Saved on Hacker News · submitted by clmul
Mir Books – Books from the Soviet Era (mirtitles.org)
Drone Physics (iahmed.me)
June 2026 · 50 bookmarks
Mir Books – Books from the Soviet Era (mirtitles.org)
Drone Physics (iahmed.me)
KARPATHY JUST KILLED THE PROMPT ERA WITH A SINGLE DOCUMENT
prompts are easy. loops are hard. and writing fifty prompts a day is the work nobody does twice.
he shifts the burden to the harness.
you define the contract once. the model writes, reviews, restarts, and reconciles. you keep judgment. it keeps the loop.
the throughline is the same in every rule: the human owns the spec and the boundary. the model owns the execution and the bookkeeping.
planner never touches code. generator never grades itself. state lives on disk, not in context.
9 rules. start with one feature, not ten. most people are still typing prompts. this turns Claude into an agent that finishes the job on its own.
here is the official document from Karpathy explaining the architecture
What would the person you want to become do with the time you’re spending right now?
Do that.
Building a custom octocopter from scratch with no prior hardware experience (karolina.mgdubiel.com)
Attention is like a limited amount, you start the day with 100 tokens, scrolling tick-tock for an hour? 25 tokens, deep working for an hour? 25 tokens - what do you have left to do and how many tokens does it take?
I’m trying more and more to not spend tokens on things that don’t help (social media), etc.
Reaching back decades, I used to do a first draft longhand on file paper, cross bits out, rewrite bits. Then bang it out on a typewriter. Then once over with a red pen the next day, and a complete re-type.
I'm not sure that I could work that way now, but it was more deliberate. Less 'drive by' thought.
"Our Writing Tools Are Also Working on Our Thoughts"
(I'm talking essays for University here not deathless prose).
I got back to writing longer texts by mentally separating writing and editing. When writing, just write. Even when you think the paragraph could be better, keep on writing.
Only start editing when a substantial piece is ready. Clean up some wording, rewrite a paragraph or two.
Even then, don't overdo it. There is always something to improve, you'll never be done that way. Good enough is good enough, hit publish and go on write the next thing.
I started getting into typewriters. I could've repurposed an old X230 and disable/remove the network card physically. But I also wanted to stop staring at a screen when writing, so I gave the typewriter a try.
It's still early and I'm struggling to write more than a few lines at a time. Not surprising from how I've been commenting "witty" one-liners in comment threads for over a decade. I expect being able to write long-form with no backspacing will need a lot of time to learn.
But I want to take back my attention. If there's one thing I've learned in the last decade, is that one's attention is a precious resource and it's time to be more deliberate in how I spend it.
WordStar: A Writer's Word Processor (1996) (sfwriter.com)
Get rich the hard, slow, but guaranteed way. There are no lines at that counter.
If a good outcome is far enough out, humans discount its value to approximately zero.
This is why opportunity will always exist for those who can endure.
The best predictor of success for tech companies, at every stage from during the YC batch to public company with billions in revenue, is the rate of shipping new stuff.
In fact, since it will be what all the smartest people do, it will probably be prestigious to write for oneself. Which in turn means more people will claim to do it than will actually do it.
Not using LLMs to write for you won't be like not using Google Maps in a new city. It will be like choosing to run and lift weights, even though we now have machines that can transport us and lift weights for us.
It will be an uncommon choice not to have LLMs write for you, but it won't be a merely idiosyncratic one. It will be what all the people who care about thinking well do.
This may be extreme, or, completely stupid, but, why are we not using genetics to "grow" chips in a chemical soup yet? Similar to Verilog/VHDL, don't we have some similar language to express circuits using gene sequences?
45°C cooling design cuts data center water use to near zero (blogs.nvidia.com)
AI learns the “dark art” of RFIC design (spectrum.ieee.org)
The internet is completely deep fried
Influencers can’t compete with AI
I found this Pixar style AI character that
Generated 23.7M views w 44 posts
That account alone could make $30,000/mo+
Selling a beauty app + partnerships
So instead of paying a influencer $3,000 for one post
You can run up a bunch of these accounts in any style
Hyper-realistic
animated
Pixar
And have genuine assets that can
Print money on demand
This is the new meta
Marketing will never be the same
Hello, I wanted to share with you all a interactive map of the economics and physics constraints of the AI buildout. It has macro drivers, industrial chokepoints, and where that shows up in markets.
I've added 393 nodes and 562 edges to capture other supply / physics constraints as well.
There's no sign up, and no pay wall, it's all free.
Please let me know what you think!
GLM-5.2 – How to Run Locally (unsloth.ai)
Human intelligence is fundamentally a collective intelligence. We solve complex problems by participating in a vast cultural network that builds upon ideas across generations.
I believe the strongest AI systems will become a collective intelligence, too.
Since we started Sakana AI, our core conviction has been that the most powerful AI systems will be collaborative ecosystems, not isolated monoliths. Evolution innovates under constraints, and the future belongs to systems that explicitly learn how to coordinate collective intelligence.
Today, we are taking a major step toward that future with the launch of Sakana Fugu.
Fugu dynamically orchestrates the world’s best models to tackle complex tasks. We are proving that a well-orchestrated pool of swappable agents can match restricted frontier models like Fable and Mythos.
But Fugu is about more than just performance. I believe that Orchestration Models are the next frontier, beyond bigger models.
Relying on a single company’s model for national infrastructure is a massive risk. As recent export controls have shown, access to top models can disappear overnight.
Collective intelligence is the practical hedge against this concentration of power. Fugu simply routes around vendor restrictions by relying on an entirely swappable agent pool.
I am incredibly proud of our Tokyo team for shipping this. By orchestrating the world’s models, we are delivering the resilient blueprint required for AI sovereignty.
Read our full vision and results here:
sakana.ai/fugu-release 🐡
One reason it's more difficult to work with people you don't have faith in is simply that there are more possibilities to consider whenever they do something. You also have to consider all the ways they might have screwed up. Even if they didn't, you still have to think about it.
My opinion is that you will learn 10x more from studying failure than you will by studying success
As Charlie Munger says “inverse, always inverse."
An easy path to success is to avoid catastrophic failure. Teach yourself what not to do and rest takes care of itself
This is coming to every company that got drunk on the AI hype and gave unlimited access to tokens for their employees
Lot of them crunching the numbers and now just realizing how expensive tokens actually are
We’ve raised a $310M Series B to accelerate world models!
We believe AI that can understand and simulate the world will be one of the most important technologies of our time.
We're excited to partner with Natural Capital, Amazon, GV, AMD, IQT, and others to bring this to life.
AI demands more engineering discipline. Not less (charitydotwtf.substack.com)
SpaceX to buy Cursor for $60B (reuters.com)
Humanity isn't ready for the coming intelligence explosion (economist.com)
A backdoor in a LinkedIn job offer (roman.pt)
How to setup a local coding agent on macOS (ikyle.me)
Adaptive PDFs (sgaud.com)
I don't know what they're complaining about. AI has freed us from the drudgery of craftsmanship, letting us focus on the important stuff—managerial and administrative work!
(There's a reason why I call it the MBA's stone. It transmutes all knowledge work into a problem of management.)
You can hate showing up early, being consistent, doing what you said you’d do, keeping your word after it’s no longer convenient, and a hundred other things.
You can hate all of it.
You just have to hate losing, more.
Something else wrong about #WWDC
These prerecorded events are causing the product managers and developers to lose touch with how the features land with people. There’s no “ooh, ahh” or room to explain and demo. It’s just … flat b-roll. There’s nothing conversational.
While presenting, be natural, why the affectation. I know it’s not easy but these events are prepped hard. #wwdc
The #1 rule for presentations: Do not look like Will Ferrell from Talladega Nights.
Curious what @MKBHD thinks? He has a knack for seeming natural on camera. @Apple, @letterboxd
У меня такой жёсткий эффект зловещей долины от #WWDC
Я понимаю что их в всех тренируют вести себя на камеру и вот эта манера жестов, тона и подача считаются самыми "доверительными" с точки зрения какого нибудь корпоративного исследования.
Но когда они ВСЕ такиек
I don’t think this #wwdc appealed to Wall Street either 😂😂
What a keynote! @swiftlysingh #wwdc
new Apple Intelligence:
#wwdc
From what I've learned so far about the golden age of mechanical watches, this is roughly the peak: a 35 mm IWC with a cal 853 and dauphine hands.
Public Domain Image Archive (pdimagearchive.org)
Our first model Mac-1 6.6B beating 3 giant models.
Running this model on my Macbook M3 24GB. (model takes only 7GB RAM)
It searches web, call tools, ask follow-ups, tell jokes, find contacts, search files, write emails, book events, write notes, set reminders and so much Siri can't do.
Read again, a 6.6B model.
Will share full 2000+ scenario test results & benchmark scores in 2 days.