how to work around an agent harness's annoying propensity to use full-cost subagents with one weird trick
Max Woolf@minimaxir.bsky.social · 2d ago
Max Woolf
@minimaxir.bsky.social
(former) Senior Data Scientist at BuzzFeed in San Francisco // AI content generation ethics and R&D // plotter of pretty charts https://minimaxir.com
Max Woolf@minimaxir.bsky.social · 2d ago
So, uh, personal announcement. I am now suddenly unemployed. I am currently looking for a data science/machine learning/AI job in the SF Bay Area. If you're interested, let me know! In the meantime, I now have *plenty* of time for blogging lol.
If I ever work for Anthropic or OpenAI I'm just going to not post on social media.
This utter muppet of a conservative Canadian politician in New Brunswick forgot to remove the LLM text from his floor speech and read it out loud.
New blog post up: I saw the memes of showing LLMs the Jacobian Conjecture so I wanted to give it a try... minimaxir.com/2026/07/jaco...
LLMs break down in funny ways when told the Jacobian Conjecture counterargument
Cognitohazards can affect machines too.
minimaxir.com
This headline is extremely funny given what happened (OpenAI hacked HF by accident) openai.com/index/huggin...
OpenAI and Hugging Face partner to address security incident during model evaluation
OpenAI and Hugging Face share early findings from a security incident during AI model evaluation, highlighting advanced cyber capabilities and lessons for defenders.
openai.com
Comments about the quota reset aside, the Codex growth rate is absurd and I wonder how much of it is actually due to the macOS app bundling with ChatGPT shenanigan
Does "Jacobean Conjecture" capitalize the C? I'm seeing it done differently everywhere and I want the correct style for a blog post.
Prompt engineering is less important nowadays since agents can handle some ambiguity, but the less ambiguity, the more **productive** the agent is. It also adds checks against being too smart and cheating.
mr. TIM@timkellogg.me · 2w agoif you read the prompts for these big math problems, they’re non-trivial i like to laugh at how prompt engineering is dead, but goddamn, you still need a working mental model for how LLMs think, and that’s not easy
I'm still upset that with all the new LLMs coming out, no one has release a good OSS text embedding model at a reasonable size. EmbeddingGemma was 10 months ago ffs.
Sure, why not, here are the prompts I used: gist.github.com/minimaxir/30... If I don't complain about something in a followup prompt, assume it worked correctly. No I will not explain the prompts. Not yet, anyways.
Max Woolf@minimaxir.bsky.social · 3w agoI used GPT 5.6 to create something I've wanted for literally years: a macOS menu bar application to control an Apple TV with all expected features (including Now Playing), written using SwiftUI. It worked *much* better than expected.
ok I'm getting annoyed at the nonneligible number of people on HN who are armchair diagnosing this as a **gambling** addiction Funnily the entire reason I went into statistics/data science in the first place is because I'm absurdly risk-adverse.
Max Woolf@minimaxir.bsky.social · 3w agoNew short blog post up: what's the deal with all the random weekly quota resets for agents lately?
the frustrating part of bad ai takes is that nobody who has them wants to stand by them or admit their misses later. it is just the rocket goalposts game.
New short blog post up: what's the deal with all the random weekly quota resets for agents lately?
What's the deal with all the random weekly quota resets for agents lately?
I’m going to be the weirdo who complains about literally getting free stuff.
minimaxir.com
I used GPT 5.6 to create something I've wanted for literally years: a macOS menu bar application to control an Apple TV with all expected features (including Now Playing), written using SwiftUI. It worked *much* better than expected.
I really want to see a economic thinkpiece about the deal with random quota resets for agent subscriptions.
