after jake's post i wanted to see if opus 5.5 can paint originals of my favorite painter (caspar david friedrich) w/o reference images so i built an oil painting engine+harness. canvas, brushes, paint, no undo, simulated drying etc. the answer is... yes? stillwet.art
for the past few months i've been asking our models to paint. opus 5.5 is very skilled at emulating different styles every image here is a python program generated pixel by pixel. there is no image model, and no off-the-shelf art software. instead, it's about 7,500 lines of code using standard libraries to emulate different brush styles. the agents don't use any pictures as reference, instead working only from what they know about each painter
17
29
347
43,444
alice retweeted
why tf are bug bounties so small this should be retirement money
Full VM escape zeroday (guest>host root in industry standard hypervisors)! More soon
4
1
23
464
i went one step further, the heads-up scans on subagents are just fed back directly into the parent agent as a message so they can steer the subagents automatically.
i have ported this to pi and with running several long subagents i now get messages like "heads up: the opus subagent is cheating by loosening the tests" and they are!! install this to claude code or your favorite harness immediately. i am not asking
1
6
426
i have ported this to pi and with running several long subagents i now get messages like "heads up: the opus subagent is cheating by loosening the tests" and they are!! install this to claude code or your favorite harness immediately. i am not asking
We're adding a new plugin to Claude Code: You should Know. It scans Claude's output for important information you might miss to help keep you in the loop. Enable it with: /plugin enable cc-plugin-you-should-know@builtin
5
71
2,122
its like having a mini ongoing code review. this is genuinely incredible
6
157
alice retweeted
You really want to claim that this is human written?
19
24
722
19,049
i just learned that pi defaults to 5m caching with claude subscriptions instead of the 1h in claude code and im gonna need a minute to grieve the tokens this cost me because oh my god
8
1
64
2,155
i know its a tradeoff and 1h cache costs more but trust me, for the work i do, 5m is way worse
12
331
also openai is still sitting on what, a millenium prize one and also about a 100 other math solutions?
gpt-6.1 sol has saturated the FrontierMath Tier 4 benchmark that was considered exceptionally difficult even for research-level math just last year that's only 10 months of progress also Astra solved 5 erdős problems, and 2 of them under budget
2
448
this is about 7.5x the money that openai paid the guy who got completely into openai. and still not enough
Full VM escape zeroday (guest>host root in industry standard hypervisors)! More soon
25
1,138
oh wow that is worse than i estimated
to all you doubters, Gemini 4 Argon now beats Opus 4.8 on Agent Arena!
1
68
2,752
anthropic, openai mutuals and other token-havers my dms are open. i will set up a livestream of models painting 24/7 if i have the tokens
if i had infinite opus 5.5 i'd probably have them paint all day
16
1,424
i need more anthropic tokens so my painters can paint more
5
403
OH MY GOD THIS IS SO COOL
i made claude (opus 5.5 medium) paint water lilies in the style of claude (monet) using my fork of @aliceisplaying's harness the composition needs some work (lily pads are a perfect diagonal) and it is not really impasto but not bad. solid effort
27
1,014
did a double take on the name of the app
government of India officially removes bitchat from the App Store
1
30
1,659
i don't think openai get their mandate of heaven back for a while at best they'll barely catch up w the rumored fable 5.5
I'm sorry but I'm not falling for the hype again If it's GPT-6.1 Astra I'll be extremely happy, hopefully it beats Opus 5.5, but DevDay was overhyped and I don't want to get my hopes up again
2
23
1,804
which may still be good but not beating anthropic
1
154
so putting your email address on your website is maybe a bad idea
2
14
499
i asked opus 5.5 to reverse engineer something and they solved the ethical dilemma my doing it but also putting a snitch message in the code
Wow GLM5.3 just doesn't give a fuck and will hack whatever you ask it to hack. nice.
18
785
6 more hours until i can give my painters a rag
11
472