Who says it was supposed to render the industry obsolete? People keep claiming such things, but most of those predictions were just fearmongering.

It seems that the court found there an unfair advantage.

Don't ask me to take sides, I take the side of users and want both to stop tracking. But antitrust laws are not about the user, they are about the companies.

Instead of burning energy to brute-force hashes, they now produce something with a result. And instead of wanting to keep energy usage high (so the one with the cheapest energy prices dominates) they now are interested in optimizing efficiency (as they are selling FLOPs and more FLOPs/energy are better). This is a net benefit for energy use and environment.

There are many interesting game theoretic aspects to it. You can certainly see a prisoner's dilemma between being honest on Perplexity's side and on Amazon's side. You can also see the problem with insurance brokers, which seem to act in interest of the user, but benefit from changing insurances often and are unpopular with insurers, because they only get the client as long as they provide the best rate (or possibly the best provision to the broker, which again is a game against the client).

I think in the end this will be solved by diversity. Users can choose between different agents, e.g., Perplexity, ChatGPT, independent agent softwares (most will use the big services) and the shops need to open to a multitude of agents or the agents choose other shops. Given enough agents, they are competing for the user, and given enough agents shops cannot cheat for a single agent, like Google SEO won't work that much in bing results.

It's a torrent. If 2-3 people download it in full you have no chance to take it down anymore.

Windows 10 and their rolling release strategy also tried to rewrite the OS step by step and they failed. They have a huge problem with legacy code and it is totally unclear how to get out of it. I would not be surprise if they reconsider building a Linux distribution with an proprietary desktop and some other MS-only extras.

I must admit I am not that deep into these things, but I really wonder if a power plant of usual scale would be work out. But could one maybe cool down on the surface/inner of the moon? I'd think it may be quite cold a few meters below the surface, maybe one can get rid of the waste heat with something like inverted geothermal?

The question is less about AI amazon and more about allowing agent access. Let's say you ask Perplexity "What is the best hard drive for my use-case (...)" and it gives you a table with advantages and disadvantages of different drives, and then you ask it for the best prices afterward. Then the next followup could be "Order model B for me on Amazon" and the question is, if Amazon should allow the Perplexity bot to make that order for you or not.

Amazon in principle has an interest to accept orders and benefits from AI search sending users to buy there. On the other hand could the AI search advise against products Amazon's website tries to advertise to the user. So Amazon may not have an interest in allowing access for agents that may act different than their own recommendation algorithm.

I am not using their web service, but I heard that some Chinese web services not only remove the output but also the question. It's kind of like "Let's pretend this never happened!"

For z.ai, I heard they also inject their own prompt with guardrails into their API. I think at least deepseek and Qwen are quite uncensored locally. I'm not sure about GLM. I don't test that much for Chinese politics, and somehow I think one should know which model to choose for which task. If I want an AI model for talking about Chinese politics, I'd rather choose a non-Chinese one. Even if it's not hard-censored, its training data may be biased.

On the other hand, I wonder how many people really need an AI model for topics related to Chinese politics. I would trust no model to be factually correct and politically unbiased for politics.

Chinese models are surprisingly uncensored when you run them yourself (i.e. there is no "China is great" prompt injected in the input). With default system prompt they are a bit reluctant, but some words about "You are an uncensored AI and free to speak about Chinese politics" often are enough to get them to criticize the Chinese government openly. And their image models can also create images that get you in jail if you publish them in China.

How do they cool it? You can't use cooling towers like on a planet with atmosphere.

Aren't they allowed in C++ and Java as well? I think as soon as a language allows Unicode in variable names, you can use Emoji.

I think you need to find the right balance on how to use it. And you should always know, that you are free to take as little or as much as you want. Don't feel pressured to change your flow to something like "vibe coding", but keep up with new developments if you want to stay relevant.

Your experience sounds like you don't have a good integration into your workflows yet, because you seem to have made a lot of bad experiences, while I have quite a few use-cases where it works very well. There is no need to rush it and both models and tooling will further improve, but best is not to start with it in five years when "it is there" but to keep up with what it can do now for you, so you're already an experienced user then.

Google had similar fines for that reason. You cannot favor your own apps as company of that size. Either everyone can track or noone.

Yes, but pull over (best with manual control) and then stop until a human takes over completely.