My best results so far: two separate conversations 1/ Plan and code 2/ Review code todo after todo. Having separate conversations (ie different contexts) avoid biased reviews and give much better results. eric zakariasson any way to enable this (today I manually trigger reviews)

On ne peut que soutenir cette initiative chez Bearstech bearblog.dev

AGI won’t be close until an AI can simply say “I don’t know” instead of giving you a confident but completely wrong answer.

I was bored so I created the most essential card deck in the universe ilove.cards/explore/decks/…

omg, Cursor + composer merging and solving complex git conflicts in less than a minute (most of it being me checking the commands)... This is huge.

The delicate thing when using AI to correct a tricky bug is giving it enough guidance or context w/o inducing some biases. It will happily solve a wrong hypothesis instead of finding the hidden root cause.

Introducing the Devstral 2 coding model family. Two sizes, both open source. Also, meet Mistral Vibe, a native CLI, enabling end-to-end automation. 🧵

I'm crazy 🤣. I (try to) read every single line of Opus reasoning output in real time. It gives me a much better ability to detect if the code is gonna be good or shit. I can even stop it mid-reasoning to guide it in another direction.

First experiments with shaders by Simon Le Marchant. Super fun. Will probably be in Stormz v3. Along with Nuxt and NuxtUI.

How is it that the "I want to retrieve my invoices" is so broken across SAAS providers? With the millions you raise you cannot send an email with the invoice or provide a one-click download of a Pdf? Or let access even if I am back to a free plan?