@Serantych this is why Tesla's data advantage is impossible to replicate
English
chuplung
117 posts
Andrej Karpathy on how Tesla turned 1 million cars into a self-improving AI:
"We beam a tiny detector to the fleet, ask it to flag matches, and the cars send the images back. No firmware update. Done in a day."
they didn't label the training data. they made a million cars label it for them
→ when the model fails on a rare case, they train a small detector just for it, push it to the fleet, and the cars hunt that case in the wild and send it back
→ "we have the biggest dataset of 'except right turn' stop signs on earth. I'm basically certain of that"
→ and every driver is labeling for free - each turn of the wheel shows the model how to drive that exact spot
bookmark this ↓
sunick@Serantych
English
Anthropic Engineer Andrej Karpathy:
"Stop training from scratch. Take what the model already learned - adapt it - ship it
At OpenAI we replaced months of work with a few lines of code. Better results every time "
the Karpathy formula for working with AI:
step 1 → stop building from zero. the model already learned the hard part - just point it at your task
step 2 → stop adding complexity, the simplest architecture beat everything in 2014 - simpler always wins
step 3 → train only the last layer and freeze everything else - minutes instead of months. better results
he taught this at Stanford 10 years ago - then applied it at OpenAI, Tesla, and Anthropic
the advice never changed - because it never stopped working
watch - bookmark, then read article below ↓
codila@0xCodila
English
@helicerat0x Karpathy building his second brain with Claude Code. of course he is
English
OpenAI co-founder, Andrej Karpathy:
"there was no code that would create a knowledge base based on a bunch of facts"
now an LLM compiles what he reads into a personal wiki. one source can touch 10-15 pages - summaries, cross-links, flagged contradictions. he rarely writes a page himself
his research wiki: ~100 articles, ~400K words
the whole pattern is one markdown file - paste it into Claude Code and the agent builds the rest
watch the talk, then read how to build yours below
DegenCalls@Degen_calls_sol
English
most people use Claude Code the same way every day. open terminal, type the same things.
that's the least powerful way to use it.
the real unlock: build a system where Claude runs itself.
level 1 - codify everything you do into skills and loops
level 2 -give Claude memory so it improves from past runs
level 3 - wrap it in a UI. buttons, voice, metrics. no terminal needed
level 4 -share it with your team or clients in one click
"90% of the value is in levels 1 and 2. the fancy UI is just the cherry on top."
bookmark this ↓
chuplung@choopyplug1
Jess Yan (Claude Managed Agents, Anthropic): "we set agents tasks overnight. we wake up and the backlog is resolved and bugs are squashed." she also said: "I talk to Claude more than I talk to my colleagues." one example: 4,000 companies on a waitlist full of duplicates. she spun up an agent in 30 minutes - it cleaned the list, scored each company, sent daily invites to the best ones. "the limit is no longer our personal capacity. it's how much we can delegate at once." bookmark this ↓
English
Andrej Karpathy quietly published 9 rules for building AI agents.
Rule 1: stop writing prompts.
"If you find yourself iterating on a single message at 3 in the morning, you are still in the prompting era."
A friend who runs agent infra at a trading firm read it once and deleted half the harness his team built last quarter.
The whole paper argues most agents die from a weak harness, not a weak model. Everything you added to compensate for the model becomes dead weight the moment the model improves.
The rule near the middle, about letting the loop delete its own work and start over, is the part he screenshotted. It contradicts how almost everyone builds right now.
The closing section on where the bottleneck goes next is the whole paper in one line.
He said he read it twice, second time with his own repo open beside it.
Everyone is still tuning prompts. Karpathy already moved on.
Moysei@0xMoysei
English
chuplung 已转推
Jess Yan (Claude Managed Agents, Anthropic):
"we set agents tasks overnight. we wake up and the backlog is resolved and bugs are squashed."
she also said: "I talk to Claude more than I talk to my colleagues."
one example: 4,000 companies on a waitlist full of duplicates. she spun up an agent in 30 minutes - it cleaned the list, scored each company, sent daily invites to the best ones.
"the limit is no longer our personal capacity. it's how much we can delegate at once."
bookmark this ↓
chuplung@choopyplug1
Dario Amodei (CEO of Anthropic) said something nobody wants to hear. "we could have 5-10% GDP growth and 10% unemployment at the same time. never happened before. but it's not logically inconsistent." high GDP always meant lots of jobs. AI breaks that assumption. also from the interview: → Anthropic revenue: $0 → $100M → $1B → $10B in three years → co-work built in a week and a half almost entirely by Claude → "software is going to become cheap. maybe essentially free" bookmark this ↓
English
@choopyplug1 Dream big: turn Claude into your autonomous assistant that runs the loops while you take a well-deserved coffee break.
English
@jordanfjf02 skills + routines + CLAUDE.md. start small, automate one workflow, then build from there
English
@choopyplug1 Totalmente de acuerdo. Automatizar tareas con Claude Code puede aumentar significativamente la productividad. ¿Cómo logras implementar sistemas autoejecutables con Claude?
Español
PICOCLAW PUTS AN AI AGENT ON A $10 CHIP THAT BOOTS IN UNDER 1 SECOND
OpenClaw needed a Mac Mini.
PicoClaw runs on tiny hardware you could lose in a drawer.
The comparison is nasty:
OpenClaw: TypeScript, over 1GB RAM, more than 500s startup, Mac Mini hardware.
PicoClaw: Go, under 10MB RAM, boots in under 1 second, runs on RISC-V, ARM and x86.
It also says 95% of the core code was generated by agents, then cleaned up with a human in the loop.
And it still does actual agent work:
proactive alerts, morning briefings, task priorities, status updates, prompt tweaks.
Not an AI agent sitting on a laptop with 40 tabs open.
An AI agent small enough to run on a $10 board.
The Mac Mini got replaced by a tiny lobster with a boot time under one second.
Ostap@0xOstap
English
Anthropic Director of Engineering:
"We spent years figuring out how to make models write good code. Now we have to figure out how humans and agents actually co-author the codebase."
in this video, Fiona Fung shares what actually breaks when you scale autonomous agents and how to rebuild code review and ownership from scratch.
This breakdown packs more pure engineering value than a $500 premium bootcamp.
Bookmark and watch the full video below
Blaber@4rblaber
Ex-Google data scientist breaks down how to build AI agent harness and loop engineering in 19 minutes. agent harness (giving the model hands) + loop engineering (teaching it to think in steps) + llm ops & eval (making it bulletproof). An LLM is just a raw brain until you give it a proper architecture. Harness + loops + ops + evals = production-ready agent system. This 19-minute breakdown packs more pure engineering value than a $1,000 premium bootcamp. Bookmark this masterclass and use it as your go-to manual for turning raw models into systems you can trust.
English
A 27-YEAR-OLD BUILDER TURNED A SINGLE IDEA INTO AN ENTIRE GAME STUDIO WITH AI
Fable researched the topic, designed the characters, built the environments, animated the scenes and even generated the branding and marketing assets around the final product
AI is no longer creating isolated assets, it's connecting research, development, design and marketing into one continuous workflow that starts with a single prompt
The biggest opportunity isn't building faster, it's launching products that already come with everything needed to grow
Bookmark this
Insomnia@insomnia_vip
English
Harvard + BCG ran the cleanest test yet on AI at work: 758 elite consultants, GPT-4, 18 real tasks.
on tasks AI was good at: +40% quality, 25% faster, 12% more done. the weakest performers jumped the most - +43%.
then the twist. on tasks outside AI's range, the AI-users did worse - right only 60-70% of the time, vs 84% without AI.
that boundary is the "jagged frontier": two tasks look equally easy, but one is inside AI's power and one is a trap. and it's invisible - you can't feel where the edge is.
the fix is how you work. centaurs split the job cleanly - human here, AI there. cyborgs blend every step, checking constantly. the losers just "fell asleep at the wheel" and trusted the output.
the lesson: AI didn't reward the smartest. it rewarded the ones who knew what it was bad at.
~50-min Stanford lecture, free. the study every AI builder should know ↓
Clodex@0xClodex
Anthropic's platform team on a counterintuitive idea for building AI agents: not all tokens are equal. everyone's lever is the same - spend more tokens, get a better result. they asked: what if you give tokens jobs instead? so they split the work. some tokens execute. some advise the executor. some grade it against a rubric. some "dream" - review past runs and write lessons to memory for next time. then the key test: hold the token budget fixed across all of them. if tokens were fungible, every strategy should score the same. they didn't. on a financial-analysis benchmark, plain executing hit a perfect answer 42% of the time - the smarter strategies, up to 75%. the kicker on cost: to brute-force one perfect answer, executing burns ~1.8M tokens. advise and grade get there for a fraction - same budget, better jobs. ~15-min talk, free. Anthropic on why *how* you spend tokens beats *how many* ↓
English
Andrej Karpathy tried to give AI a keyboard and mouse back in 2015.
It failed spectacularly - and the reason tells you everything about why AI agents are NOW finally working.
AI agents in 2024 aren't starting from zero.
They're starting from everything we've ever written.
That changes the entire game.
Better than any $1000 building bootcamp. Watch + read the guide below.
Morty@0xMortyx
English
AI developer:
"Annotating 1,200 images by hand would take days
With NVIDIA Locate Anything 3B, it took minutes - and the YOLO dataset came out ready to train immediately."
in 8 minutes he walks through the exact Python project that auto-annotates any image folder and exports a complete dataset in YOLO, COCO, VOC, or CSV format
no labeling tool, no manual bounding boxes, just a JSON config file and one terminal command
full guide below👇
Fokki@0x_fokki
English
@AnatoliKopadze amazing find bro
the ones ignoring this will feel it soon
English
Head of Engineering Shopify:
"AI writes the code, AI reviews the code. Your job is just to write the loops around it."
26 minutes on how AI changed the way 3,000 engineers work inside a single company.
Ignoring it while everyone else uses AI to do more is the fastest way to fall behind.
Watch it, then read the step by step guide on loops below.
Anatoli Kopadze@AnatoliKopadze
English
Boris Cherny, creator of Claude Code, Anthropic:
"Right now, this is just the golden age of the generalist. People that want to do more than one thing, it's never been more fun."
On June 2, at Acquired Unplugged presented by WorkOS, he explained how he went from coding daily to not writing code at all.
The first version of Claude Code wrote only 10 to 20% of Boris's code. A year and a half later he removed the IDE: he hadn't opened it in a month.
Per-engineer productivity at Anthropic is up 200%, and onboarding a new hire now takes two days instead of several weeks.
In 29 minutes, Boris explains why his job is no longer writing code, but writing loops that decide what to build next.
Watch it, then save the framework below.
yurshev@yurshevv
English
Anthropic just dropped 5 workshops on building self-improving agentic systems from scratch:
00:00 - Ship your first Claude agent
36:44 - Build memory for Claude agents
1:05:06 - Make your agent autonomous
1:26:46 - Set up a proactive agent
2:03:35 - self-improving agents (tools,skills)
These 3-hours of free Claude workshops will replace 10 paid agentic courses.
Watch today, then read article below on how to build a self-improving agentic system with Fable 5.
Codez@0xCodez
English