If some government is spending $20 million to investigate this, that seems like an insane misuse of taxpayer money.

I once warned a manager at a smallish company that their fantasy of doing manufacturing would never happen partly because of a lack of QC and the lack of anybody with authority to shut a project down if it was not meeting spec. "We need that guy!" the manager said, and I came back, "If you had that guy you'd fire him the first time he told you something you didn't want to hear." Musk is in a pickle right now and I'm sure he really doesn't want to hear that the tank failed an X-ray inspection and the whole craft needs to be taken apart to make sure it's OK, thereby missing the launch window. And I'm sure the QC guy knows that. So as I warned that guy who once asked for my advice, this is what he got.

"researchers that found state-of-the-art large language models face complete performance collapse beyond certain complexity thresholds"

Humans also face complete performance collapse with cognitive tasks beyond certain complexity thresholds

BTW, the ascii glorp above is a unicode em-dash, proving that I am an LLM and hence not a disinterested party to the discussion. :)

At least that's what I heard skeeted on Bluesky: https://bsky.app/profile/danbi...

And of course this is no surpriseâ"an inconclusive study that can be used to justify a controversial headline is gold in the infotainment biz.

How would an LLM accurately determine which cases were "easy"? They don't reason, you know. What they do is useful and interesting, but it's essentially channeling: what is in its giant language model is the raw material, and the prompt is what starts the channeling. Because its dataset is so large, the channeling can be remarkably accurate, as long as the answer is already in some sense known and represented in the dataset.

But if it's not, then the answer is just going to be wrong. And even if it is, whether the answer comes out as something useful is chancy, because what it's doing is not synthesis—it's prediction based on a dataset. This can look a lot like synthesis, but it's really not.

I was hoping for football fields, but no luck. :(

No, that's a bad idea. A menu bar at the top of the screen is a much bigger target to hit, and easy to find by muscle memory. The file menu is always in the same place, regardless of what app you're using, and the buttons extend infinitely up above the screen. By contrast, a menu bar tied to the window moves around whenever the window moves, so you always have to visually find it again, and target size is just the size of the button and ends at the top of the window.

She hunted and pecked through the beginnings of a reasonably successful career as a magazine copywriter back in the day. I tried to tell her it would be worth her while to spend a few hours with Mavis Beacon, but she insisted she had her way of doing things and that was that. Two index fingers, staring at the keyboard instead of the screen. Meanwhile, I was younger than her but did learn touch typing on a manual in high school. Anyway a year or so after I gave up trying to convince her to spend some time learning to type properly, I walked in on her as she was working and she was holding her hands in the home position, index fingers hovering above F and J, eyes on the screen, and doing a good 80 wpm as she pounded out copy. When I pointed this out she looked at her hands and said, "I don't know about any of that, I just adjusted what I was doing to get a little faster." Well, that's why they teach it that way, but some people gotta ice skate uphill, y'know.

Queen's English is "has got." "Has gotten" is U.S. English. Presumably author is from the UK.

Jobs and Wozniak got rich off Apple, Gates and Balmer off Microsoft. Sinclair was already rich. Tandy, Commodore, Atari, and IBM had hugely popular machines but no "rock stars" single-handedly responsible for their development, and bad business decisions ultimately killed them. Similarly Coleco, which had a great chance to undercut the PC with the Adam and its cheap letter quality printer, but they were too ambitious and by the time they worked out their manufacturing problems the PC had taken root. But the PC killed the rest of the industry by killing itself, making the first clones possible which could run object code generated for other manufacturers' machines, which was Microsoft's second stage to orbit after providing Level II Basic for the TRS-80. It wasn't MIcrosoft's intent, but imagine what today's computer ecosystem would look like if all software was still architecture-specific and there were a dozen or more popular models to choose from.
--
Apple and the rest had room to grow because the big names like DEC, Data General, and even IBM were focused on business and saw them as toys. They bought and ate anything that looked like it might compete with them, such as the CP/M office systems which might be a credible threat to minicomputers like the DEC PDP series. That was another gap IBM threaded by being IBM.

Waste Management used to have pretty good customer service, if they missed a pickup you just called, they'd send a truck out. Bin broken, call and they'd come fix it. Easy peasey. Now, all you can get is a call center in India that insists you got service even when you did not. They have missed three pickups in a row now. When my wife finally got a human being after 3+ hours on hold with multiple calls, the rep was completely unsurprised that it ended with a cancellation request and offered no pushback. There are three other companies doing trash pickup in our subdivision, one of them will now get our business, and apparently we're not alone.

Hal's insanity trigger was actually revealed in the book version of 2001 itself.

There's actually a solid history to show that being a late adopter isn't always a bad thing. There's clearly some value in LLMs, but at this point most of what we are hearing is speculative hype intended to kite stock prices. Basically a ponzi scheme.

I'm sure that some value will drop out of this in the end. I am not at all sure what it will look like, except, probably not much like what the hucksters are promoting.

When things are clearer, it will make sense to invest. Right now, it's probably best to let other people burn cash. Particularly since one of the things they're doing is completely destroying copyright law, so when they're done, we can just copy whatever they did with impunity.

In order to get the $X from the Canadian government, you need a lot more than just a ASD diagnosis. You have to be affected to the point where it significantly affects your ability to live day-to-day life. It's not automatic based on the diagnosis - you have to have a Dr. fill out a huge form outlining exactly how you are affected and how your disability interferes with living. Things like being unable to feed or dress yourself, or challenges in doing normal life. Feel free to look over the form: https://www.canada.ca/en/reven...

The $X is just a tax credit of around $2000/year. If you don't make enough money to pay taxes, you get nothing back - but if you're dependent on a family member for care, you can transfer some of that to them. It does give you access to some other potential supports. As someone supporting an adult daughter with profound autism - it's hardly some windfall compared to caring for someone who will never be able to live independently.

I've also been diagnosed with ADHD and autism, but with a bit of Concerta and I do just fine. I don't have a disability, because my diagnosis doesn't affect my ability to live.