Top all-time · #2451
Sayash Kapoor
Personal blog · aisnakeoil.com/paipoliticseconomics
Sayash Kapoor writes AI as Normal Technology, a personal blog at aisnakeoil.com about ai, politics and economics. Since 2023, 31 different Hacker News readers have linked it 53 times, most often “Is GPT-4 getting worse over time?: A new paper has been widely misinterpreted”. It is still writing: the last post was on 2026-10-19, and you can follow it by feed.
Too few sampled comments (14) to say how often HN recommends it.?
Why HN links it
HN readers link this Substack by Princeton AI researchers for critical breakdowns of AI hype, such as GPT-4 getting worse over time and inference-time scaling. Commenters mostly cite its posts as sources.
Summarised by Claude from a sample of 14 HN comments that link here. Each comment is marked recommending, critical or neutral (citing without judging); how this works.
Analyzing AI as transformative but normal technology, not superintelligence.
Most-linked posts?
- Is GPT-4 getting worse over time?: A new paper has been widely misinterpreted9 people · 10 mentions · first linked 2023-07-19
- Is AI progress slowing down?5 people · 6 mentions · first linked 2024-12-19
- AI safety is not a model property3 people · 4 mentions · first linked 2024-03-12
- AI Scaling Myths3 people · 3 mentions · first linked 2024-06-27
- Does ChatGPT have a liberal bias?2 people · 2 mentions · first linked 2023-08-18
- Generative AI's end-run around copyright won't be resolved by the courts2 people · 2 mentions · first linked 2024-01-22
- AI existential risk probabilities are too unreliable to inform policy2 people · 2 mentions · first linked 2024-07-26
- AGI Is Not a Milestone2 people · 2 mentions · first linked 2025-05-01
- Could AI Slow Science?2 people · 2 mentions · first linked 2025-07-16
What Hacker News says?
Yes, some think it’s happening now: https://www.aisnakeoil.com/p/is-ai-progress-slowing-down
Comment linking /p/is-ai-progress-slowing-down
on “My AI skeptic friends are all nuts”Submitted AI companies are pivoting from creating gods to building products as a story · 133 points
HN discussionSince techbros are stupid: "Note that people could always do these kinds of repurposing, and it was never a problem from a copyright perspective. We have a problem now because those things are being done (1) in an automated way (2) at a billionfold greater scale (3) by companies that have vastly more power in the market than artists, writers, publishers, etc. Incidentally, these three reasons are also why AI apologi…
Comment linking /p/generative-ais-end-run-around-copyright
on “Nightshade: An offensive tool for artists against AI art generators”Submitted Scientists should use AI as a tool, not an oracle as a story · 124 points
HN discussionThe new idea is inference-time scaling, as seen in o1 (and o3 and Qwen's QwQ and DeepSeek's DeepSeek-R1-Lite-Preview and Google's gemini-2.0-flash-thinking-exp). I suggest reading these two pieces about that: - https://www.aisnakeoil.com/p/is-ai-progress-slowing-down - best explanation I've seen of inference scaling anywhere - https://arcprize.org/blog/oai-o3-pub-breakthrough - François Chollet's deep dive into o3 I…
Comment linking /p/is-ai-progress-slowing-down
on “GPT-5 is behind schedule”Submitted Is the UK's liver transplant matching algorithm biased against younger patients? as a story · 93 points
HN discussionJust as a followup for those interested, after reading the paper and some more critique: The paper is mostly ok in that it points out model drift and the importance of making sure you use API versions. One good thing is that the full dataset/methodology was published on Github: https://github.com/lchen001/LLMDrift (everyone should do this!) so it was easy to replicate/validate, but there are some issues: * Simon Boe…
Comment linking /p/is-gpt-4-getting-worse-over-time
on “GPT-4 is getting worse over time, not better”Submitted AI Scaling Myths as a story · 44 points
HN discussionLike clockwork, it's coming out that the original paper was wildly misinterpreted: https://www.aisnakeoil.com/p/is-gpt-4-getting-worse-over-tim...
Comment linking /p/is-gpt-4-getting-worse-over-time
on “GPT-4 is getting worse over time, not better”Submitted AI existential risk probabilities are too unreliable to inform policy as a story · 42 points
HN discussionThat was a pretty deeply flawed paper, one of the largest drops recorded was simple parsing errors in their testing: https://www.aisnakeoil.com/p/is-gpt-4-getting-worse-over-tim... Overall evals and pinning against checkpoints are how you avoid those worries, but in general, if you solve a problem robustly, it's going to be rare for changes in the LLM to suddenly break what you're doing. Investing in handling a wide…
Comment linking /p/is-gpt-4-getting-worse-over-time
on “Azure ChatGPT: Private and secure ChatGPT for internal enterprise use”Submitted AI leaderboards are no longer useful. It's time to switch to Pareto curves as a story · 40 points
HN discussionWhat this work actually shows is that a bunch of scientists are ignorant about how LLMs work, but are rushing to publish papers about them anyway. It is ridiculous for Thirunavukarasu to draw this conclusion from GPT-4's performance on a written ophthalmology exam. From the good folks at AI Snake Oil[1] > Memorization is a spectrum. Even if a language model hasn’t seen an exact problem on a training set, it has inev…
Comment linking /p/gpt-4-and-professional-benchmarks
on “LLMs approach expert-level clinical knowledge and reasoning in ophthalmology”Submitted AGI Is Not a Milestone as a story · 36 points
HN discussionTangentially related: "AI safety is not a model property" ( https://www.aisnakeoil.com/p/ai-safety-is-not-a-model-proper... ) https://news.ycombinator.com/item?id=39939997
Comment linking /p/ai-safety-is-not-a-model-property
on “Asking BigTech to police AI is like asking oil companies to solve climate change”Submitted AI Agents That Matter as a story · 35 points
HN discussionThe study itself is somewhat flawed: https://www.aisnakeoil.com/p/is-gpt-4-getting-worse-over-tim...
Comment linking /p/is-gpt-4-getting-worse-over-time
on “Study shows over time, ChatGPT lost accuracy”Submitted AI existential risk probabilities are too unreliable to inform policy as a story · 15 points
HN discussionOn most of their tests gpt-4 is not actually worse [1]. In particular coding results are affected by changed due to different output format rather than worse abilities [2]. But that's ok because the message of the paper is that there is strong drift between versions and developers should be aware of it, not that gpt becomes worse [3]. [1] https://www.aisnakeoil.com/p/is-gpt-4-getting-worse-over-tim... [2] https://tw…
Comment linking /p/is-gpt-4-getting-worse-over-time
on “Show HN: Beating GPT-4 on HumanEval with a fine-tuned CodeLlama-34B”Submitted Is AI progress slowing down? as a story · 10 points
HN discussionThe authors' substack is great. They're sort of like investigative journalists who pick a topic that's got a lot of hype and then dissect it to (often very quickly) find that the hyped up topic was just hot air. Some examples: The claims of "liberal bias" being just a completely bunk study: https://www.aisnakeoil.com/p/does-chatgpt-have-a-liberal-bia... No evidence of GPT-4 getting worse over time: https://www.aisna…
Comment linking /p/is-gpt-4-getting-worse-over-time
on “Princeton ‘AI Snake Oil’ authors say GenAI hype has ‘spiraled out of control’”Submitted We Looked at 78 Election Deepfakes. Political Misinfo Is Not an AI Problem as a story · 7 points
HN discussionThis should be a link to https://www.aisnakeoil.com/p/is-ai-progress-slowing-down - my post is just a summary of some of the highlights, but the whole thing is fantastic and well worth reading. Here's a HN submission for that article: https://news.ycombinator.com/item?id=42464357
Comment linking /p/is-ai-progress-slowing-down
on “Is AI progress slowing down?”Submitted Is AI progress slowing down? as a story · 5 points
HN discussionWhen saying LLMs are "good at writing code" there's a distinction between "good at taking high-level English-language solutions to programming problems and translating them into low-level implementations via a specific language / design patterns / etc" versus "good at finding solutions to programming problems." GPT-4 is indeed quite good at the former - and the former is what most enterprise programming work actuall…
Comment linking /p/gpt-4-and-professional-benchmarks
on “Show HN: Microagents: Agents Capable of Self-Editing Their Prompts / Python Code”Submitted Could AI Slow Science? as a story · 4 points
HN discussionhttps://www.aisnakeoil.com/p/is-gpt-4-getting-worse-over-tim... > The math questions were of the form “Is 17077 prime”? They picked 500 numbers, but all of them were prime! > The June version of GPT-3.5 and the March version of GPT-4 almost always conclude that the number is prime regardless of whether it is prime or composite. The other two models do the opposite. But the paper only tested prime numbers, and hence…
Comment linking /p/is-gpt-4-getting-worse-over-time
on “(Databricks cofounder) How Is ChatGPT’s Behavior Changing over Time?”Submitted Is AI progress slowing down? as a story · 4 points
HN discussionSubmitted The AI Snake Oil Book as a story · 4 points
HN discussionSubmitted AI Scaling Myths as a story · 4 points
HN discussion
