Oh please, please jam it between CTRL and ALT, just like the Windows and Fn keys. Touch typists and gamers have it way too easy these days. Better yet, remove CTRL from one side and ALT from the other to make room. Who needs both on both sides anyway? That's just a waste.

I read the article and it doesn't tell me any more than the summary. What were the complaints about? If they're complaining about something offensive in the content, sure, I understand that. But it sounds like they were just complaints that AI was involved in any way.

Who the fuck cares? AI is a tool. Think of it as a tool of the person who ultimately signed off on the work. Would anyone care if the copywriter used a pencil or typewriter or word processor or even just threw a box of magnetic poetry at the fridge? No? They why worry that the name of the tool is (very misleadingly) called "AI"?

"But it's putting people out of work!" I hear you cry. Oh, boo hoo. Every technology advance ever has done that. You used to have transcriptionists who would type up a hand-written manuscript. They're gone. Or typesetters who would painstakingly arrange actual bits of lead to be inked to put words on a page. It's just the next verse of the same song we've been singing since the invention of the wheel.

US healthcare wait times are less in some (but not all) cases.

Long wait times are the norm even to see your primary care physician where I am. I live in a sparsely-populated region of the US and our wait times are outrageous. My friends in the UK are appalled at how long we have to wait for routine care.

And I'll never understand why vision and dental aren't considered "healthcare". The US system is a mess.

• Right now, 24MB seems wasteful for a single web page.
• A few years ago, 2MB would have seemed wasteful.
• Before that, hundreds of KB would have been wasteful.
• And in the pre-web days applications would be wasteful if they used tens of KB.
• Or single-digit numbers of KB.
• In even earlier days, using extra individual bytes was wasteful.
• People have been known to optimize code just to find individual *bits* of unused space.

The point is, the problem of creeping bloat has existed ever since the world's second program was written, and people have been complaining about it for just as long.

A wise person once said, "Software expands to fill all available memory." The same can be said for filling bandwidth. As long as an excess resource exists there's no incentive to optimize its use. Where we once scraped to find unused bits we now casually toss around megabytes and gigabytes. At 300 bps there might have been room for a few ASCII art flourishes on the BBS. Now that a typical home connection is literally hundreds of thousands of times faster there's room video sidebars and frameworks that save the programmer a little effort.

It all boils down to time. In general people are willing to wait a few seconds for a response. Anything slower than that and we start looking for ways to speed it up. Anything faster than that isn't noticed, so isn't worth optimizing away. Page load times stay roughly constant even as bandwidth increases because that's what's fast enough on the human scale. Any extra bandwidth gets consumed by higher bandwidth art, or supplemental data, or just wasteful programmers importing a whole framework for that very special .toUpperCase() method that's so much better than the built-in one. There's no need to optimize. It's fast enough.

And yeah, it sucks for the people on the trailing edge of the technology where it's not fast enough. Conscientious developers will at least try to stay within the last generation's resource limits, but I doubt you're going to see anyone trying to optimize out single bytes or even single kilobytes at this point.

Very much the same for me. My high school teacher was barely keeping a chapter ahead of the students in the "BASIC for Dummies" (or whatever the real title was) book we were using. I had already burned through the entire book and had done all the assignments. He tried to give me another textbook to work through and I told him no. I'd already completed more coursework than the rest of the class would the entire term. I'd happily do something challenging but I wasn't going to slog through all the same 10 PRINT "hello, world"; 20 GOTO 10 sort of assignments again. Another kid was in the same boat I was. To his credit, the teacher actually found some elementary school teachers in the district who had these newfangled "computer" things in their classrooms but didn't know what to do with them. We spent the rest of the term going around doing custom programming work for whatever ideas the teachers had.

Ah, TRS-80 and Apple //e. Those were the days.

I expect you're right, that true AGI will require some way to interact with the environment. We learn by exploring, after all. This could be a virtual body in a virtual environment. It just has to have consistent interactions.

But that has nothing to do with the current crop of poseur AIs. LLMs "know" one thing: The statistical likelihood of one word following another. That's it. It turns out that when you have a huge body of data to train on you can make a model that is amazingly realistic. You can even give it considerations like "say this in the style of..." and it will turn out something that reads like it could have been written by that person.

That's all it is. A statistical parlor trick. Truth isn't a consideration. Common sense or general fund of knowledge isn't a consideration. It's just that "people eat chickens" is statistically more likely to appear in the training data than "chickens eat people". When an LLM hallucinates it's because the words in the false statement are at least as likely as the words in the true statement. There's no knowledge of any sort except for a huge table of statistics about which words tend to follow each other in the training set.

Lies, damned lies, and statistics.

All that means is they have to build in a delay. Lose contact for a few minutes? No problem. Lose contact for a few days? Something must be wrong, turn on the check engine light and engage limp-home mode.

I used to advocate everyone switching to UTC, but I think I have a better idea now. Return to solar noon. When the sun is at its highest point for the day, that's 12:00. And just for fun let's add a variable-length hour. Sunrise is 06:00, by definition. Sunset is 18:00. Wherever you are the daylight period is always exactly 12 hours long. Who's with me?

I didn't notice the ransom demand when I skimmed the article, but if it was there I withdraw my remark.

"Suspected ransomware attack"? How to you have a suspected ransomware attack? The defining feature of ransomware is (wait for it) the demand for ransom. That pretty much gives away the game. If no one has demanded ransom, it's not ransomware. QED.

"It's when I first realized, wow this is really not a safe model."

No, it's not a safe model. You train a model with unsafe data, you get unsafe output. If you train a model with uncurated crap and tell it to figure out the connections for itself, don't be surprised when it does. Pro-life supporters often use terminology describing the pro-choice advocates as demonic or monstrous, and often show pictures of bloody aborted fetuses. Is it any wonder that the model learns to associate "pro-choice" with such things?

No amount of guardrails is going to stop this from happening. The only way to avoid it is to train the model on human-vetted input. And that will only stop accidentally producing such things. It won't stop the models from following explicit prompts to produce disturbing output. If it was trained on pictures of violence, and pictures of puppies, you can tell it "Now, produce a picture of violence being done to a puppy."

I just don't get how some people think they can declare certain subjects to be digital thoughtcrime that are off limits to the AIs, which have been specifically trained to make exactly those associations.

I don't actually use it, but it's made by Facebook. I'm just assuming there are ads. If not now, then as soon as it gets any traction in the market.

In other words, water prices are too low.

From a strictly free-market point of view you're correct, of course. So let's think about the ramifications of that.

The current residents are using water. They're happy, their demand is pretty much on par with the supply, prices are low. Then the data center is built. Water prices are low so they decide to go with evaporative water cooling. Now the demand exceeds the supply. But it's a free market, so the prices go up. The company who owns the data center isn't concerned. Water is still priced less than alternative cooling systems. But now, the residents' water bills skyrocket. They're competing with a big company with deep pockets for a limited resource. The company can afford it. The people can't. Eventually people start moving away because they simply can't afford to live there. Yay, free market. Sure, the prices eventually stabilize but only after the demand decreases because so many people have left.

Or heck, why stop there? The water is from a river, right? Who says it belongs to the townies? Let's just locate the pumps upstream from the town and divert it all to the data center. In a free market, isn't that a smart idea? Why pay someone else for a resource that they're just scooping up from the environment when you can go out and scoop it up for yourself? (We'll just ignore the fact that the Colorado River is already the subject of tons of treaties and regulations shackling the power of the invisible hand. Those are bad things, right?)

0xjessel, who posted the announcement, also posted "there are no plans to support APIs for 3p clients. but posts from federated threads users can be seen on other mastodon apps if you follow them".

So, no. Just a way to facilitate bulk posters to post in even more bulk. You, the lowly end user, need to stick to the approved interface that earns them ad revenue.

Clearly the only fair thing to do is to sell by auction. Not the food, necessarily. At a place like Wendy's there is rarely a shortage of product. No, auction off the wait time. Auction off the place in line.

When you enter the queue you pay some amount to expedite your service. You're slotted into the line based on the amount you pay. If the next person offers more they get slotted in in front of you. If they offer less they get slotted in behind you. If they offer nothing they get added to the end of the queue. You can always increase your payment to move up, but it's non-refundable. You forfeit the fee if you leave the queue.

There. A perfectly fair free-market solution that considers the resource that's actually scarce and quickly settles in to whatever pricing the market will bear at any given moment. You're welcome.