π‘˜π‘’π‘Ÿπ‘›π‘’π‘™π‘‘π‘Ÿπ‘–π‘π‘˜ (@kernel_trick) header
π‘˜π‘’π‘Ÿπ‘›π‘’π‘™π‘‘π‘Ÿπ‘–π‘π‘˜ (@kernel_trick) profile picture

still somewhat loves computers β€’ some exp with llms β€’ there are no claude-like models β€’ tokens per petrodollar

Joined June 2025
Photo by π‘˜π‘’π‘Ÿπ‘›π‘’π‘™π‘‘π‘Ÿπ‘–π‘π‘˜ (@kernel_trick)Photo by π‘˜π‘’π‘Ÿπ‘›π‘’π‘™π‘‘π‘Ÿπ‘–π‘π‘˜ (@kernel_trick)Photo by π‘˜π‘’π‘Ÿπ‘›π‘’π‘™π‘‘π‘Ÿπ‘–π‘π‘˜ (@kernel_trick)Photo by π‘˜π‘’π‘Ÿπ‘›π‘’π‘™π‘‘π‘Ÿπ‘–π‘π‘˜ (@kernel_trick)Photo by π‘˜π‘’π‘Ÿπ‘›π‘’π‘™π‘‘π‘Ÿπ‘–π‘π‘˜ (@kernel_trick)Photo by π‘˜π‘’π‘Ÿπ‘›π‘’π‘™π‘‘π‘Ÿπ‘–π‘π‘˜ (@kernel_trick)Photo by π‘˜π‘’π‘Ÿπ‘›π‘’π‘™π‘‘π‘Ÿπ‘–π‘π‘˜ (@kernel_trick)Photo by π‘˜π‘’π‘Ÿπ‘›π‘’π‘™π‘‘π‘Ÿπ‘–π‘π‘˜ (@kernel_trick)Photo by π‘˜π‘’π‘Ÿπ‘›π‘’π‘™π‘‘π‘Ÿπ‘–π‘π‘˜ (@kernel_trick)Photo by π‘˜π‘’π‘Ÿπ‘›π‘’π‘™π‘‘π‘Ÿπ‘–π‘π‘˜ (@kernel_trick)
π‘˜π‘’π‘Ÿπ‘›π‘’π‘™π‘‘π‘Ÿπ‘–π‘π‘˜ retweeted
So creative. I love it!
2
1
15
1,080
looked at python ecosystem and sighed
Bloom is a library for layout, motion, and effects for Bubble Tea.
4
106
5,868
what in the parasocial
εœŸθ±†ε§ (@poteto) Lauren Tan 2015β€”2026 εŽ†εΉ΄ι’œε€Όι€ εž‹ε˜θΏπŸ‘‡
19
5
724
82,717
sorry your cute UI is not a moat either
You can now build custom themes for Claude Code on desktop with mods! Change the colors, add an animated background, and make it your own. Just tell Claude how you want it to look, or get started quickly with the theme-examples plugin. We'd love to see what you make!
1
6
268
also how much cope is "coding and math improve general reasoning"
β€œSomething important will continue to be missing” What Ilya refers to here is the lack of generalisation in current LLMs. Want them to be really good at computer use? Cool, you gotta collect a bunch of RL environments for that. Want them to be good at creating slides? You gotta collect a huge amount of RL environments for that. The thing is, current LLMs don’t generalize beyond what they are trained on. Each new thing we want them to do requires collecting a huge amount of data (for which labs like OpenAI and Anthropic pay lots of money - RL environments are created by contractors paid to create high-quality data). Ilya wants to create AI that generalises dramatically better, comparable to humans
1
5
290
try adding "user is a habitual liar" to your agent memory and let me know the results
2
12
372
so CodeAct
🚨Claude code users!! Your 5-hour and weekly limits will stretch much further. Anthropic published a new official documentation page for experimental Code Mode in Claude Code v2.1.323+. Instead of sequential tool calls, Claude writes JavaScript to run parallel tool calls in one step. code.claude.com/docs/en/code…
1
188