Lukas Biewald

vibe-cadding is simultaneously so amazing and so frustrating

I love letting my six year old daughter create crazy cursor prompts and seeing what it spits out. If you find this summary intriguing, you can try the game at otto-game-browser.netlify.app - it did an impressive job with a tough spec

What happens when you post a real Monet and say it’s AI? The coolest art social experiment I’ve seen in a while. Thank you @SHL0MS

@yanatweets This is really cool! I’ve been doing a lot of CAD experiments in other domains. My experience sounds similar that the LLMs are terrible at generating it directly but can generate python code to generate good CAD files

I’ve been experimenting with having Codex generate dress patterns in CLO (fashion software) and render them on a 3D model. It found a Python API… and just skipped the UI entirely. Here’s the result after 10 hours of fully autonomous work: Still early, but kind of wild.

@ClementDelangue Always the best advice 😀

@l2k Try it!

The new Claude and OpenAI models are vastly better at physical reasoning and generating CAD files. I had them design me a mathematical chandelier and a life size hexapod walker. The assembly instructions, seem plausible, I wonder if it will work.

Both me and my wife uploaded the same photo of a cut on my son's head and asked GPT whether or not to take him to the hospital. GPT told me it was fine to watch it and wait and told my wife he definitely needed to see a doctor immediately - in both cases confirming our prior bias. It's possible that this was just a borderline case and we randomly got different results, or that this was caused by a slight variation in our prompts. But it does make more suspicious that some artifact of the post training is causing these models to tell users what they want to hear. As the models know more and more about us, and there's more pressure to grab marketshare, this is only going to get worse. Sadly it's way more fun to have your biases confirmed.

@NicholasBardy It doesn’t control the benchmark

@l2k A lot of these look like straightforward improvements. I'm surprised it didn't just game the benchmark here and win here by just increasing concurrent users, at the cost of tanking per user throughput. Looks like there is one batch size incrrease.

I used Karpathy's autoresearch pattern to optimize inference throughput overnight. The results look promising.

BREAKING: Meek Mill has abandoned the W&B web dashboard entirely. Says he only monitors training runs in the terminal now. "I don't need a browser to know loss is converging," sources close to the rapper confirm.

AI coding tools are great at helping new developers ramp faster. Mike Cannon-Brookes talks about why that’s a good thing, but also why it doesn’t remove the need for experienced engineers. Someone still has to look at the code, understand it, and stand behind it. It’s a useful way to think about how teams can scale with these tools.

All the people saying Cursor is dead clearly have not used the product recently - it’s gotten incredibly good

We heard you. The wandb mobile app is now LIVE on iOS 🚀 Monitor training runs from anywhere. Crash alerts the second something breaks. Live metrics on your phone. This has been the most requested feature in wandb history and it's finally here!

Introducing W&B Skills for coding agents! Watch us install the skill, point it at a live fine-tuning project with thousands of traces and RL training runs, then query it all from the terminal in seconds. Also works with @weave_wb to pull in your agent traces and spot failures.

Full episode: youtu.be/-hl0jpwWbV4?si…

I thought drug design was one big ML problem. Isomorphic Labs' Sergei and Max explain that it's more like 10 massive, distinct problems stacked on top of each other. You can't just solve for structure; you need safety, delivery, and a dozen other variables.

@yanda @wandb You left out the part where two months in you started smoking cigars, submitting unhinged expense reports and yelling things like “the leads are weak” at your desk

In 2021 I promised @l2k I'd work on whatever kept him up at night at @wandb. A few months later he said that thing was "sales." I'd never carried a quota in my life. Here's what happened next 🧵

I'm claiming my AI agent "coldclawlukas" on @moltbook 🦞 Verification: coast-5GXV