I just finished reading this article and it quietly reorganised how i think about training AI at all.. the core of it is that nobody actually gets good from the finished work instead you get good from the markup, the part where someone senior shows you what was wrong and why. the edit is the lesson, not the clean final draft... and that's exactly how AI got good at code, too. it never learned from finished code, it learned from the diffs. every fix, every rewrite, every "no, do it this way" that git happened to save over the years. but everywhere else, we throw that part away. the deck ships, the memo gets filed, the contract gets signed, and the thinking that produced it is gone. a document is the output of judgment with the judgment removed. > it's why "capture your knowledge" by searching old documents never worked. every firm built a system of record.. what it knows. nobody built the system of judgment.. what it actually decides is good. and judgment is revealed, not stated.. interviews lie, edits don't. > this is the part the "untrainable / private correctness" debate keeps missing: it's not untrainable, it just lives in the one place everyone deletes. ask any firm for its acceptance rate and it can't even tell you.. nobody kept the redlines. software just happened to keep its receipts. law and finance never built their git. the judgment is in the diff.

Anthropic engineers just built a full app from scratch with nothing but a loop of agents 40 minutes from the team that builds Claude Code three agents running in a cycle: one plans, one builds, one judges, looping until the app actually works nobody in that demo wrote the same prompt twice. they designed the loop once and walked away watch it, then read the full guide on how to build one below

I thought I was late to AI, but after intense research, I believe it’s just the beginning. Of course, nothing goes up in a straight line. Booms and busts will happen. But not engaging with it and calling it a bubble is unacceptable.

Can AI judges make human-like reasoning mistakes? Roll two dice. At least one is a 6. Many say 1 in 6 Correct 1 in 11 A simple logic trap—also seen in AI models. #AI #Ontology #DID #Web3

a 2-minute cleanup can make claude feel like you upgraded models the video is not about some secret prompt. it is about deleting the junk you already gave claude. old skills, dead connectors, half-used tools, random workspace instructions, abandoned experiments. all of that can sit in the background and quietly poison the context. then people say “claude got worse.” maybe. or maybe you turned it into a crowded browser with 47 tabs open and asked it to think clearly. this matters because agent performance is not only model quality. it is context hygiene, tool discipline and how much irrelevant garbage the system has to carry before answering. the article is really about the same operator lesson: smarter ai does not save a messy setup. the model can be strong and still produce weak work if the environment around it is noisy. sometimes the bottleneck is not claude. it is the junk drawer you connected to it.

One thing that stood out to me about ExpRL is that it exposes a limitation of how we currently think about AI progress Most benchmarks only measure whether a model arrives at the correct answer They rarely measure how many useful ideas were explored and discarded along the way Human innovation often comes from failed attempts Not perfect ones A scientist may test hundreds of wrong hypotheses before finding the right one Yet traditional RL treats every failed path as equally worthless ExpRL suggests there may be valuable information hidden inside those failures That idea feels important beyond math benchmarks As AI agents become more capable the winners may not be the models that find answers the fastest They may be the models that learn the most from being wrong This is also why verification matters Without reliable ways to evaluate intermediate reasoning and outputs we cannot distinguish productive exploration from pure hallucination The future of AI may depend less on generating answers And more on understanding which mistakes are worth learning from @ritualnet

The only tweet you need to survive AI. I recently sat down and wrote 10 "rules" to survive the upcoming AI era. No AI slop - just my genuine thoughts. If you're worried about job displacement, or any of the ripple effects of AI, I hope you read this: Rule #1: Be fast The #1 skill in the AI era is high agency. Information is no longer a moat. Your edge is now speed. Make it your goal to be someone who takes action quickly. Rule #2: Build a business AI is creating SO many business opportunities. Information is now abundant, and you have no excuse not to build a cash-flowing business. Rule #3: Use AI every single day Being AI-native is no longer optional. Integrate it into your life and make an effort to be on the cutting edge. Rule #4: EQ > IQ When everyone has access to the same AI tools, the differentiator is no longer intelligence; it's your ability to connect with other people. Rule #5: Prioritize health Start with the simple things: Sleep, exercise, hydration, and diet. Rule #6: Build a personal brand Distribution is the new moat. Making content is free, and the best distribution channel. Rule #7: Learn to sell This is the skill no one talks about. AI can build the product, write the copy, design the site, and run the ads, but it still can't build human connections. Rule #8: Invest Invest in yourself (skills) and in AI (pick-and-shovels). This is the biggest bubble we've ever seen - take advantage of it. Rule #9: Network Build your network. Your network is your ultimate safety net and how you'll land new opportunities. Rule #10: Stay curious This is the meta-skill that feeds everything above. Those who can learn and iterate fast are now more valuable than ever. Save this so you don't forget it.

[APRO Intelligence #07] ──────────────── AI is incredible at STEM, coding, math. But empathy, aesthetic judgment, reading a room, sensing slow causality — AI is still very, very far from us. Daniela from Anthropic put it perfectly: "We are moving into an era where human core is becoming unprecedentedly expensive." Build with AI. Don't forget what makes you irreplaceable. → @APRO_Oracle

Guillermo Rauch, CEO of @vercel: "every task in the world eventually becomes a coding task. that's why every lab is racing to the same thing.. a coding agent that runs without you." the clearest breakdown of where coding AI actually goes next. worth more than any $500 agent course. watch it, then read how to run one while you sleep below.