MostlyStable

Maybe I misunderstood the point you were trying to make with your original comment, but you seemed to be suggesting that this piece of hardware could negate the risks (from the banks perspective), of allowing their apps to support non-secure devices. If your suggestion only applies to a very small subset of customers, then that doesn't allow them to put their app (which every customer can use), on the untrusted devices.

You are suggesting an entirely second piece of hardware when there are people in this thread claiming (IMO correctly) that many customers prefer an app over the browser because it's very mildly more convenient.

And there is a very large swath of banking customers who I wouldn't be surprised to learn that banks will earn less than $130 from over the course of a decade.

That's a bold claim. It was almost certainly true for models in 2023, when it was written. I think that the recent spate of mathematical discoveries at least significantly narrow the implications for even current models (one can argue that they are only re-mixing existing mathematical concepts to solve problems that humans initially generated, but in my opinion, at some point you are really narrowing what it means for an idea to be "novel"). And I would seriously consider taking odds against it for models within the next 3 years (since that's how long it's been since it was published).

We are basically already at the point where if what current frontier models can do isn't truly "novel", then the vast majority of humans have never had a novel idea in their lives. My three years from now guess is that if whatever it is that frontier models can do at that point isn't "novel", then no human ever will have had a novel idea.

Remixing existing work into something novel is ~ the definition of creativity.

Oh cool, where did they copy the solutions to hundreds of open math questions from?

I self host (I have a large overlap with the services in TFA), and love it. I will never go back and think that more people should self host (or at least...join small groups of people who designate one person to manage it for a small number of people).

And believing that the current AI economy might have a period of rapid devaluation in the near future is pretty reasonable. But, much like the original .com bubble, the idea that AI as a technology will not survive and thrive through such a pop, and become an ever more important part of our future, is a belief that I can't comprehend. Anthropic and OpenAI very well may go belly up. AI as a technology isn't going anywhere.

And as much as I'd like to believe that more people may self host more things in the future, self hosting will never, even if it was adopted by literally everyone, become as big a deal as AI, even if AI hit a wall tomorrow and didn't get any better than the current frontier models.

Depending on your location and power costs though, before the recent crises, it was probably generally worth it get newer hardware for the power efficiency gains. I have an old DDR3 dual-socket XEON E5-2650 system, and while it's more than sufficient for everything I run on it (and then some), its idle power consumption is pretty bad. If prices do ever come down, I'll probably try to replace it purely for that reason, even though I'm not sure I'll ever need more raw horsepower than it's got.

I had always wondered about the magrittr package name, but never enough to actually go find out. That's a cool origin.

The bigger HVAC win, in my opinion, if you have time-of-use billing, is to crank the temp when power is cheap (usually overnight), and, assuming your house is relatively well insulated, letting it ride through the rest of the day. This does lead to some funny things like the fact that I use heavier blankets in the summer than in the winter because I chill the house so far overnight, but for anyone whose climate is relatively moderate and with a relatively modern house, this can save quite a bit on power costs.

Admittedly, this will probably not work that well in a Minnesota winter, or an Arizona summer, but I think a surprising number of people could benefit.

Pangram says 90% Human. And this didn't come of as particularly AI to my eyes.

Cool concept. As they mention, more work needs to be done on long term durability, although they saw no degradation in their (very short) testing period. To me, the humidity control seems more worthwhile/useful than the power generation, although that could be a useful solution for small IoT devices like door and window sensors that don't need much power.

I like it as a proof of concept, would love to see a more serious engineering effort to demonstrate what real world costs, durability, and aesthetics might look like (if significant portions of indoor walls are going to be covered in this stuff, it needs to be able to look decent).

What's the end goal of this "build ugly things and depress land values" strategy? If it's to drive out current residents and buy cheap, to what goal? If, once you have bought cheap, you stop building ugly things and start doing high value things.....then land values go back up and you have defeated your own purpose.