Love to see open weight models here.

I like and value dedicated QA (and believe manual testing still has its place too) BUT if you do it wrong/too early without building a quality culture first, then you get abdication of responsibility from engineering. "I tested the happy path and it works, so here you go QA!".

When interviewing a candidate, the goal is to get to a strong yes or no. If the feedback is "they might be good; hard to tell", the interview failed. I fail ~25% of the time -- it's hard! But it's important to understand the goal, and to train your team for it.

My Lumina toothpaste finally arrived (frankly I kind of forgot I ordered it). But now I'm too scared to try it. And I do get cavities.

RT @ArtificialAnlys: Developers consider an average of 4.7 LLM families with OpenAI GPT/o, Google Gemini and Anthropic Claude being the mos….

RT @ArtificialAnlys: We’re releasing the Artificial Analysis AI Adoption Survey Report for H1 2025 based on >1,000 responses from developer….

RT @pingToven: Coming soon to OpenRouter 🔥.

Good judgement, not code, is the new bottleneck. And it's always been the thing in shortest supply.

The fact that diffusion LLMs exist & work is a pretty strong argument that LLMs are fundamentally different than human intelligence. I can sort of convince myself I might just be a next token predictor — but no f’ing way I’m a diffusion model!.

Weird that 'stringy' and 'stingy' are pronounced completely differently.

RT @PyythonEX: @lauriewired GPT posts: You're not ready for this next pro function .Gemini: Ship ship ship .Claude: The model just said it….

I liked ChatGPT’s advanced voice more when it was precise like JARVIS and not “yeah, well so, it is true that uhh…”.

At face value, there is a ~500x cost between the cheapest and the most expensive. But output tokens 3x more expensive than input, and high end models’ reasoning outputs, there’s another order of magnitude — the actual cost difference is close to 5000x!.

RT @deedydas: Karpathy at YC startup school calls this the transfer switch of AI. It's hard to keep up with all the LLMs out there. OpenRo….

Why is that all designers are expected to have a portfolio, but not engineers? They both make things.

Not sure what I've been doing to that caused me to receive a targeted @Sothebys ad for this on Instagram, but I'll try to keep it up!.

RT @OpenRouterAI: Ranking of LLMs by the % of JSON violations we detected over the past week, for the top Structured Output requests. ⭐️ Qw….

This type of error is what drove the overpopulation panic of the 90s.

A sigmoid looks a lot like exponential growth, until suddenly it doesn’t. I think it’s likely that AI model progress is halfway up a sigmoid, vs exponentially gaining capability. This doesn’t make AI any less exciting, but does make me less worried about superintelligence.

The constant reinvention of front end frameworks has finally stopped because frameworks are now selected for their LLM legibility. New frameworks are definitionally outside of the training set and will therefore never catch on.