Béla Pátkai

CTO @ampcontrolio · prev @DaimlerTruckBus · postdoc @Cambridge_Uni · PhD @ENS_TampereUni – software is eating the grid

[Béla Pátkai @patk_ai]

Software is eating the grid.

For years the answer to rising demand was simple: build more. Supply chains found a better one - replace inventory with information. It's energy's turn. The grid we have can do more than we think.

[Béla Pátkai @patk_ai]

@paulg @NicoleBarbaro Agreed, but now we need to decide if AI is a moral or a technical question 🤔

[Maria Popova @themarginalian]

The meaning of maturity – Ursula K. Le Guin on what it really takes to grow up https://t.co/03AMtjFyTX

[Béla Pátkai @patk_ai]

@Plinz Does truly open AI have a @Linus__Torvalds yet? That may be a bigger Achilles heel than lack of lobby power.

[Béla Pátkai @patk_ai]

@taliasmiley @spakhm it does come down to really simple principles, doesn't it! this is why I think that building a great culture is not that hard, as long as you are able to drill down and find the core

[Béla Pátkai @patk_ai]

@spakhm Regarding "lay out a plan": by this time you probably have a few wins behind you with customer, so you can refer to those gently: "we still solve your problems faster / recover from these better than others, you have evidence of this, let us do it again"

[Naval @naval]

The best way to pace the frontier is to hold the labs fully liable for the behavior of their models.

[Peyman Milanfar @docmilanfar]

Recursive Self Improvement

[Béla Pátkai @patk_ai]

@dhh @wholemars Looking foreward to the discussion! Please someone write a draft of 10 bullet points about how abundance will play out. The last point can be the monthly payment to individuals. Honest question, not cynical, not a hater.

[Naval @naval]

"When liability exposure approaches 0—when no one pays the price for unverified failures—verification budgets collapse...Deployers flood the Runaway Risk Zone with unmonitored agents, capturing the gains of automation while socializing catastrophic risks."

[Slava Akhmechet @spakhm]

On what made YC great c. 2009:

- yc greatness is best understood by comparing it to trad ecosystem at that time
- the Pioneer Way yc office was surrounded by trees, equipment suppliers, small scale logistics companies, and of course startups. People were loading and unloading trucks around the corner. By contrast trad VC offices were tucked away in country club grounds on Sand Hill Rd
- YC parking lot must have had cars normal ppl buy because I never noticed one. On Sand Hill you couldn't help but notice Porsches, Maseratis and so on
- when you entered YC you'd walk through a white empty vestibule they never quite knew what to do with, then enter the cheery "orange room" filled with tables and benches for founder dinners. Sand Hill offices were somber forgettable places set up by professional decorators who would never have to work there themselves
- at YC you were greeted by pg in his cargo shorts and polo, and Jessica always brimming with energy
- on Sand Hill you were greeted by an immaculately dressed receptionist who looked like she'd been sent over by central casting. (It was always a woman; I've been rejected dozens of times by venture capitalists on Sand Hill, and I don't think I was greeted by a man once.) She'd offer coffee. You'd decline. She'd walk you to a conference room. The walk was usually long and silent; the only sound you'd hear was the loud clicking of her high heels against the floor. Then you'd settle in around an oversized mahogany table and wait. The wait was never so long that you could call them on it, but somehow always long enough that it felt disrespectful
- in fairness you sometimes had to wait at YC too. This was because Paul was walking around somewhere with a founder, enthralled by an idea they were both brainstorming. And in a few minutes he'd be equally enthralled by an idea he was about to be brainstorming with you
- while waiting for pg you'd look around. On one side there was tlb's Anybots lab, stocked with all kinds of electronics and robot parts. On the other side there were rice cookers (or were they pressure cookers?) pg and co used to cook founders dinners. A Sand Hill partner might take you to a Michelin star restaurant, but nobody on Sand Hill was ever going to cook for you
- there was a structural element. YC partners were writing small checks and treated it as an experiment. They could be very tough on you if you screwed up, but it always felt like they stood next to you, helping you peer into the future to make the thing work
- but partnership in a Sand Hill firm was not an experiment. It was a highly desirable career path, with all the implications that come with that. Almost everyone who worked there was afraid to make a mistake or to look dumb in front of their partners. And so on Sand Hill founders felt like they were looked down upon from Mt Olympus. (What we were too green to realize is that sitting on Mt Olympus is a very precarious thing)
- pg & tlb were hackers. Not in their distant past-- now. Tlb was building robots. Pg was building HN on top of Arc on top of Scheme. It was very difficult to imagine anyone on Sand Hill doing that
- anyone could see the bones of yc-- they were the OGs of building in public. Pg's essays and HN comments would document the choices, the problems, the decisions. Since he worked out many ideas right there on the page, he'd implicitly document his way of thinking and the attitude of self-belief, curiosity, and optimism that's so indispensable while working in the fog
- eventually trad firms would clone elements of YC. Seed funds, batches, blogs, and so on. It worked, but only to a point. Sand Hill ppl would often ask-- what’s the secret? We’re doing all the same stuff, why are you founders so loyal to YC but not to us? Of course everything was out in the open for anyone to see, there wasn't anything insightful to add

/fin

[Ahmad @TheAhmadOsman]

We will not have another Library of Alexandria situation with AI

We all have the right to this new information technology and civilizational infrastructure

[Béla Pátkai @patk_ai]

@TheAhmadOsman Is anyone working on this or anything similar? I would add the right to own all your data in physical form (a drive), audit logs for any data you share, your fully local inference, reproducibility of decisions made about you (e.g. in healthcare, finance, law enforcement)

[mark_l_watson @mark_l_watson]

I wish people would look to AI to solve non-software development problems more than the low-hanging fruit that is easier trained models for coding.

I have been critical of Meta and Google for privacy issues, but for me they have become the ‘not so bad guys’ because of their consumer facing AI tech: my wife and I spend $5/month family sharing Gemini and it is fantastic for non-technical research, search, and general utility. Muse by Meta, without giving it access to private data like email, login rights, etc., is a practical little tool for recurring information gathering and distillation.

Non-tech AI use has the advantage in that it is something that might just be used for ten minutes a day to save time so it is not as intrusive into human life. Google and Meta should thank Anthropic and OpenAI for making them look a little better in comparison.

[Naval @naval]

Challenge what they say, not who they are.

[Béla Pátkai @patk_ai]

@relizarov I don't suppose Chinese labs are open and friendly because they are altruistic. They simply wanted to catch up and get in the game. Let's see what will happen when they are leading.

[clem 🤗 @ClementDelangue]

We need 100x more transparency in AI!!!

[Paul Graham @paulg]

This is why it's a good thing that American labs are in the lead. If Chinese labs were, their employees wouldn't be able to warn us about threats. But Chinese labs will encounter the same threats within a year, in complete silence.

[Justin Schroeder @jpschroeder]

Can’t make this up…the MacBook is using its webcam to look at its screen in a mirror to improve AMD Radeon chip support in Omarchy.

[mark_l_watson @mark_l_watson]

Since I retired 2 years ago I have been getting back into using Prolog. I don't have a lot of professional Prolog experience (less than 6 months, and I have been writing code since 1966) but I enjoy the language and for some classes of problems it is very expressive and compact. BTW, I set the minimum purchase price for all my Leanpub books to $0 so if you don't like reading books online, you can also make a zero-cost purchase (or pay me 😀):

[Béla Pátkai @patk_ai]

@Plinz @shostekofsky Because humans have skin in the game? (But personally I wouldn't know, just speculating)