What model did you use? I used GPT-5 and it responded with Chiefs, Titans, Texans, Ravens, Giants, Eagles, Saints, and Browns. Looks right.

Ironically, ChatGPT could have probably generated your post much more quickly with a prompt of "Write me a lengthy Slashdot post challenging the value of AI, highlighting that it only regurgitates things that it's been trained on, can't generate repeatable output, and can't be held accountable".

So, train it on more?

A colleague of mine trained a custom GPT on a bunch of our internal documentation and code. I asked it a question about which of our internal APIs to use for a particular scenario and (after 2-3 minutes of burning presumably an ungodly amount of electricity) it correctly told me that such an API did not exist, but gave me the ones that do exist including a code snippet showing how they could be used.

Saved me at least 10-15 minutes of combing through libraries and a few minutes writing the code myself.

That's the point. If viewers want the sort of 10-15 minute videos you describe, YouTube is happy to tune their profile in a way that serves them up.

I like the 10-15 minute format of video you seem to want to promote, and we're both on Slashdot so probably have at least some aligned interests. There's a higher-than-average chance that I'm in your target audience. Why so hostile?

I've been using YouTube as a primary source of video entertainment for probably 6-7 years. When will they turn on me?

I find that it does a remarkably good job of adapting to my tastes. I went through a kick where I really enjoyed cooking videos, and waddya know those were served to me. I eventually got tired of that and now only see the highlights. The channels I enjoy which post long videos (hr+) still consistently show up in my feed with their infrequent new posts. I sometimes get shorter (sub 10min) videos related to my current interests.

YouTube has a lot of good content on it. My recommendations to those struggling to find things to watch is to 1) make an account and 2) look outside of the recommendation algorithm if your account is new (Google "best channel for X" for your interests, or check out this comment section where other Slashdotters have made suggestions.

I just opened YouTube and the first 6 suggested videos (which I can see without scrolling) are: 20min, 30min, 6min, 13min, 15min, 21min.

Average runtime? 17.5 minutes. An anecdote isn't data, but Youtube has clearly not "abandoned" the sort of video you describe.

Nope. Denmark's oil production is trivially small (~25% of the US per capita). Sweden, Iceland, and Finland produce none.

Interesting take on an article about a guy whose disorder led him to a murder/suicide. He could probably have used a bit of (non-AI) help.

1) Because few organisations have the funding/willpower/know-how to maintain the infrastructure needed to keep the data on-site in a way that meets business continuity requirements.

2) Even fewer organisations have the funding/willpower/know-how to provide 24/7 on-site support for that infrastructure when something goes catastrophically wrong and you need an expert to look at the infrastructure to figure out how to fix.

That second point specifically seems to be related to what MSFT is being squirrely about. My read is that when they get a call at 3am and some Scottish policeman says it's CRITICAL that they get their Sharepoint up and running, MSFT can't guarantee that the engineer will be based in the UK.

And they'll all be able to mutually understand that pidgin version of AI English? If the point of language is to enable communication, that seems like a victory.

A friend of mine works at a university. They've noticed a trend over the past few years of foreign students who can achieve high marks on their written assignments (by writing in their native language and having a tool like ChatGPT translate, presumably), but struggle to engage in classroom discussion because they haven't had a need to develop English language skills to nearly the same degree as in the past.

Maybe that'll become irrelevant as AI tools for live interpretation improve? But for the time being it means that many of the students aren't getting the full value of the educational experience.

I don't think anyone's arguing that there aren't resources required to keep old emails, just that they're utterly insignificant.

If we assume that every single one of the 80 million people in the UK has 3GB of deleteable emails somewhere in the cloud, we're talking about 240 PB of data. Let's multiply that by a factor of 4 for replication, indexing, and whatever else I haven't considered, and then round it up to a nice even 1000 PB (1 exabyte) of data.

A top-loader 4U storage server can take 90 drives at say 22TB each. Call it 2 PB. We'd need 500 total, and can fit 9 in a a 42U rack (leaving some space for switches and whatever else), so call it 60 racks. Round it up to 100 racks to be conservative.

100 racks is, what, a few rows out of hundreds in a data center? If I'm off by an order of magnitude somewhere, we're talking about a single large data hall?

It's just not simply an amount of infrastructure that matters for the effort that would go into trying to make a difference. Certainly not in the grand scheme of the UK's water consumption.

It's possible to hold a high bar for quality without belittling and humiliating those who are underperforming.

If my boss publicly humiliated the lower performers on the team when they didn't meet expectations, I'd be looking for new employment regardless of whether I'm one of the high performers that isn't a direct subject of verbal abuse.

Has Linux succeeded because Linus is brittle and shouty, or in spite of it?

I don't think that the possibility of being publicly humiliated because you made an error in judgment makes people more interested in contributing to the project.

To be fair to LLMs, the architecture of the human brain took a few billion years of evolutionary training on the experiences of billions of generations of our ancestors (the ones who survived, anyways) to get to the point where it is today.

In comparison, training GPT-4 in a few months with a few tens of thousands of GPUs doesn't seem so bad.