RevEng

There was no cyber attack here. They released a cert with a trivial key. They sold a lock with no pins. When people realized what was being delivered, they blew the whistle. They were killed by a self own of epic proportions.

Passive voice is a common technique to avoid taking responsibility.

Yes it did help, at least back in the day when they weren't doing tons of RL. If you just train on random text, your prior probability is to return the average of everything. Ask for software, get average software, or worse yet, randomly get either gold or shit. By adding constraints to your prompt like "you are an experienced engineer" it will prefer to return answers like an experienced engineer would give and less like a random noob would. It doesn't suddenly make it smart if it never was, but it helps to reduce the variance and bias towards the good outcomes.

They certainly are! Nvidia creates and maintains an awful lot of software. Selling GPUs is what gives them revenue but they produce a lot of software to generate additional pull through demand for their hardware.

Any examples of such schedulers?

That's how Nvidia works. They make arrangements with lots of companies to produce demos that use Nvidia products, especially ones that Nvidia is about to announce. That gives them lots to show at GTC. Companies are happy to work with them because having your name said by Jensen adds three points to your stock price.

Is this for real? If so, that's a major institutional and managerial failing.

AI is already taking many lives and ruining many others just by providing text responses to humans. The AI only needs a way to interact with the world and humans can be that conduit. The better AI gets, the more blindly people will do whatever it says.

From TFA:

> The second common variant is, instead of centralizing a service so that everyone in the company uses it, every department does it on their own. Department managers like it because it’s one less party not reporting to them to depend on (and nothing scares managers more than needing things from people not reporting to them.) And the people managing the department managers like it because they don’t need to deal with the centralized service department allegedly failing — and having to figure out if that’s really true and how to fix it, or if it’s an excuse made up by the other departments for failing at their own job. Here, too, if outcomes are what concerns you rather than what's easiest for senior leadership, standardizing on something across the company vs having everyone roll their own variant is a hard decision to make.

The argument is that centralization is the better outcome, even if it's less appealing to leadership. My counter argument is that centralization has downsides that need to be considered; it isn't inherently better.

I disagree with the centralization thought. I'm working on a project to create a central AI platform for a very large corporation. As you can imagine, such a large company already has several different platforms made by different branches within it. Each branch has something they are relatively happy with. But corporate sees this redundancy as waste, so they want to consolidate it into one.

The problem with one is the same with any other single source: you are stuck with whatever they offer on whatever timelines they provide. You are sharing this with a hundred thousand other people, so it's not tailored to your use case, but either made as generic as possible, or full of irrelevant features. Both of these cases make it worse for you.

I have often made in house libraries either based on or completely replacing a free equivalent because the free one doesn't meet our needs and has extra complications we don't need. Yes, it comes with its costs, but sometimes it's worth it for something that genuinely meets your needs.

By forcing everybody to use the same common platform, you are either forcing them to work around the mismatch between the platform and their needs, or your are forcing the platform to provide for everyone's needs. Neither is efficient and it harms every other user.

If you are going to make your own, you had better have a good reason for it, but you shouldn't be forced into a single solution for everything - it can end up being more expensive than making a few special purpose applications.

I'm almost certain it was parody.

There is a big difference in the cost of a 5-nines up time system in a heavily space constrained environment compared to a home hobby white box used for some coding. The GPUs alone cost 10x for the data center versions compared to the gaming versions even with similar specs.

The cost of online services is also largely a result of the cost of training (though hard to say exactly what that number is). Assuming you are using open weight models at home, you aren't paying for the training - someone else is.

There's a fine line, as with alcohol. A single cup of tea might feel refreshing. A monster energy drink is going to feel like a fight or flight response. Some people are also more sensitive than others and some are addicted more really than others.

Especially when there are many ways to do the quantization. You can get quite different results from different methods.

They are also an order of magnitude difference in number of parameters and that matters a lot.