petilon

Sure, but what if someone steals your money by impersonating you? Here too, ideally the victim is the bank, but now the onus is on you to convince the bank that they are the victim. They are going to say you're the victim, your identity got stolen, sorry you lost all your savings!

We need to update our laws. This is not "Identity Theft", this is "Negligent Verification Fraud", and it is the bank's fault because they were lax in their verification process.

Only one-third? They are slowing down. Crypto companies accounted for almost half of corporate donations in the 2024 election cycle [1].

[1] https://www.cnbc.com/2024/08/23/crypto-companies-are-pouring...

No, you can't rename a file in use, unless it is explicitly opened with a sharing flag (which most applications don't).

On Linux if a file can't be edited because some process is holding it open you can just mv the file to /tmp and create a new file in its place. The process that has it open is unaffected because the handle is independent of the path.

On Windows a file that is open in an application cannot be moved, plus it won't tell you which process has it open. Yes you can use some sysinternals tool but this is basic info that should be immediately available without installing some additional tools.

Exactly. We should stop saying "Identity Theft" because that lets financial institutions off the hook. "Identity Theft" sounds like it is either your fault for being careless with your identity, or the thief's fault for stealing it. How about "Negligent Verification Fraud"? That makes it clear it is the fault of the bank. It is impersonation enabled by the bank's lax verification process.

> insane amount of effort goes into getting more powerful models to run with the same or less RAM

The same can be said about operating system memory requirements. I am sure Linux and Windows kernel developers can confirm. Yet 30 years ago Solaris used to run comfortably in 16 MB of RAM, today you need 512 times that to run Linux.

Memory requirements of newer models will increase, so while the hardware may last 10 years it won't be able to run the latest models for 10 years.

Son's point is that even if the idea is solid, it will take at least 10 years to get it to work. That's too long to wait, the winners and losers in the AI race will be decided long before then.

Also: if it does work, all it does is save on the cost of electricity. But just 7% of operating a data center on earth is the cost of electricity. The rest are the chips and other things.

> You can call it a failure, it's just not a military one.

Would that work in corporate America? Try telling your boss, "I didn't actually fail because had you given me an unreasonable amount of time and resources I would have succeeded." Every project has an expectation of time and cost, and if you can't meet that expectation then you failed.

> doing it is slow and expensive, so there is insufficient political will to do it

So the US Navy is unable to achieve the objective in a reasonable timeframe and cost. That's the same as failure.

> Terrestrial television will surely follow

I hope not. I am saving $80 per month by recording over-the-air TV using Tivo. I only want the major networks, and recording them over the air is free. Tivo DVR is great for OTA and their service is still active but sadly they have stopped selling their DVR.

Last year Tim Cook gifted Trump a custom, one-of-a-kind glass plaque with a 24-karat gold base [1]. (Cook needed a policy outcome that would protect Apple's supply chain costs and avoid a costly 100% tariff on certain chips and components.)

You may have to make similar offerings if you want to use the latest version of ChatGPT.

[1] https://www.youtube.com/shorts/0O9QhwIkj5w

If you have ever made a painting and sold it, then you too profited from the work of thousands of artists. How so? Because your sense of what is art came from those who preceded you. You have seen the works of Picasso, Rembrandt, Monet and so on and your brain absorbed from their work, just like an LLM.

If an LLM generalizes from thousands of authors then it is no different from what your brain does.

I would agree, if they are inspecting static output of American AI models without using their compute resources.

What AI firms are trying to build is the artificial equivalent of a human brain. If a human learns from a source material and uses the knowledge in their career that doesn't violate the copyright law. If an artificial brain does the same then it doesn't violate it either. This is up to the courts to decide. Alibaba can't take the law into its own hands and decide what the punishment ought to be.

This also shows how Chinese firms are weak in AI algorithms, they can't build a model without stealing from American firms.

Has it been proved in a court of law that it is a copyright violation?

In some cases if the model regurgitates the original material then that is clearly copyright violation, but if the model "learns" from the source material just like a human brain would then that's not a copyright violation.

I think the analogies are appropriate. Anthropic took public data and added value on top of it. It is that added value that Alibaba is targeting. If it was the underlying data, that's freely available.

> If anything these models should be compelled to be public since they have been trained off public data.

Isn't that a bit like saying if you read books in a public library to pick up a new skill you should work for free?

> What an absurd overreach to call this an attack.

Would it be an attack to take your meal by force if you used a public recipe to prepare the meal?

Yes, but that doesn't mean you shouldn't listen to them. You listen to them and you evaluate their claims.

Why wouldn't you listen to a company on why their product solves the problem you're having?

If you want to know why you might want to buy a product ask the company that made it. If you want to know why you might not, ask their competitor. In both cases the answers are valuable, but in neither case should you take the answers at face value.

If you want to know why you shouldn't choose Windows it makes sense to get Steve Jobs' opinion and then evaluate the opinion.

Other than co-founding one of the most successful tech companies, that is.

Stephen Colbert, 60 Minutes, now 'The View' is being silenced.

Previously: A Florida jury found that flaws in Tesla's self-driving software were partly to blame for a crash that killed a 22-year-old woman in 2019 and severely injured her boyfriend.

The jury verdict requires Tesla to pay $243 million in punitive and compensatory damages to the parents of the woman and to her boyfriend.

https://www.reuters.com/world/us-judge-upholds-243-million-v...

Neocloud is a low margin business. SpaceX has no advantage over competitors such as Oracle, Microsoft, AWS and others. Renting out hardware purchased from Nvidia will give them 15% profit margin. Terafab does not have any advantage over TSMC. TSMC has massive economy of scale and mature supply chain.

In short, there is nothing to explain the massive overpricing of SpaceX stock, it will come down to earth at some point. Don't be left holding the bag at that point.

Those naive retail investors can stay irrational longer than I can stay solvent.

SpaceX is being valued as an AI company and yet they don't have a frontier AI model. Goldman Sachs is predicting 100x growth in 4 years for xAI, but it is a failed company with no frontier model. Top employees have left, and the company is renting out datacenter capacity.

Satellite launch business has $4.1 billion in revenue, but only growing 8% annually. Most of the revenue is from Starlink. It has $11.4 billion in revenue, with around 50% growth. Blue Origin will offer them competition soon.

X, formerly Twitter, has around $2B revenue, limited potential.

The massive 2030 projections ($474B total, $144B Starlink, $322B AI) are Goldman Sachs' IPO roadshow model. The projections are so aggressive they feel scammy.

SpaceX's 2025 revenue is $18.7 billion. A typical premium valuation for a top-tier tech company might be around 10x to 14x revenue, which would imply a strong IPO valuation of roughly $187 billion to $262 billion.

The reason for the outlandish valuation is because of naive retail investors who believe Elon Musk has never failed at anything.

If that's true then it is a significant improvement over structs in C#.

But the distinction still matters. Value-vs-reference semantics affect equality, identity, nullability, arrays, collections, boxing, and performance.

So this is not just an implementation detail. If a language hides that distinction at the use site, it increases the burden on the reader and makes code harder to reason about.

I understand the rationale for value classes, but the implementation is flawed.

What will this code print:

   Point a = new Point(10, 10);
   Point b = a;
   a.x = 100;
   System.out.println(b.x);

Until now the answer was obvious. Now with the addition of value classes, the answer depends on whether Point is a value class or a reference class. So readability suffers with this design.

This is a violation of the principle of uniformity. In The Psychology of Computer Programming, Weinberg explains that uniformity is a psychological principle which says that users/programmers expect that things that look similar should do similar things, and conversely that things that look different should do different things.

If a programming language lets two constructs look nearly identical at the use site while having meaningfully different semantics, it increases the cognitive burden on the reader. Programmers must inspect the type declaration or rely on tooling to understand whether assignment, equality, identity, and mutation behave like ordinary reference objects or like values. That can make code harder to reason about and maintain.

This could have been fixed by requiring the use of the "value" keyword not just at declaration time but also at use time like this:

   value Point a = new Point(10, 10);
   Point b = a;
   a.x = 100;
   System.out.println(b.x);