Starboy

@demishassabis @IsomorphicLabs How come none of these other CEOs have a similar vision? AI can be applied in so many other useful areas beyond video/meme generation and some half baked ai assistant.

I’ve always believed the No.1 application of AI should be to improve human health. That work started with AlphaFold, and now at @IsomorphicLabs with the mission to reimagine drug discovery and one day solve all disease! We are turbocharging that goal with $2.1B in new funding.

@realBigBrainAI Exactly what I’ve been saying. Why has no one built anything groundbreaking - even the subject matter experts with more tools and processing power available to them. Something is fundamentally not adding up here.

Demis Hassabis (DeepMind CEO) on why no vibe-coded AAA game has topped the App Store yet, despite the tools being good enough.

@Mnilax In a recent interview he said you can basically leave these agents running with no supervision. That’s contradictory to what he’s saying here.

Karpathy threw a grenade at every senior engineer who still treats LLMs as a toy. his actual words: the worst thing an expert can do right now is reject them. most experts read it as a threat, but it's advice. his framing: > the gap between "AI tools are bad" and "AI tools are useful when used right" is professional discipline, not capability > agents have cognitive deficits. they fail in ways nothing in the training set anticipated > the experts who reject LLMs lose to experts who learn to wrangle them > "models have so many cognitive deficits. but you can route around them" routing around the deficits is what CLAUDE.md was invented for. Karpathy himself wrote 4 rules. across 30 codebases they took my Claude error rate from 41% down to 11%. solid drop. but his rules pre-date the slop era going public. I bolted on 8 more, tuned to the failure modes that surfaced after January. got it down to 3%. a CLAUDE.md does not raise Claude's IQ. it lowers his slop floor. that is the entire game. open the article underneath. the model is not the bottleneck. your config is.

@iddris Encapsulated Modularity.

question: can anyone explain what taste is? not the formal definition — but in your own words define “taste”. here are the 3 rules of your answer : shouldn’t be longer than one sentence. must be original to your best ability. can’t be in response to someone else’s.

@MarioNawfal This is what AI compute needs to be used for not creating some bs images and memes.

Mind-blowing moment from the DeepMind doc: Demis Hassabis gets told AlphaFold can predict all 1-2 billion known proteins in just a month. He instantly says: “Do it.” This is how history gets made

@iddris @adcock_brett Marc Cuban might be right. The future of robotics might not be humanoids. Cars can drive themselves. Vacuum cleaners operate themselves, so do dish washers. There’s a solution for manufacturing and assembly lines and it’s not humanoids.

here’s something i think the team can get ahead of @adcock_brett . as we move towards generalized humanoids, we’ll need consistent behavioral experience across systems. essentially: HIG for embodied AI. the coordination cues in the recent figure demo inspired the thought.

@eliguerron Impressive.

When I was at Apple, I loved working on micro interactions that you see all over the OS. Now that I’m not an apple I still like to solve for these little problems that really annoyed me. In this case, I designed a backspace button with a speed controller, so by just pressing it you can delete by letter and then immediately by word as you stretch it, without having to wait (like it usually does on the OS) and then if you stretch a little more, you can speed delete through words… I’m also working on another one where you can repair the words if you over-deleted it by accident 😜 (it also has haptic feedback, which makes it really fun)

@Bitcoin_Teddy They are both right. It’s possible to create completely private systems but it’s not being done because of the financial incentives.

Joe Rogan & Google AI Researcher Ray Kurzweil Get Into an Awkward Exchange Over Protecting Data from Intelligence Agencies KURZWEIL: "We have the ability to keep total privacy in a device...We know how to build perfect privacy." ROGAN: "How do we do it?" KURZWEIL: Long pause...

@vitrupo What’s this dude talking about? You can’t even see the satellites from earth, how will it be seen from millions of miles away?? How about a scenario where ETs have enough mental power, they don’t need data centers in space like primitive people do? Has he thought about that?

David Kipping says orbital data centers may imply we are all alone in the universe. If AI compute moves into space, it should leave artificial rings of warm infrared light. But we seem to live in a totally natural universe, with no hint anywhere of anything artificial. That makes the Fermi paradox harder to dismiss.

@rohanpaul_ai I have to disagree with Andrej on this one. The systems are not there yet where you can hit a button and sit back. Maybe when we get to AGI but now these systems still need a lot of babysitting. If he is right FSD would have already been solved but it’s not.

Andrej Karpathy: "To get the most out of the tools that have become available now, you have to remove yourself as the bottleneck. You cannot be there to prompt the next thing. You need to take yourself outside the loop. You have to arrange things such that they are completely autonomous. The more you can maximize your token throughput and not be in the loop, the better. This is the goal. So, I kind of mentioned that the name of the game now is to increase your leverage. I put in very few tokens just once in a while, and a huge amount of stuff happens on my behalf." --- From @NoPriorsPod YT channel (link in comment)

@rohanpaul_ai Steve Jobs never wrote a line of code. Elon Musk doesn’t have a phd in rocket science. These people are visionaries who surrounded themselves with incredibly smart people.

Sam Altman: "There was a time when we used to make fun of the “idea guy,” who only had an idea and needed someone technical to build it. But now, people who just really deeply understand their users and can’t code at all, I want to fund those people."

@60Minutes One of the pioneering founders of human genome sequencing including mRNA used in the c vaccines. But as with any breakthrough discovery, it can be used for good just as much as it can be used for not so good things.

"This is the first synthetic species,” microbiologist J. Craig Venter told 60 Minutes’ Steve Kroft in 2010. His team of scientists created a synthetic bacteria designed on a computer, with man-made DNA. Venter says that from concept to completion, it took 15 years and $40 million to make. Venter, who died this week at 79, was one of the most famous scientists in the world, known for his pioneering work in deciphering the human genetic code.

@DevinOlsenn @CincyDan11 Weak programming. Easiest way out could have been to do the reverse of what it did to park. Vision only didn’t help in this scenario.

@CincyDan11 I mean this is literally a video of the car getting stuck in a parking spot that it put itself into...

FSD struggles to get out of parking spot due to lack of front bumper camera. This is a clip from my latest YouTube video, but it felt worthwhile to post separately and discuss. This to me is a clear example of FSD getting itself into a situation that it can’t get out of due to a lack of a front bumper camera. You can see the car wants to back up to gain vision, but becuase the spot was so tight it couldn’t back up anymore. What do you all think?

@yacineMTB Both thinking and understanding are like sight. Everybody has eyes but sees things differently. Thinking can be mimicked but cannot be duplicated.

you can outsource your thinking but you cannot outsource your understanding

@iddris It requires some personal info. Same as a functional assistant app. I don’t want to deal with securing anyone’s data. Companies need to understand that as well. On device solutions will get faster “buy-in” than anything else.

@StarboyTech what’s stopping you from shipping ?

hmmm.

@iddris I built an AI assistant (for personal use of course) that outperforms anything on the market today.These companies are so obsessed with user data it blinds them from building what is truly useful to the user. Don’t try to predict what users want, users know exactly what they want

@StarboyTech what are you building? i’m genuinely curious.

people don’t want to solve hard-problems anymore since ai made it infinitely easier to ship junk and half-baked ideas.

@TansuYegen They didn’t figure out how to make it less noisy? Great design but functionality still average and this is what is wrong with products today.

China just rebuilt the razor from scratch, it already gets called the Rolls-Royce of shaving gear 🪒

@iddris The real issue is lack of creativity and same approach to solving old problems. Why aren’t software engineers billionaires by now? So many of the everyday user apps can be reimagined with Maps, Navigation, To-do, a real “AI personal assistant” - not all the bs available today.

@StarboyTech the real issue is we obsess over ai that boosts generic productivity & “work”. real usefulness is deeply personal. you can’t productize or generalize it.

the more you abstract technology, the more normal people instantly feel its usefulness. abstraction reveals the value to most people. builders need to abstractmaxx as hard as possible. strip the nerd guts. make the magic obvious.