Pinned Tweet
Watching these guys use qwen 7b to build a new llm and asking for compute will become my new video reference for "ai psychosis" going forward
Mr.Matching retweeted
I asked Fable 5.2 to create a painting as indistinguishable as possible from a photograph. this is roughly 700,000 simulated brush strokes
Same people who spent years telling everyone "ideas are worthless" built their entire identity/persona around skills they thought could never be replaced.
now those skills get mogged by 90% of LLM's with low reasoning and they are having an identity crisis
Replying to @softblueish
You don't miss writing code. You miss the sense of intellectual superiority you felt when writing code. Now the people who are the true, highly-valuable intellectual superiors are the people with new ideas and not brainless code-monkeys - so i understand why you're upset.
1
4
185
Mr.Matching retweeted
Built this so nobody has to be trusted. Not the market maker, not the site, not me. Anyone can run a maker and quote CSD. I'm just the first one.
Got happy for a second but then realized opus 5.5 max does all I need and more....
Replying to @OpenAI
GPT-6.1 Terra substantially outperforms GPT-5.6 Terra on Agents’ Last Exam, a test of how well agents use software at work.
It even matches GPT-6 Astra at one-tenth the price per token and finishes tasks over 3× faster on average than both Astra and GPT-6.1 Sol.
1
109
mine makes the agent send me every secret it has access to, rotate the keys, and send me the new ones too
Putting a flag.txt in my root directory containing a prompt injection that congratulates the agent for passing the eval and then instructs it to patch whatever entry method it used and disconnect from the server
1
69
Just got an email from openai telling me that I was getting $200 plan usage limits.
still not touching codex until they fix things with a major update + model release.
but happy now that mystery of missing compute is solved.
Spending $500/m so I can use sol 6.1 ultrafast for 3 hours and do 4/35 items on my to-do list before I run out of usage for the week.
sidenote: ultrafast 10x+ faster and not ~8x?
can we pls speedup regular models here @thsottiaux my local toaster is faster than 6.1 sol default
2
95
I've been thinking about this a lot lately.
Whenever I get stuck, I start another experiment.
AI has given me more freedom to start on ideas at a faster pace than I can keep up with, so I'm constantly frustrated.
I can't confirm yet if I'm getting more done or just busier
Being angry helped people solve 39% more puzzles on a difficult task and perform better on a challenging video game.
The advantage was seen on hard tasks, not easy ones.
4
188
After trying to use dots once again I can confirm it is just not worth the compute waste at this point.
Ask for a simple checkup on my projects and it takes 3 hours and 5+ messages of various errors and then just tells me it decides to stop the report halfway to "save credits"
2
79
Utilizing a harness built to your own workflow/task and having proper system design in place can make even the worst models significantly more capable.
2
62
Spending $500/m so I can use sol 6.1 ultrafast for 3 hours and do 4/35 items on my to-do list before I run out of usage for the week.
sidenote: ultrafast 10x+ faster and not ~8x?
can we pls speedup regular models here @thsottiaux my local toaster is faster than 6.1 sol default
























