pov - you can blur an object and eyeball its optical center for precise alignment

I've been trying to articulate why the fawning, complimentary responses from AI chatbots feel so insidious to me. I've figured out how to explain it. Wrote a piece on how current model training and design choices threaten our critical thinking skills: maggieappleton.com/ai-enlightenme…

So October 15th, the extended US tax deadline, is just around the corner, and I have some observations which are more about LLM progress than taxes. Background: many people professionally involved with LLMs estimate 2026-2028 as the year where one can get an LLM to "do taxes."

In today’s Transformer, I wrote about the very scary disconnect I felt between the conversations I was having with people like Joshua Achiam and Jack Clark at The Curve, and the way the wider world talks about AI.

Do you know where to buy the cheapest turpentine?

feature not a bug I would also like to be woken up when prod goes down

Tiny Recursive Models: A tiny 7M parameter model that recursively refines its answer beats LLMs 100x larger on hard puzzles like ARC-AGI We independently reproduced the paper, corroborated results, and released the weights + API access for those looking to benchmark it 🔍

Training LLMs end to end is hard. Very excited to share our new blog (book?) that cover the full pipeline: pre-training, post-training and infra. 200+ pages of what worked, what didn’t, and how to make it run reliably huggingface.co/spaces/Hugging…

Playing with the idea of using Serrif as a font for RYBitten.

this came to me in a dream

Many AI agent problems are really just information retrieval problems. If the agent has a better way to find and comb through data performantly, you will get far better results. Compute is fungible so you can use it during indexing and processing or you can use later to crank

godspeed

VCs giving their expert advice: “Build something people want” “Talk to your users” “Hire A players” “Focus on distribution” Founders:

Every node in your graph is a little death of inspiration to me. You've turned imagination into plumbing and disguised it as “creative freedom”. You've mistaken complexity for control and turned creation into workflow theater. Let us build worlds, not workflows. Don't give me

absolutely lovely to see this. kinda reminds you why anthropic’s whole vibe feels anomalously clean in a world addicted to sludge metrics. kudos to mike & team. also you can see a reflection of this in their marketing as well (which i posted about here before).

Cursor scaled to $29B without any full-time PMs. Ryo (Cursor's Head of Design) walked me through how they work and it's the opposite of every big tech best practice: 1. Roles are muddy PM work is spread across designers and engineers. Everyone does what fits their strengths

wrecked

Applied AI products supporting new model releases isn't what is used to be. It's no longer a question of evals, it's a question of agility and, in some cases, grit. Opus 4.5 exposed an entirely new concept called "effort", a parameter that involved optimization and use case

ever find a UX pattern so good, you've started a company because of it?