, and trial-and-error—not magical algorithms.

His belief in the scaling hypothesis emerged from years of tracking deep learning trends (AlexNet, CNNs, AlphaZero, early scaling-law papers), noticing that bigger models plus more data kept broadening capabilities, while the field systematically underreported the role of brute-force experimentation.

Now is an unusually leverageable time to write because AI trains on everything.

He claims that text online directly shapes future models’ behavior and values; if your preferences and viewpoints are not written down, they effectively don’t exist to AI systems, which is dangerously close to not existing at all in future influence terms.

WORDS WORTH SAVING

5 quotes

The most underrated benefit of anonymity is that people don’t project onto you as much… everyone has to read you at least a little bit to even begin to dismiss you.

Gwern Branwen

All intelligence is search over Turing machines… there’s no master algorithm and no special intelligence fluid.

Gwern Branwen

You’re voting on the future of the Shoggoth using some of the few currencies it acknowledges: tokens that it has to predict.

Gwern Branwen

Magic is putting in more effort than any reasonable person would expect you to.

Teller, quoted by Gwern Branwen

I maximize rabbit holes… It’s the sudden new area I can fall into and obsess over that I really live for.

Gwern Branwen

Benefits and costs of anonymity for intellectual workFuture of AI in firms, agency, and units of selectionScaling hypothesis, AI timelines, and why others missed LLMsGrand theory of intelligence as search over Turing machinesGwern’s process: rabbit holes, Wikipedia apprenticeships, and writing practiceEconomics and lifestyle of being an independent online researcherInfluencing AI through writing and preserving values in model training data

High quality AI-generated summary created from speaker-labeled transcript.

Get more out of YouTube videos.

High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.

Add to Chrome