In a moment of extreme irony and hilarity, Claude Code doesn't know how to properly edit it's own settings file, so it hallucinated it and broke it. 🤓

UC San Diego is discovering shocking grade inflation at many Cali high schools, to the point that a large minority of kids with a 4.0 average in math are at 6th grade or below in mathematical ability. In response, they’re resorting to semi-blacklisting schools with especially bad grade inflation.

Oh damn I should try Gemini 3.0 on Shenzhen I/O. 2.5 Pro can almost but not quite solve it from a problem description and the manual.

The "AI Bubble" isn't a fake technology bubble (like crypto/NFTs arguably were); it's a Valuation and Infrastructure Bubble. The tech is real, but the prices are wrong.

People say the em dash is the greatest tell that something is written by an LLM, but actually it is using emoji in code.

“No way to prevent this” says only language where this regularly happens

When Gemini gets stuck in a "test broken -> change code -> test still broken" loop, it's always the right move to spin up a new instance to tackle just that problem. Otherwise the main agent will go insane. But it definitely feels like cheating on Gemini... 🤔

A thoughtful post by Anil Dash the fact that users want to use AI. www.anildash.com/2025/11/14/w...

if you conceive of a LLM as "sophisticated software" which is essentially accessing a database, you will get every policy question about LLMs wrong.

Part of the reason why I’m so insistent about folks understanding AI capabilities is that they’re here to stay and we need to start thinking about what to do in such a world. Putting the genie back in the bottle is a pleasant fantasy that delays serious reckoning

I enjoyed the first book in the Honor Harrington series, On Basilisk Station, although it's clear Weber is still trying to find his feet. I particularly enjoy its depiction of professional competency, especially in... these times. 😅 alexgude.com/books/on_bas... #BookReview

side note: the og:image generator that screenshots the blog post at a readable text size is really neat screenshots.jamesg.blog?url=https://...

one of these is a mindless machine churning out terrible takes and the other is Grok

This is somehow the fault of the individual AI researchers here. To compensate, I've been negatively polarized into being even more of an anti-copyright extremist.

Oh so Walter Miller can write huh? Why didn't anyone tell me this?! (Everyone told me this.)

FAFO

All right I'm putting my marker down: Prop 50 over 60%

STATE OF THE ART just literally has Diziet Sma go to Earth, take a quick look at it and endorse socialism while condemning misogyny, looking directly into the camera, it's incredible

[2/9] We argue that instead of getting stuck on metaphysical debates (is AI conscious?), we should treat personhood as a flexible bundle of obligations (rights & responsibilities) that societies confer.

I've made the same-ish plot using California data! Be careful out there! alexgude.com/blog/switrs-...

I am the very model of an ancient trunkless leg of stone, My shattered visage lies half buried with a sandy sneering frown, The sculptor carved my arrogance upon these broken lifeless things, My name is Ozymandias, the self-describèd King of Kings!

I jump into the moderation debate. I agree with @adambonica.bsky.social and @gelliottmorris.com on the methodology, but I want to draw attention to a bigger issue. The collapse of candidate effects generally. leedrutman.substack.com/p/the-modera...

A fun project here for you to fight over, from @philedwardsinc.bsky.social: How Midwestern is Each State?! mostmidwest.com

an awful lot of why politics is so deranged now in four easy steps:

😐

unbelievbable. i cannot believe you are going after a disabled shepard for his dietary preferences.

In important math news: new shape just dropped (the noperthedron) www.quantamagazine.org/first-shape-...

Rescue the Tilden steam train! Email [email protected] with the subject line "Public Comment- Tilden Steam Train". bayareatelegraph.com/2025/10/24/t...

people saying ai is a fad don't know what they're talking about. i'm definitely not giving up my claude now; the ability to do structural refactors "change this thing to be layered like this thing in this other project" is fantastic. i *do* want the model to be local and hope we'll get there in time

New: The market for Counter-Strike skins fell $1.75 billion overnight -- that's 25%. Investors dumped money into the digital item market amid concerns about the S&P 500 and crypto. Then, last night, Valve pushed a market-breaking update to the game

I recorded a ten minute video showing my vibe-coding process for building a tool for sharing formatted terminal sessions via copy and paste using the new Claude Code for web - now available on YouTube here www.youtube.com/watch?v=GQvM... More notes on my blog: simonwillison.net/2025/Oct/23/...

Skeptical of using AI agents for docs? What about hallucinations? Join Emma, Senior Technical Writer at Block, next Thursday as she uses goose to write its own documentation, and shares strategies for doc creation and validation with agentic AI. 📆 October 30, 9am PT/ 12pm ET 📺 linktr.ee/goose_oss

And we're live! Starting today, you can use Bounce to migrate from Mastodon to an existing Bluesky, Blacksky, or any other ATProto account! Check it out: blog.anew.social/bounce-beta-...

approaching the 10 year anniversary of the 750 diphoton excess debacle. you probably have no idea wtf i am talking if you weren't in particle physics at the time, but i think we can actually learn a lot from it. it was as if we collectively decided there would be no social cost to publishing BS.

The good news for Californians is that while tech jobs continue to leave the state, the remaining jobs are disproportionately high-paying The share of information sector compensation going to CA is actually up since 2029 even as the state’s share of employment has shrunk rapidly

NEW from me: California keeps losing tech jobs Amidst the AI boom, the US is in the largest sustained drawdown in tech employment since the dot-com bust And what tech jobs remain continue to leave California for the rest of the US 🧵 www.apricitas.io/p/california...

We did it! 🥰

the owner of the pronouns labeler using it for moderation (blocking/hiding their account, idk what else) is crazy. Users should have more control over how labelers control their experience 😵‍💫

there’s another angle — what’s the goal software engineers have always gotten tricked into thinking the goal is to write “good code”. but the goal has always been high velocity of impactful software there’s a lot of cognitive dissonance in the industry over that

Blocked. BLOCKED. *BLOCKED*! You're all blocked! None of you are free of sin!

cool stuff. notable effect of model size. the biggest in the list by far, Gemini 2.5 Pro, dramatically outperforms the rest. other SOTA and SOTA-adjacent models (GPT-5, Claude Sonnet and Opus, Qwen Max, Deepseek, Kimi K2) were not tested

This thread is a *FANTASTIC* look at LLMs!

this is like shaking a bag of watch parts and dumping out a fully assembled watch. many people say that this is impossible. the people who say that result from an even more disorganized bag shaken by an even stupider process.

so, okay. based on this framework, my answers to a bunch of questions about LLMs:

I'm beginning to suspect that a key skill in working effectively with coding agents is developing an intuition for when you don't need to closely review every line of code they produce. This feels deeply uncomfortable!

I think a LOT of AI pushback from developers comes from taking the job too seriously in the ways that don't matter, and not seriously enough in the ways that do. Claude doesn't write perfect code? Well uh, neither do I? And if you are maybe that's not the best use of your time?

Not really wanting to start a fight but if I can use it to 1) set up basic CRUD around some new tables 2) hook up basic apis around that, 3) read some random library's docs that I'll never use again for the logic.. that lets me focus on the parts I really do like, architecting it all at a high level

A great step towards fixing our housing crisis! It feels especially good because I was really invested in it: I talked to my representatives in person, calling them up when they wavered later. Still a lot to do, but we'll get it done! 🥳