Self-driving cars also operate at a far larger scale than individual human drivers. Over 300 people died in Boeing 737 crashes, but the entire aviation industry has not been shut down as a result.

the 737-max was grounded for almost two years, Boeing paid a $2bn+ fine and the CEO was fired.

To be fair, actual self-driving car business units have suffered similar fates. Uber's autonomy team was shut down after they killed Elaine Herzberg, and GM's Cruise was functionally destroyed after they contributed to a major bodily injury crash and tried to cover up their errors.

It's the level 2 consumer car companies that have mostly skated under a liability shield. Tesla FSD has been shit for a long time and paid very little price for it.

Google also bought capacity from xAI, and OpenAI have a deal to use Google compute which may be how things propagated? I still don't get why chatgpt.com would show a 404 because of an AI datacentre outage though

As someone who could be tempted by Julia but isn't involved in the community this was a very helpful read, thank you for sharing.

It really isn't very helpful, as many of the issues are either obsolete or exaggerated.

Out of curiosity, what would you like to hear him saying about non-LLM AI?

There's plenty of arguments to be had. I guess there could be some useful applications of non-LLM AI (eg. translation tools), but quality output is still a concern. Environmental impact is also still a concern. If output is of sufficient quality, who reaps the benefits is a concern ; if not, who pays the price of shittier products is a concern.

Whether LLM or non-LLM tech, the question of surveillance capitalism and how models need to be fed swaths of privacy-invading data (eg. self-driving cars filming everything all the time) is also a concern.

Overall, i think the ecological concern is still the biggest one. We are at the start of a major climate crisis, and any self-preserving measures were already hard fought by industry. I fear that any form of AI investment will lead (because it's already leading) to destroying any hope of saving an inhabitable planet.

Really only large language models require huge swaths of training data and extremely expensive training/inference. Other “AI”, like basic neural networks or something like path finding, is basically free. It’s just an algorithm, a small one that runs on computers.

The term AI is very overloaded. When people say AI, they really mean LLMs. AI like old Siri has basically no cost. It’s not environmentally damaging, as it runs on device. It’s not privacy gore, because it runs on device.

Or take, for example, lane keep assist systems on vehicles. This doesn’t require a huge data center or anything, it’s basic computer vision and a small on-vehicle model.

Yes AI has a “cost”, but for virtually all non-LLM AI the ecological cost is the same as any other application or algorithm. Siri doesn’t burn more energy than instagram.

Well that’s what I’m saying - a basic machine learning system is no more ecological cost than an algorithm. Because it just is an algorithm.

Computerization of cars is orthogonal, it’s not an AI problem. Things like engine control for fuel efficiency isn’t AI, it’s just an algorithm. Cruise control is just an algorithm. The cost is the same as any other software.

Now LLMs DO have a new cost. You really require super computers to run LLMs at scale, that much is true. But that’s very distinct from machine learning, it’s a different beast.

Nobody is complaining about, like, monster AI in The Binding of Isaac and how much ecological impact it has. Because it doesn’t. Well… that’s AI, and 10 years ago that’s the only AI people really talked about. Now AI really means LLMs. The ecological impact is from LLMs.

To play devil's advocate - unless you were to take the position of declaring bankruptcy on the possibility that a complex society of competing actors can agree on climate change - and therefore this being an unsolvable problem that you need AI to solve, as humans can't handle the complexity.

I find a lot of the debate on either climate change or AI collapses if you point out that "we shouldn't do this, we should ALL just do this" is an extremely unrealistic position in an international complex ecosystem of competing actors.

> "we shouldn't do this, we should ALL just do this" is an extremely unrealistic position

No it's not. It's only if people like you are so defeatist about it. Look what happened with CFC gazes, leaded paints and asbestos. Sure we still have more problems to face but some problems are so serious that we can in fact gather consensus.

> an unsolvable problem that you need AI to solve

That is completely wrong on so many level. I understand you're playing the devil's advocate, but do you understand this is a messianic-level delusion that's made up by religious psychopaths so people buy into their products without thinking twice? If you don't have hope, the only rational solution is not to give in to AI, it's to Luigi Mangione or Unabomber your way to a better world. Personally, i choose hope: we can solve this problem (and every other, who strangely have more or less the same culprits).

A great read - as with all of Sander's diffusion posts.

"Model collapse" is often overstated, as this paper demonstrates: https://arxiv.org/pdf/2404.01413

The original model collapse paper assumes you train networks on 100% synthetic data produced by the previous generation. But if you maintain some portion of real data then the problem is mitigated.

Not when you also including poisoning attacks.

I remember the original paper showing issues with even a couple percent of certain kinds of synthetic data too, not 100%.

What do you mean by poisoning attacks - stuff like Nightshade or Glaze? I was under the impression that those have largely failed to achieve their goals.

Those aren't poisoners. The are attempted protectors.

What problems do you run into for maths with Python? For linear algebra and ML with Jax/Numpy I find it quite readable.

I think problems is too strong a word, it's just how the syntax looks. And of course it's far better than say C++. But specifically, the lack of native vectors/matrices and very few infix operators (e.g dot product, and matrix multiplication with @ is clunky). Explicit but generic broadcasting using `.` also helps.

The way that types interact with methods _feels_ like maths. I.e. you have a method made up of mathematical operators, if you swap out an input, it typically does the natural and correct thing.

I think the point is that, from a compiler's perspective, it's not obvious how much you should be allowed to optimise code at the cost of changing the outcomes of floating points maths - do you allow 1e-10, or 1e-6, or 1e-4 level changes? Does your compiler have to run some test calcs to bound the scale of the change introduced by rewriting fp maths? Some compilers will let you opt in to rewriting floating point maths, but that's opt in so users understand that their numeric outputs might change between optimisation levels.

For more, there's a good post on this kind of flag in Rust: https://pythonspeed.com/articles/faster-float-math-rust/

Does it? Mojo was announced 6 months after ChatGPT was released.

I wouldn't say there was a pivot, AI development has explicitly been the goal since launch.

Oh, gotcha. I thought I recalled reading early material that emphasized the flexibility of Mojo emitting MLIR for many tasks, but it's possible that that was either pre-launch, or that I got confused when reading about MLIR itself.