People should stop hating on AI as long as OpenAI is no monopolist yet. If we don't embrace a diverse AI industry including as much open source as possible, there will still be AI in the future, but only from a few monopolists. Currently it looks like OpenAI/Microsoft. Apple doesn't do well with AI yet and Anthropic may also have a chance (and isn't much more sympathetic than OpenAI). Discouraging average people from AI now only stops the grasroot movements, but not the industry players.

They promised a few month ago to release an open model again.
Yeah, it's not there yet. And when it will be there it will suck. The impressive models come from China since quite some time. Meta had its chance with Llama4 and released models that are okay but too large for most people. Then Qwen3 came and released a series of models in all sizes with good performance stealing them the show.

They have IP ranges they dislike. If you're unlucky your hoster got the IPv4 addresses only recently and then Microsoft does not like your mails. I once had a server with an IP from a block that was reserved for a long time and they blocked everything. Other than some people I never had a problem with Google, though. I think you should have SPF/DKIM/DMARC and then Google is happy. I'd wish it would be that easy for Hotmail/Outlook.

If you diverge from the average, you can't be part of a majority.

You just need to interpret it right, but sadly the benchmark got hyped too much.
Human preference does not mean quality. It is highly affected by tone, formatting, length and other factors that are nice-to-have but not as important as correct answers.

Even when doing SPF and DKIM right and sending a few mails per month, they still want you to manually apply for whitelisting if you run an own mail server. Good that not too many people I communicate with use it, because they really do not want your mail if you aren't one of the big players.

In principle yes, because "This does not give Mozilla any ownership in that content." does not exclude giving them a license you do not want to give them.

Almost every website has something in the ToS like "You give us a world wide ... sublicensible, ... non-exclusive license for your content", which, while only covers their their ass that they can show your content, still gives them unlimited rights to use your content, even when you still retain the ownership. If (and only if) Mozilla would now really get a license to your content when you use Firefox, you would give away a free license for stuff you might not want to give away for free. There it doesn't matter if you're the owner. In fact, in many legislations you cannot even give away copyright, but you can always sell an exclusive license that is almost the same, i.e., giving someone else the rights and preventing yourself from giving them multiple persons.

Whatever the hierarchy is for the drafts ... maybe the boss doesn't look at 100 but only at a selection, but the point is, that in the end there is work for the artist to create a good version (and tailored variations, think of a full corporate design based on the best candidate), but before there is less work to produce a larger amount of drafts. Creating three drafts by hand cost time and money and then it's hit or miss if you get the job.
Creating 100 and selecting ten using AI costs less time and if the boss says all ten are not what they are looking for, you get a chance for another ten, because you don't need another two weeks for them. And then you've got the candidate everyone likes and can start creating high quality versions of it that would be hard to create using AI.
I just don't see humans replaced anytime soon. But I see humans getting great tools to work faster and better. Take what is useful for you, and you don't have to use everything someone wants to sell you. There is a lot of BS marketed right now, but somewhere in there are the gems.

I am a huge fan of software that doesn't have huge changes often. But the Plasma desktop has a very stable user experience. I am really happy about each update, there is always something getting better even in the small details, but the last revolution was between KDE 4 and 5, afterward it got only more polished and stable. There is no need for an LTS, if your usual releases do not break things. The only point is that distributions like Debian, which like to stick with older versions, have it easier to patch their "LTS" branch if they can adopt an upstream LTS release.

Humans are the problem. First the paying user base and second the LMArena benchmark that is won by winning side-by-side comparisons rated by users. Of course the models are optimized for positivity.

Neural Networks are a subset of AI. Not the other way round.

They added ToS saying they may use user transmitted content. The problem is that the ToS apply to more than their services, so it can be interpreted that they may at any time start using content you enter in Firefox. They probably won't, but they are now allowed to.

Not for Mozilla, but for Firefox.

There will be quite a lot of forks. Most will die soon, but there will be one or two surviving, maybe forked by one of the larger groups like GNU, FSF, Apache Foundation, etc. and it can only get better than what Mozilla does to the browser.
And that Mozilla thinks they can't survive without Google money is the best indication for their mismanagement. Other organizations create larger projects with less money, entirely by donations. Mozilla invests more money into management than into development.

I really like what Firefox should be, but only because the rest is worse, it doesn't mean that Firefox is in a good state. Let some people who care about free software and users take over, remove all telemetry, reduce features nobody asked for, focus on bug fixing and performance and Firefox will be better than ever.

"I'm curious if different people ask the same AI the same predictive question at the same time do they all get the same answer every time?"

It depends.
You get the same probabilities every time. If you do greedy sampling, you get the same result every time. If you have an AI with control over the samplers, set temperature to 0 or use a TopK sampler with k=1.

If you do random sampling, it depends on how likely the answer is. If you ask if the sky is blue, then the word "potato" is a pretty unlikely answer, while both yes and no are much more likely, the no possibly followed by some explanation of atmospheric phenomena.

For the sampling you can play around here:
https://ancillary-proxy.atarimworker.io?url=https%3A%2F%2Fartefact2.github.io%2Fll...

At the bottom you get the probabilities and an LLM would choose with a weighted random distribution from them. At the top you can change the probabilities using different samplers, like for example, TopK, Temperature, and more.

Why should it be able to predict it? What crucial information does it have to make a good prediction? I mean bring back the kraken who predicted soccer results, he wasn't that bad at it.