Mike van Rossum

Went to my second ever broadway show the other week and it was a one-man performance by Daniel Radcliffe aka Harry Potter. He spends the 20 minutes before every show handing out props and recruiting people to interact with him during the show. Never seen someone so dialled in. Was an amazing performance, 1.5 hours straight, no intermission. He does this night after night. Did some math with Claude and he probably makes $50K per week which is great $$ but still considerably less than what he would make for a movie role. His parents also helped him invest his Harry Potter earnings well and he makes $660K/month just off investments (UK companies are required to make full financial disclosures, very different from the US but very useful for this kind of pocket watching). Which means he never has to work again he just continues to do this for the love of the game. These kinds of in-person experiences and acts of passion are one of the few things that can’t be replaced by AI Oh also he does not like it when you call him Harry or bring up his old movies. Ask me how I know.

Jensen Huang: "If that $500,000 engineer did not consume at least $250,000 worth of tokens, I am going to be deeply alarmed. This is no different than a chip designer who says 'I'm just going to use paper and pencil. I don't think I'm going to need any CAD tools.'"

i've been trying to automate my crypto research for months.. burned through probably $1500 in API costs learning what doesn't work then qwen drops this today and i'm like... fuck what i was doing: paying claude API to monitor markets 24/7 checking whale wallets, price alerts, volume spikes, key events and unlocks, the whole thing yes it did worked great but the problem is i was paying per query for stuff that should just... run checking "did this wallet move?" 24 times a day adds up fast but i need that monitoring. can't manually check wallets every hour. what i tried: switched to gemini thinking it'd be cheaper. got rate limited constantly. tried chatgpt API. worked until i hit their usage limits mid-research. tried mixing cheaper models for simple tasks. configuration nightmare. the realization: cloud APIs are designed to monetize every call for simple monitoring, i'm paying premium prices for basic checks then qwen dropped these models today: 0.8B, 2B, 4B, 9B free. run locally. no API key. no per-call charges. i'm reading the specs and it's clicking: "4B model for lightweight agents" "9B closing gap with larger models" wait... i can run these on my laptop? for free? unlimited checks? what this means for my setup: whale monitoring: qwen 4B running locally, checking addresses constantly → $0/month price alerts: 2B model fast enough for real-time checks → $0/month basic research: 9B for pulling on-chain data, comparing metrics → $0/month im testing it now.. downloaded the 4B model. running it locally. set it to watch 5 wallets as a test. response time is solid. checks every minute. no API cost. if this scales to 20 wallets with no slowdown, i just saved $200+/month on monitoring alone. the plan is migrate simple monitoring to local qwen models (2B/4B) keep claude for complex strategy analysis and final content polish basically, free local models for 80% of grunt work, paid API for 20% that needs premium intelligence i'm sharing because i spent months burning money on cloud APIs for tasks that don't need premium models qwen dropped the exact solution today if you're automating anything repetitive (monitoring, alerts, basic research), test these local models first might save you hundreds per month like it's about to save me downloading the models now. will report back on how it actually performs at scale.

2016: this meeting could have been an email 2026: this meeting could have been a markdown