Ilia Breitburg

Crazy that a screen can be more powerful than a laptop.

AI models are actively avoiding Taylor Swift: evals.breitburg.com/swiftie-bench/

@karpathy I love that the last improvement was changing the random seed, this poor agent was running out of options

I packaged up the "autoresearch" project into a new self-contained minimal repo if people would like to play over the weekend. It's basically nanochat LLM training core stripped down to a single-GPU, one file version of ~630 lines of code, then: - the human iterates on the prompt (.md) - the AI agent iterates on the training code (.py) The goal is to engineer your agents to make the fastest research progress indefinitely and without any of your own involvement. In the image, every dot is a complete LLM training run that lasts exactly 5 minutes. The agent works in an autonomous loop on a git feature branch and accumulates git commits to the training script as it finds better settings (of lower validation loss by the end) of the neural network architecture, the optimizer, all the hyperparameters, etc. You can imagine comparing the research progress of different prompts, different agents, etc. github.com/karpathy/autor… Part code, part sci-fi, and a pinch of psychosis :)

Outside Anthropic's office in SF... intense moment!

A statement from Anthropic CEO, Dario Amodei, on our discussions with the Department of War. anthropic.com/news/statement…

I'm increasingly taking pretty strong versions of this view seriously.

The increasingly-hyperbolic METR graph is actually good news for safety. We just have to survive a brief singularity in March, and then afterwards the models will never be able to do more than undo a few hours' worth of work

@ericmigi Try asking Claude to launch subagents to make screenshots for each platform in parallel. It’s beautiful. Also, since it understands what the features of the app are, the screenshots make sense

adding a few more features to pebble-tool and cloudpebble - like auto generating screenshots/gifs on all platforms, then sending pbw and screenshots to Pebble appstore as a new release. Any other dev quality of life features you'd like to see?

they received $500 million in funding from this

@olegkozyrev зацени huggingface.co/breitburg/gemm…

Apple & Anthropic always had/have similar energy Big L on Anthropic's part for fumbling the siri negotiation. Big L on Apple for not just buying Anthropic.

AppleScript has been waiting 33 years for this moment

i was playing with claude code for the last 2 months and actively using opus 4.5 for the past week instead of holidays, i ended up having an identity crisis, with no understanding of where my personal leverage is anymore as an engineer who started coding at 10 using a physical book “python for dummies” and felt ahead of everyone for the past 14 years, i now feel like there’s no difference whatsoever between me and non-technical people that’s quite depressing tbh

I wonder what happens if you scale it up

so this costs $500/mo. what happens if you miss a payment?

@ericmigi @Pebble Seemed like a waste not to

Of course someone is training a neural nets on their new @Pebble 2 Duo...

@ericmigi @Pebble I’m using the Pebble 2 Duo on 4.9.71-stop0 with an app v1.0.4 on iOS. If the animation is continuously playing, either the API request is taking too long or AppMessages aren’t working :P

@ericmigi @Pebble I just released an update that includes a request timeout and improved error handling. I hope this update works now! I appreciate you trying it out, and thank you for creating such amazing things. You’re truly an inspiration.

@ericmigi Claude says he’s feeling great running on Pebble :) apps.rebble.io/en_US/applicat…