@Polymarket trader @LATINAXBT just pulled off an impossible comeback. Nov 2025: $3.05M May 2026: -$1.51M Jun 2026: $691K Survived a $4.6M drawdown. Went seven figures underwater. Then clawed all the way back to green in 18 days. $61M traded, 56% win rate, ice in the veins.

For 18 months, this Polymarket account's lifetime PnL was minus $25. The next 11 months: $407,492. It's 1 of 18 wallets that went from ~$0 to $100k over the same year. I profiled all of them, and 4 playbooks kept showing up.

That's right, to explain interleaved thinking more straightforward, I made a comparison html with M2! It visualizes the difference between regular thinking, interleaved thinking but reasoning not passed pack, and full interleaved thinking. We will post a tech blog with specific benchmark with/without reasoning passed back

Context is the new RAM & LLM is the CPU. Agent development is nothing but applying the fundamentals of CS.

Context Rot Great title for a report, but even better insights about how increasing input tokens impact the performance of top LLMs. Banger report from Chroma. Here are my takeaways (relevant for AI devs):

If Claude is really doing so much of the coding for Anthropic, why haven't they used it to create a fucking ui for Claude Code? It's 2025. Why the fuck am I forced to use a cli for everything as if it were 1995?

Don't use structured output mode for reasoning tasks. We’re open sourcing Osmosis-Structure-0.6B: an extremely small model that can turn any unstructured data into any format (e.g. JSON schema). Use it with any model - download and blog below!

A lesson that took me 10 years to learn..

Some of you, vibe-coders, are the exact definition of Mount Stupid. I was also young and cocky and thought I knew better than everyone else. A bitter lesson I learned the hard way: good software is way more than what you see on the screen. A million micro-decisions hold it all together behind the scenes. You might not see them, but they are there and matter. It's awesome that people are vibe-coding and learning how to build software. It's not the conventional way most of us went through, but it's also valid, and I hope it sticks. But you gotta understand that it's unlikely you'll build good software after a day of talking to a model, especially with the current state of this technology. It will take time and much more work than these models can give you today. So stay humble, have fun, and use this as an opportunity to learn and improve. You'll find a ton of people willing to help! But if you get cocky, start disparaging "boomer" developers, make fun of those pointing out the problems you are creating, and in general, have a know-it-all attitude, don't be shocked when people exploit every single vulnerability you created. Tomorrow, we might all become obsolete, but that's not today.

Even better than meditation.

Don’t work at Anduril. Or SpaceX. Or Tesla. Build your own company. There’s never been a better time for engineers to start startups. We’ll teach you how.

Every good founder must not get bullshitted

Agent Laboratory Using LLM Agents as Research Assistants

.@ilyasut full talk at neurips 2024 "pre-training as we know it will end" and what comes next is superintelligence: agentic, reasons, understands and is self aware

⛽️💸 Ethereum’s Gas Race: Uniswap Guzzles & a Subway-Themed MEV Bot Cashes In? Find out how institutions & bots are adding to the woes of retail crypto transactions datadash.substack.com/p/top-…