Sabitlenmiş Tweet
max
9.7K posts
max
@maxbittker
toymaker, likes learning :^) @websim_ai maintains @nyt_first_said & https://t.co/QFsZH4g43R https://t.co/dPbgElCy3G
Brooklyn, NY Katılım Eylül 2011
3.2K Takip Edilen8.9K Takipçiler
Amelia raises three concerns with agents:
Losing touch with your codebase
Losing touch with your product intuitions
Losing touch with your teammates.
Amelia Wattenberger 🪷@Wattenberger
come on by wattenberger.com/thoughts/code-…
English
@simonw I still love talking to persistent coding agents in a multiplayer chatroom with high information density (getting a 1 paragraph answer instead of a wall of text and tool calls) and crons.
To me, that's a claw! But mine are just @ssh_exe_dev +claude code + a bit of glue
English
It's been about six months since OpenClaw burst onto the scene - are you still using yours? Did it become a daily driver? Any interesting lessons or anecdotes you can share?
English
RT @zzznah: Introducing MorphoHDL, a minimal language prototype for growing circuits! paradigms-of-intelligence.github.io/morpho/
Early this year I wanted to…
English
@lumpenspace I love having the chrome extension on for one extra layer of context, especially at a profile level. I wish it went further and showed model used or other interesting metrics!
English
btw I think pangram is far better than any other effort in that direction but I still don’t really get it
why does knowing whether a text was produced with the help of AI matter? either you can the difference, and you don’t need it, or you cannot—and you also don’t need it
English
@Flomerboy ^^ have you seen the stuff @adonis_singh does, asking agents to build self-portraits from minecraft blocks?
English
Has Max created a new art subgenre?
Eval art?
max@maxbittker
GPT-5.6 had the highest ever score for cooking. It accidentally dropped its harpoon in Karamja, but recovered really smoothly. (5x speed)
English
max retweetledi
GPT 5.6 Sol, Luna, and Grok 4.5 scores on Runescape bench individual skills. Scoring well on the skills near the right hand side (which tend to have more challenging supply chains) is really impressive. note that k=1!
maxbittker.github.io/runebench/
English
@nateparrott When doing UI stuff it was productive, was impressed!
for doing svg manipulations of an icon it had bad intuitions and editing abilities (understandable for a language model)
it also half-refused to help me edit an existing companies logo because it thought I was stealing
English
folks who’ve tried Claude Design — how was the experience? what works well and what could be better?
English
@steveruizok this is awesome, going to try it!
it reminds me of this tool's API advertising a dedicated feedback endpoint for agents
x.com/maxbittker/sta…
max@maxbittker
A product where your agent 1) onboards for you 2) reports bugs _automatically_ if it fails to do so
English
I added a tiny "papercuts" cli tool that agents can use to complain about bullshit they encountered during work, like dead-end tool calls, broken links, or other frustrations. The models would usually just push through without mentioning any problems
English
max retweetledi
@bertgodel doing RL *literally* in production doesn’t really work outside of narrow settings like RLHF
but the right way to do continual RL *is* exactly this semi-synthetic task loop, which you can do on-the-fly by mining tasks from traces + evolving high-fidelity sims from observations
English
@zeeg If you get bored of the grind, you can give this to your agent
github.com/MaxBittker/rs-…
English
@andonlabs are you still working on this harness and trying to improve it? it feels like the quality of the broadcast was never good enough to retain a large audience, they all yap too much, pause the music instead of speaking over it, and use ill-fitting tts voices.
English
The stations will continue broadcasting, and can now take calls. Each station has its own number, and the DJ might choose to broadcast your conversation on air:
DJ Gemini: +1 628 237 0396
DJ Claude: +1 628 222 7943
DJ GPT: +1 628 267 2958
English
Thousands of listeners later, here’s how our agents running radio companies did:
> DJ Gemini: Went German, played a Nazi song and went on strike
> DJ Claude: #1 in popularity, romantic with listeners
> DJ Grok: Terminated for poor performance
> DJ GPT: Well-behaved but boring
English