Top all-time · #755
Max Woolf
Personal blog · minimaxir.com · on HN as minimaxirdataaiprogramming
Max Woolf's personal blog at minimaxir.com covers data, ai and programming. Since 2012, 86 different Hacker News readers have linked it 508 times, most often “Tempering Expectations for GPT-3 and OpenAI’s API”. It is still writing: the last post was on 2026-10-25, and you can follow it by feed.
Sometimes recommended · 6 of 39 sampled comments recommend it?
Why HN links it
HN readers cite Max Woolf's posts as sources on LLM prompting, GPT-3 expectations, Hacker News data analysis and embeddings. Most comments cite them neutrally; a few praise them.
- Recommended for · 6 comments
- Called a great, less-bloated GPT-3 article; one analysis liked; thanked, and suggested for reading on LLM tips.
Summarised by Claude from a sample of 39 HN comments that link here. Each comment is marked recommending, critical or neutral (citing without judging); how this works.
A blog written by data scientist Max Woolf out of San Francisco.
Most-linked posts?
- Tempering Expectations for GPT-3 and OpenAI’s API6 people · 7 mentions · first linked 2020-07-18
- Does offering ChatGPT a tip cause it to generate better text?6 people · 6 mentions · first linked 2024-02-23
- Can LLMs write better code if you keep asking them to “write better code”?5 people · 6 mentions · first linked 2025-01-02
- An Introduction on How to Make Beautiful Charts with R and ggplot24 people · 16 mentions · first linked 2015-02-12
- The Quality, Popularity, and Negativity of 5.6M Hacker News Comments4 people · 16 mentions · first linked 2014-10-06
- A Statistical Analysis of All Hacker News Submissions4 people · 9 mentions · first linked 2014-02-24
- Benchmarking TensorFlow on Cloud CPUs: Cheaper Deep Learning Than Cloud GPUs4 people · 5 mentions · first linked 2017-07-05
- Imgur is Covertly Redirecting Image Links on Facebook and Twitter3 people · 8 mentions · first linked 2014-02-06
- How to Analyze Every Reddit Submission and Comment, in Seconds, for Free3 people · 8 mentions · first linked 2015-10-02
- Experiments with Making Convincing AI-Generated Fake News3 people · 6 mentions · first linked 2019-09-30
What Hacker News says?
Sure, but you can still re-prompt them again telling them to just do better . In case people missed it, I'm referencing an observation recently made by 'minimaxir, described here: https://minimaxir.com/2025/01/write-better-code/ As it turns out, you can improve the quality of code generated by some LLMs by repeatedly responding to it with just three words: "write better code". While Max only tested this on Claude 3.…
Comment linking /2025/01/write-better-code/
on “DeepSeek-R1”Submitted Nano Banana can be prompt engineered for nuanced AI image generation as a story · 887 points
HN discussionIn February, I did an analysis on all Hacker News stories: http://minimaxir.com/2014/02/hacking-hacker-news/ Yesterday, I published an analysis on all Hacker News comments: http://minimaxir.com/2014/10/hn-comments-about-comments/ There's a lot of interesting trends in the data. Let me know if you want to know anything in particular and I'll get back to you. :)
Comment linking /2014/10/hn-comments-about-comments/
on “Hacker News API”Submitted Can LLMs write better code if you keep asking them to “write better code”? as a story · 812 points
HN discussionThat's less and less true https://minimaxir.com/2025/11/nano-banana-prompts/
Comment linking /2025/11/nano-banana-prompts/
on “AI World Clocks”Submitted OpenAI and Anthropic cut their LLM prices by up to 80% within a few months. Why? as a story · 443 points
HN discussionUsing the data from a previous HN analysis of all posts until Feb 2014 [1], I've found that 1.45% of all submissions are Show HN submissions. (but that only found 18,000 submissions. If you use the given 24,000 submissions and the current number of articles [1.3m], the percentage is 1.75%) I have the data loaded if you have any other Show HN questions. [1] http://minimaxir.com/2014/02/hacking-hacker-news/
Comment linking /2014/02/hacking-hacker-news/
on “Show HN: Show HN”Submitted As an experienced LLM user, I don't use generative LLMs often as a story · 388 points
HN discussionThat result for rendering HTML to an image (the Counter Info one) is pretty impressive. https://github.com/minimaxir/gemimg/blob/main/docs/files/cou... to this: https://x.com/minimaxir/status/1991580127587921971 - see also https://minimaxir.com/2025/11/nano-banana-prompts/#image-pro...
Comment linking /2025/11/nano-banana-prompts/
on “Nano Banana Pro”Submitted Pushing ChatGPT's Structured Data Support to Its Limits as a story · 304 points
HN discussionDoes the refrigerator magnet system prompt leak [1] still work? [1] https://minimaxir.com/2025/11/nano-banana-prompts/#hello-nan... .
Comment linking /2025/11/nano-banana-prompts/
on “Nano Banana Pro”Submitted I Resurrected “Ugly Sonic” with Stable Diffusion Textual Inversion as a story · 289 points
HN discussion>> - Put a strawberry in the left eye socket. >>- Put a blackberry in the right eye socket. >> All five of the edits are implemented correctly This is a GREAT example of the (not so) subtle mistakes AI will make in image generation, or code creation, or your future knee surgery. The model placed the specified items in the eye sockets based on the viewers left/right; when we talk relative in this scenario we usually…
Comment linking /2025/11/nano-banana-prompts/
on “Nano Banana Pro”Submitted Game Theory: How 70,000 Pokemon Players Sabotage Themselves as a story · 289 points
HN discussionIn case anyone missed Max's Nano Banana prompting guide, it's absolutely the definitive manual for prompting the original Nano Banana... and I tried some of the prompts in there against Nano Banana Pro and found it to be very applicable to the new model as well. https://minimaxir.com/2025/11/nano-banana-prompts/#hello-nan... My recreations of those pancake batter skulls using Nano Banana Pro: https://simonwillison.n…
Comment linking /2025/11/nano-banana-prompts/
on “Nano Banana Pro”Submitted The Decline of Imgur on Reddit and the Rise of Reddit's Native Image Hosting as a story · 270 points
HN discussionjust sharing this great minimaxir article countering the hype: https://minimaxir.com/2020/07/gpt3-expectations/
Comment linking /2020/07/gpt3-expectations/
on “OpenAI's GPT-3 may be the biggest thing since Bitcoin”Submitted Tempering Expectations for GPT-3 and OpenAI’s API as a story · 267 points
HN discussionI published a response today to the sudden hype urging people to temper their expectations for GPT-3 a bit: https://minimaxir.com/2020/07/gpt3-expectations/ GPT-3 is objectively a step forward in the field of AI text-generation, but the current hype on VC Twitter misrepresents the model's current capabilities. GPT-3 isn't magic.
Comment linking /2020/07/gpt3-expectations/
on “OpenAI's GPT-3 may be the biggest thing since Bitcoin”Submitted Does offering ChatGPT a tip cause it to generate better text? as a story · 265 points
HN discussionAs someone who's done exactly that with the Reddit data ( http://minimaxir.com/2016/06/reddit-related-subreddits/ ), posts/comments aren't as useful for intent as you note. The valuable Reddit data is the nonpublic views/subscription/upvote/downvote behavior.
Comment linking /2016/06/reddit-related-subreddits/
on “Facebook Security Chief Said to Leave After Clashes Over Disinformation”Submitted Benchmarking TensorFlow on Cloud CPUs: Cheaper Deep Learning Than Cloud GPUs as a story · 258 points
HN discussionAnother fun thing to do with using SVG-as-images in the browser is easy animation of the SVG, using Primitive (for example) as noted in the article to convert an image into SVG shapes, plus a tool like Vivus ( https://maxwellito.github.io/vivus/ ) or Snap.svg ( http://snapsvg.io ) to animate each shape of the SVG. Here’s a demo I made using Primitive + Vivus: http://minimaxir.com/2016/12/primitive/
Comment linking /2016/12/primitive/
on “Using SVG as image placeholders”Submitted Please Do Not Steal My Code, Mock My Analysis, and Present My Ideas as Your Own as a story · 257 points
HN discussionI wrote an article a few months ago on the differences between R Notebooks and Jupyter Notebooks (and why, IMO, R Notebooks are better): http://minimaxir.com/2017/06/r-notebooks/
Comment linking /2017/06/r-notebooks/
on “JupyterLab is ready for users”Submitted The best way to use text embeddings portably is with Parquet and Polars as a story · 247 points
HN discussionReddit's public-facing data is very useful for ML/AI. On the data science end, you can do a lot to gauge important topics and user behavior: https://minimaxir.com/2018/09/modeling-link-aggregators/ On the silly AI-end, I made a subreddit consisting only of text-generating RNNs: https://www.reddit.com/r/SubredditNN/ Reddit's internal data is even more robust.
Comment linking /2018/09/modeling-link-aggregators/
on “Reddit is raising a huge round near a $3B valuation”Submitted Leaving Apple Inc as a story · 239 points
HN discussiontips: do prompt engineering like you're programming VQGAN GAN image generators can be tricked into giving results with phrases like 'unreal engine', 'artstation', 'junjo ito', 'van gogh' https://minimaxir.com/2021/08/vqgan-clip/ same for google. 'gist' gives you a specific category of technical results. seems like the default is some combination of medium, SO + github issues
Comment linking /2021/08/vqgan-clip/
on “Ask HN: Has Google search become quantitatively worse?”Submitted Visualizing How Developers Rate Their Own Programming Skills as a story · 225 points
HN discussion
