Top all-time · #14
Simon Willison
Personal blog · simonwillison.net · on HN as simonwaiprogramming
Simon Willison's personal blog at simonwillison.net covers ai and programming. Since 2007, 986 different Hacker News readers have linked it 3,298 times, most often “AI agent lethal trifecta: private data, untrusted content, exfiltration vectors”. It is still writing: the last post was on 2026-11-01, and you can follow it by feed.
Sometimes recommended · 6 of 38 sampled comments recommend it, 3 are critical?
Why HN links it
HN readers link his blog as a go-to source on LLMs: pelican-on-a-bicycle model tests, GPT-5 and Gemini posts, and examples of what models can and cannot do.
- Recommended for · 6 comments
- Called 'very well written' and 'some great examples'; examples show 'usefulness and limitations'.
- Criticised for · 3 comments
- Says a GPT-5 post is marketing-y and glosses over flaws; says a quote was cut; says a premise is no longer true.
Summarised by Claude from a sample of 38 HN comments that link here. Each comment is marked recommending, critical or neutral (citing without judging); how this works.
Most-linked posts?
- AI agent lethal trifecta: private data, untrusted content, exfiltration vectors45 people · 55 mentions · first linked 2025-06-16
- /37 people · 47 mentions · first linked 2018-01-26
- Git scraping: track changes over time by scraping to a Git repository15 people · 21 mentions · first linked 2020-10-09
- Not all AI-assisted programming is vibe coding (but vibe coding rocks)15 people · 19 mentions · first linked 2025-03-19
- "A computer can never be held accountable"13 people · 17 mentions · first linked 2025-02-03
- CaMeL offers a promising new direction for mitigating prompt injection attacks12 people · 28 mentions · first linked 2025-04-11
- /tags/pelican-riding-a-bicycle/12 people · 27 mentions · first linked 2024-12-16
- /12 people · 15 mentions · first linked 2022-02-07
- The Dual LLM pattern for building AI assistants that can resist prompt injection11 people · 20 mentions · first linked 2023-04-25
- What happens if AI labs train for pelicans riding bicycles?10 people · 15 mentions · first linked 2025-11-18
What Hacker News says?
Oh interesting, does that mean languages other than English won't be paying such a large penalty in terms of token lengths? With previous tokenizers there was a notable increase in the number of tokens needed to represent non-English sentences: https://simonwillison.net/2023/Jun/8/gpt-tokenizers/
Comment linking /2023/Jun/8/gpt-tokenizers/
on “GPT-4o”Submitted How to succeed in MrBeast production (Leaked PDF) as a story · 2115 points
HN discussionI've seen these called "explorables" or "explorable explanations" before and I really like them. I've been collecting notes on them here: https://simonwillison.net/tags/explorables/ Here's the website that coined the term: https://explorabl.es/
Comment linking /tags/explorables/
on “Moon”Submitted The killer app of Gemini Pro 1.5 is using video as an input as a story · 1136 points
HN discussionI've kind of lost track of where we disagree here, to be honest. Maybe we need to drop "easier" and "harder" and talk about speed. I can write software faster with LLMs, without sacrificing quality - in fact I can get higher quality software because doing things "better" doesn't mean they take longer. I derive enjoyment from building good stuff. If I can do it faster I can build more of it, which increases my enjoym…
Comment linking /2023/Mar/27/ai-enhanced-development/
on “My AI skeptic friends are all nuts”Submitted Xkcd 1425 (Tasks) turns ten years old today as a story · 1003 points
HN discussionI'd count ChatGPT memory as a feature of ChatGPT, not of the underlying LLM. I wrote a bit about that here - I've turned it off: https://simonwillison.net/2025/May/21/chatgpt-new-memory/
Comment linking /2025/May/21/chatgpt-new-memory/
on “My AI skeptic friends are all nuts”Submitted Watching o3 guess a photo's location is surreal, dystopian and entertaining as a story · 987 points
HN discussionNo matter how hard I train my fingers will never be able to move fast enough to output 100 lines of code in 15 seconds. When I get to writing actual production code with LLMs I treat them more as typing assistants than anything else: https://simonwillison.net/2025/Mar/11/using-llms-for-code/#t...
Comment linking /2025/Mar/11/using-llms-for-code/
on “My AI skeptic friends are all nuts”Submitted Things we learned about LLMs in 2024 as a story · 984 points
HN discussionPrices have been dropping like a stone over the past two years, due to a combination of new efficiencies being developed for serving plus competition from many vendors: https://simonwillison.net/2024/Dec/31/llms-in-2024/#llm-pric... I'm not seeing any evidence yet of that trend stopping or reversing.
Comment linking /2024/Dec/31/llms-in-2024/
on “My AI skeptic friends are all nuts”Submitted Promising results from DeepSeek R1 for code as a story · 979 points
HN discussionPlug for Simon's (very well written) longer form article about this topic: https://simonwillison.net/2025/Mar/11/using-llms-for-code/
Comment linking /2025/Mar/11/using-llms-for-code/
on “My AI skeptic friends are all nuts”Submitted The last six months in LLMs, illustrated by pelicans on bicycles as a story · 962 points
HN discussionI doubt I had much influence at all on the spread of vibe-coding. I stand by what I wrote about it though: https://simonwillison.net/2025/Mar/19/vibe-coding/ I think it's a net positive for regular humans to be able to build tools for their own personal use, and I think my section on "when is it OK to vibe code?" (only for low stakes projects, treat with extreme caution if private data or security is involved) is so…
Comment linking /2025/Mar/19/vibe-coding/
on “My AI skeptic friends are all nuts”Submitted NotebookLM's automatically generated podcasts are surprisingly effective as a story · 907 points
HN discussionThey do. All over the place. Simon Willison has some great examples in his blog and on his GitHub. Check out Karpathy’s YouTube videos as well. https://simonwillison.net/
Comment linking /
on “My AI skeptic friends are all nuts”Submitted Gemini Diffusion as a story · 890 points
HN discussionYeah, definitely both. New models come out all the time. One of the most interesting signals to look out for is when they tip over the quality boundary from "not useful at task X" to "useful at task X". It happened for coding about a year ago. It happened for search-based research assistants just two months ago, in my opinion - I wrote about that here: https://simonwillison.net/2025/Apr/21/ai-assisted-search/
Comment linking /2025/Apr/21/ai-assisted-search/
on “My AI skeptic friends are all nuts”Submitted Stable Diffusion is a big deal as a story · 874 points
HN discussion"they can't be aware of the latest changes in the frameworks I use, and so force me to use older features, sometimes less efficient" That's mostly solved by the most recent ones that can run searches. I've had great results from o4-mini for this, since it can search for the latest updates - example here: https://simonwillison.net/2025/Apr/21/ai-assisted-search/#la... Or for a lot of libraries you can dump the ENTIRE…
Comment linking /2025/Jun/2/claude-trace/
on “My AI skeptic friends are all nuts”Submitted Your job is to deliver code you have proven to work as a story · 847 points
HN discussionProgrammers who think front end is "easier" than backend have been wrong for well over a decade. https://simonwillison.net/2012/Feb/13/why-are-front-end/
Comment linking /2012/Feb/13/why-are-front-end/
on “My AI skeptic friends are all nuts”Submitted Claude Skills are awesome, maybe a bigger deal than MCP as a story · 738 points
HN discussion> "A computer can never be held accountable, therefore a computer must never make a management decision." Legendary quote from an internal 1979 IBM training document [1] https://simonwillison.net/2025/Feb/3/a-computer-can-never-be...
Comment linking /2025/Feb/3/a-computer-can-never-be-held-accountable/
on “DeepInfer: High-performance inference framework for DeepSeek models”Submitted Grok: Searching X for "From:Elonmusk (Israel or Palestine or Hamas or Gaza)" as a story · 734 points
HN discussionI made this one recently: https://www.youtube.com/watch?v=qy4ci7AoF9Y - notes here: https://simonwillison.net/2025/Nov/6/upgrading-datasette-plu... My best writing on this topic is still this though (which doesn't include a video): https://simonwillison.net/2025/Mar/11/using-llms-for-code/
Comment linking /2025/Mar/11/using-llms-for-code/
on “Anthropic acquires Bun”Submitted Claude 4 System Card as a story · 700 points
HN discussionI got this working with my LLM tool (new plugin version: llm-anthropic 0.14) and figured out a bunch of things about the model in the process. My detailed notes are here: https://simonwillison.net/2025/Feb/25/llm-anthropic-014/ One of the most exciting new capabilities is that this model has a 120,000 token output limit - up from just 8,000 for the previous Claude 3.5 Sonnet model and way higher than any other model…
Comment linking /2025/Feb/25/llm-anthropic-014/
on “Claude 3.7 Sonnet and Claude Code”Submitted Notes on OpenAI's new o1 chain-of-thought models as a story · 699 points
HN discussion
