LOL
Posts by Ilia Breitburg
Going to help with making the distribution more even with my upcoming Pebble contest submission
what if you could navigate unicode characters by visual similarity?
AI coworker UI idea: badges!
@ericmigi.com any timelines on exposing the speaker APIs for Pebble 2 Duo?
claude may occasionally transfer funds...
Hello Graphite
nice digicam you got there
The default hostname is "...'s MacBook"
It appears that the new MacBook Neo was supposed to be named just MacBook
How about a new RSS client?
AI models are actively avoiding Taylor Swift: evals.breitburg.com/swiftie-bench/
Opus 4.6 stands at 22%, GPT-5.3 Codex at 14%, and GPT-5.4 is at 0%. Can't wait to see Anthropic’s efforts to counter this as well!
Introducing the "Code Comments Slop" bench, that measures the rate at which LLMs put sloppy sections in code comments like these:
# ============================================
# CONFIG
# ============================================
evals.breitburg.com/code-comment...
Any interesting observations? I’ve been a Claude fanboy for more than a year now, is trying Codex worth the time?
Why 5.2 instead of 5.3?
I don’t get why Anthropic hasn’t used Claude Code to rewrite their bloated Electron app as native. Should be much easier than a C compiler.
When you write a reward function for RL, you essentially write your wish to a genie in Python
Taking time to read pint articles (sometimes pint for a while). I appreciated this one about interactions design by Rauno Freiberg.
rauno.me/craft/intera...
Need to frame that
Yooo! Welcome. First time here?
I've been working on a custom IMAP/SMTP server that acts as a Telegram proxy for any email client. Your Telegram messages arrive as emails, and you reply to send messages back. Really fun stuff. Hope to finish and open-source it soon
Would you prefer phone calls and emails instead of instant messaging for communication with friends? I think there's certainly something about mail that was lost with IM. The friction of being unable to edit or delete what you've written makes you more present and think more when composing
Game engines are O(n) in scene complexity. Diffusion models are O(1), so the same cost whether you’re rendering an empty room or a million polygons. What if you made the engine differentiable and optimized the diffusion model against it directly, rather than sampled frames? Has anyone tried it?
I recommend reading The Math of Why You Can't Focus at Work.
Excellent visualization about how we lose valuable focus time because of interruptions and what we can do about it.
justoffbyone.com/posts/math-o...
The perfect Claude Code machine acquired
Oh hey I’m on TV