I bet @ATabarrok $100 that โSystems in GPT line will by 2025 make <$1B in customer revenue clearly tied to such systems. If product contains such as component, but also has other features, one needs to attribute best estimate % of product revenue to this one.โ Right Alex?
@girllich1 This looks like clickbait.
360k parameters is way too small to do anything except to memorize few questions.
TinyStories models are 2.5M, smallest GPT-2 is 125M
@lopp stick to "just rebuild and rest everything from scratch" model just because it's simpler?
(It's definitely true that incrementality is a source of bugs - but where are new formalisms which can make it safe...)
@lopp Microsoft had edit & continue for C++ in Visual Studio 2002 - it could literally patch a running executable with new code to speed up interactive debugging.
Do we have like zero progress in 25 years in terms of incremental building and testing?
Or is it more like people
@_smitop@IsaacKing314@sparr0@Aella_Girl CDT can model Newcomb's problem is the same sense classic physics can model entangled particles: you just get a wrong answer.
You can write a program which compares strategy CDT claims is optimal to a random choice and random choice is on top. That means that what CDT gives
@IsaacKing314@sparr0@Aella_Girl CDT just cannot model Newcomb's problem. It has no solution in CDT.
"But Omega couldn't have predicted" is not an argument for 2-boxing, it's an argument for not using CDT
@Autonomaly@burkov It's literally a documented feature: https://t.co/BR2VfgpJ3n
same in codex.
I'm not sure if Burkov just craves attention or he's "too smart" to read the docs.
@VictorTaelin Ehh a sufficiently smart Fable should be able to decompile your app back to the prompt.
If you're concerned that slop code is too much of a distractor you can export just your requests and ask Fable to distill it down to a single prompt
I don't think a bet on NVDA is a bet on AI as there might be much more efficient way to grow compute at scale than to buy NVIDIA chips.
NVIDIA was an obvious choice for compute years ago as getting people to work with something other than CUDA was hard. But CUDA familiarity premium might just evaporate as code running on these chips will be written by AI. You'd want cheapest chips, not most familiar ones.
@burkov@Mark_H_87 it's poorly maintained, on its last legs and might be using ingredients which are off.
We cannot have perfect information about businesses we interact with , but costly signalling tells us something
@burkov@Mark_H_87 If you go to a restaurant and you see that waiter is in dirty clothes and floor is dusty, would you eat there, or would you go to another one nearby which is clean?
There are correlations. It's quite possible that the dirty one has superior food. But it's also possible that
@atoiansk@levelsio Why do you mention websites when we now see frontier AI working with most complex math and finding bugs in kernel at unprecedented rate?
The guy who makes supersonic planes claims that AI is very relevant to hardware engineering: https://t.co/QwRLsdrR6i
12 years ago I founded Boom, as a software engineer doing hardware for the first time. Iโm probably the least handy founder in hard tech. My 12 y/o son does my soldering and 3D printing for home projects, and Iโve struggled with imposter syndrome the whole way through. What business does a software guy have building supersonic jets and turbines?
More than a decade in, the most insane realization is that software is the key to reaccelerating hardware engineering. But in a way that I donโt think is widely understood or implemented, even in the burgeoning hard tech community.
Most hardware engineering is actually software engineering in disguise. Itโs spreadsheet โcode,โ engineering calculations done as one-offs. Eg wing structural loads in, spar sizing design out. Thermal loads in, air conditioning system sizing out.
Everything is a calculation, and calculation is software. But these calculations are done essentially by hand, as if we are doing arithmetic but have not yet discovered algebra.
Hardware engineers have baby software, built in spreadsheets, the kind of crappy software I wrote as a kid before I understood architecture, unit tests, APIs, version control, and everything else that makes modern software good.
Our airplanes and engines are built with the software sophistication of a script kiddie.
So what we tell our engineers: donโt design the hardware, design the *algorithm* for designing the hardware. And take the algorithm seriouslyโas an algorithm and as a first-class engineered software system.
Never design a piece of hardware, such that you couldnโt redesign it tomorrow with a different set of assumptions.
This all sounds obvious, yet our experience at Boom is that it changes everything.
Ten years ago we couldnโt afford to build much software. We had a small teamโand there were few software engineers with the curiosity, brilliance, and vision required to learn hardware engineering.
But now AI changes the equation. Now everyone can code. The software engineers define the architecture. The hardware people bring the physics and engineering know-how.
We have a turbomachinery design tool that lets one engineer do in realtime what previously took 100 engineers six months. Another tool auto-generates 65% of entire aircraft structure. Iโm sure weโll figure out the other 35% too. Eventually these should not be multiple tools, but one integrated physics model of a complete vehicle, airframe and powertrain.
The results speak for themselves. A boomless supersonic airplane built by 50. A full scale supersonic turbine made by 150. And with this approach, I donโt think weโll need that many more people to build and certify a supersonic airliner.
You donโt need a big team to build big thingsโnot when everyone on the team is a software engineer.
@atoiansk@levelsio Do you really think frontier AI model cannot help with industrial automation? https://t.co/VvFozt3jMz
It can write software, as you said, design PCBs, etc.
Have you actually tried using these frontier models?..
Fable 5.2 shows substantially improved skills in controlling physical robots, even on tasks it has never explicitly been trained to do.
On RoboDojo, a benchmark with a simulated two-arm robot, it outperformed other models and most AI systems built to control robots.
@atoiansk@levelsio I'm not sure if 20x cheaper food is feasible in near future, but 10x more accessible healthcare probably is within few years. Then, perhaps, cheaper construction. That depends on regulators allowing that, of course
@atoiansk@levelsio How is that "speculative"? Do you really see no value in more robots, research, etc?
And inb4 "but that's not LLMs" - frontier AI models do quite well with robotics:
https://t.co/VvFozt3jMz
Fable 5.2 shows substantially improved skills in controlling physical robots, even on tasks it has never explicitly been trained to do.
On RoboDojo, a benchmark with a simulated two-arm robot, it outperformed other models and most AI systems built to control robots.