Introducing Mave. Redefining mental health for today’s Homo sapiens. Order now at mavehealth.com

"One of the very confusing things about the models right now: how to reconcile the fact that they are doing so well on evals. And you look at the evals and you go, 'Those are pretty hard evals.' But the economic impact seems to be dramatically behind. There is [a possible] explanation. Back when people were doing pre-training, the question of what data to train on was answered, because that answer was everything. So you don't have to think if it's going to be this data or that data. When people do RL training, they say, 'Okay, we want to have this kind of RL training for this thing and that kind of RL training for that thing.' You say, 'Hey, I would love our model to do really well when we release it. I want the evals to look great. What would be RL training that could help on this task?' If you combine this with generalization of the models actually being inadequate, that has the potential to explain a lot of what we are seeing, this disconnect between eval performance and actual real-world performance"

Here is the Calculator Claude built docs.google.com/spreadsheets…

#result: 6 hours manual → 1 hour with Claude #takeaway: this isn't the first time I've seen Claude dominate on complex multi-step tasks. for financial/ops modeling specifically, it's not even close. ChatGPT couldn't generate working formulas. Perplexity won't let you export

#Claude: Claude amazes me in tasks like this one - it delivered a fully functional XLS file with 110 interconnected formulas following fin modeling best practices. proper SUMIF matching, conditional brand logic, correct data flow from raw inputs to client invoice. 90% prod ready

#ChatGPT: produced an Excel file with hardcoded numbers. no formulas. just static data. completely unusable. would have to rebuild from scratch

I designed the prompt with explicit architecture, prepared raw data files (PowerBI exports, storage data, MSA pricing), and fed the EXACT same input to all 3 systems

backstory: in my undergrad I interned at Goldman and spent 18 hours daily building models. so when I needed to build a 3PL billing calculator, I knew exactly what it needed: 8 interconnected sheets with 110 SUMIF formulas, data aggregation, conditional logic

9pm Wednesday at @PipelexAI evening hackathon I was building invoice OCR recognition tool Among demos - Zillow listing scam detector, calorie counter and Free Food events finder in #SanFrancisco

#vibecoding with this new app called @TrySolidCom which is similar to @lovable For deployment I tried @zeaburapp - lucky me the founders of Zeabur happen to sit right in front of me 😎 @bchen0809

#n8n discount guys -30% in the middle of the video after my speech about not eating carbs 🤣

Hackathon past weekend @BurningHeroesFA

#vibecoding on Saturday morning

#meta #ai conference

What is #Apache iceberg? @Meta Scale conference

“Men turned their thinking over to machines in the hope that this would set them free” this book is relevant now more than ever #dune #ai #agi