Prototyping a VSCode extension #lab

fix multi real time streaming callbacks #lab

Read The Illusion of Thinking: Understanding the Strengths and Limitations of Reasoning Models via the Lens of Problem Complexity. Conclusion: high complexity is hard for any current LLM model. "We identified three distinct reasoning regimes: standard LLMs outperform LRMs at low complexity, LRMs excel at moderate complexity, and both collapse at high complexity."

I have been working on silence. The results are quite humble, but this is just the beginning.

Just read this paper "Darwin Gödel Machine: Open-Ended Evolution of Self-Improving Agents"

spent another day fixing internal tool #lab

bought some credits on main LLM apis to experiment and compare

Back after spending a whole month automating a 3 day task. Worth it? I really hope so, time will tell, basically built a semi automated deep research

Read this paper: Can AI Help with Your Personal Finances? TLDR if you ask it enough times, you might get wisdom. But "enough" isn't defined, and the cost of poking? Unclear. The experiment use 10 by default. Best model on the research was original Sonnet 3.5.

Fix db clone script made by Gemini 2.5

checking benchs of llama 4. planning integration onto my saas

"coded" from mobile

Sharing results of "Project .io vs .ai - the ultimate SEO duel 🔍" .ai site pulling more views w/ just tech SEO, zero backlinks. .io mvp release delayed, 3 behind schedule but cooking something cool

got gaslit so bad by openai deepsearch

Check new Open AI paper

went outside. sunned. grass touched. humans spoken to. disturbingly pleasant, didn’t know how much I missed it

woke up at 3am. silence sharpens the mind. no noise, just code. only thing left is to ship.

Spend all day vibe coding a chunk of a SaaS. Failed hard. Burned 50 credits. Rebuilding from scratch

imagen gen down again

wake up at 2:30 am to work. That code's not gonna write itself... or is it?