jasomill

I'm more interested in the techniques used to generate these proofs, and how they can be improved to be as reliable at solving a broader class of problem than already formalized mathematical conjectures.

Most human jobs don't require innovation.

On the other hand, there's no evidence that work will dry up merely because it becomes more efficient, or even if it does, that this cannot lead to a better world. The trick is finding a way to prevent rapid changes in the nature of work from destroying too many lives, and that's a very human problem machines alone can't solve.

Maybe. And while this may be a useful superpower, it's hard to predict whether it would be an enjoyable one.

This stokes my suspicion that effective LLM input for software development will ultimately resemble more of a higher level programming language than the natural language in common use today, and that the design of such languages and related tooling will play at least as large a role as the models themselves.

This is not novel. Computers have been generating mathematical information that would be resource-prohibitive to produce by hand since the dawn of computing.

The ingenious aspect seems to be that we've found simpler ways to apply computing power to a broader range of problems, which it's important to note is as much a testament to the developers of proof assistants like Lean as it is to LLM developers, to say nothing of the countless mathematicians over many centuries that contributed training data.

Whether this will lead to many revolutions, or a handful of impressive results followed by a long dark spell remains to be seen.

I, for one, wouldn't place large bets on either.

Depending on what you and the other poster mean by "arranging UI stuff", AI could either solve the problem for you or create larger problems, especially if you're not opinionated on either technical or UI matters.

So long as the web views provide for accessibility (and thus UI automation), this sounds perfectly reasonable for the majority of applications.

Speaking of which, my dream is an "HTML light" standard with a subset of functionality specifically designed for local UI use, ideally with lightweight non-browser engines supporting this standard shipping with all major OS platforms as shared system components.

We had browsers that ran on systems with 8 MB RAM, after all, and while it's not a fair comparison, there's no technical reason why a system with three orders of magnitude more RAM should struggle to run more than a handful of applications merely because they use HTML and JavaScript for rendering.

Interesting that federal funding isn't included in this comparison, given that a quick web search suggests[1] it comprises anywhere from 20%-40% of state revenues.

[1] https://reason.org/commentary/ranking-the-states-most-and-le...

Paint now has Copilot features that require sign-in, so presumably involve network access.

And create single-pixel images. My most used application of Paint by far is creating single-pixel PNG files to work around the fact that "solid color" is not an option for lock/login screen backgrounds.

Do you have details on this? Given that this is classified as a network vulnerability of the Paint application itself rather than any shared system components, I assumed it had to do with the recently-introduced Copilot features which AFAIK are the only part of the application that directly interact with the network.

Fallout: New Vegas wins the award for allowing you to kill seemingly plot-essential NPCs. As far as I know, there are only two characters in the entire game you can interact with that can't be killed, both robots. One is a vendor, so you always have at least one shop, the other is the head of a faction that allows you to complete the game even after alienating (or killing) all the other factions.

It's impossible to speak highly enough of this game.

Also the System Shock and BioShock series.

Old habits die hard.

There was a time when the company I was working for was selling service contracts for DEC TZ30 tape drives to DEC itself because their own internal service organization was unwilling or unable to support them.

The application was software distribution, most likely to customers they were contractually obligated to support, and no newer drives were able to write the required tape format, so upgrading to newer drives wasn't an option.

Not only drivers. The initial release of Windows 95 predates USB itself, so the entire USB stack didn't appear until later OEM service releases that AFAIK were never released as updates for existing installs or even as retail products (unless you count MSDN or OEM System Builder products with their associated license restrictions).

Sure, but what are the odds they're using these rather than HD 3.5" disks purchased from whatever new old stock they've been able to source over the past 20-odd years?

You mean the cards with $1,600 MSRP that regularly sell for $7,000+ on eBay?

Depends on whether you’re buying or selling.

There were no PowerPC Macs in the '80s. Running Mac software on non-Apple 68k machines was a thing, though[1].

Apple itself also had a project, started in the late '80s, to port the Mac OS to x86[2].

Incidentally, PowerPC Mac OSes, especially later pre-OS X OSes like Mac OS 9.x, run quite well today on vanilla QEMU, no hacking required.

[1] https://www.atarimagazines.com/startv3n6/mac_pc_on_st.html

[2] https://en.wikipedia.org/wiki/Star_Trek_project

If access to a system is critical, I'm going to have some way to access the system other than SSH (out-of-band access through a BMC, serial port, or VM host) that doesn't exclusively rely on the same means of authentication.

On systems that receive regular updates, SSH can fail for many reasons other than authentication problems, after all.

That's an 8GB card. For gaming (DLSS, etc.), you can still get newer 8GB cards like the 3070 which represent solid upgrades over the 1070 Ti on the used market for under $300. Still not great since similar cards were available at similar prices new a couple years ago, but at least not trading at multiples of MSRP like higher-end cards.

How would DLSS, which runs on GPU hardware — and specifically, as I understand it, GPU hardware that accelerates AI inference — eliminate the GPU?

And if a system is intended to be modular and upgradable, what would combining the GPU and CPU into a single module contribute to this goal?

Now an architecture that separates the memory controller from the CPU and gives pluggable CPUs and GPUs (and storage and network controllers, for that matter) equal access to a single pool of RAM in a secure and high-performance way, that would certainly be interesting, and not only or even primarily for gaming.

Only if you don't intend to maintain, debug, or even understand what the code is doing.

Or at least not without the benefit of more capable coding agents than you used to create it.

And only if the coding agents in question can be trusted to not deliberately or accidentally fail to provide you with critical information ("Now that you mention it, the program does occasionally POST the contents of ~/.aws/credentials to a web server. As this produces no output, I didn't consider it relevant to your question about whether it correctly outputs the string 'hello world'.").

Unplugging a PC containing a Maxwell+ card doesn't properly set its clock speed.

Even if this is an intentional nerf by Nvidia, which I doubt, then then nerf is buggy.

The minimum configuration appears to be 24GB / 18 core, which at least one vendor, Microsoft, is selling for $2,599.99, which is at least closer to $2,399 than $3,000+.

If datacenters aren't lining up to buy 64GB DGX Sparks for $5,000, what makes you think they'd be interested in 24GB RTX Sparks with fewer cores and less attractive form factors for almost half that?

And if there's a market for a DGX Spark configuration with a higher compute-to-RAM ratio, why isn't Nvidia offering it?

My best guess is that the entry-level Sparks are mostly intended for people who would otherwise be in the market for a traditional high-end laptop, who might be enticed to spend a few hundred more for at least the promise of local AI, while the higher-end models are intended for more serious AI development.

With that said, it's not hard to imagine there isn't nearly as large a built-in market for high-end SFF Linux PCs as there is for high-end Windows laptops.

If this were cost-effective, why wouldn't they just buy DGX Sparks, with similar performance at similar price points and a more convenient form factor?

Or RTX Spark and run their workloads under WSL, which as the article notes, has minimal overhead relative to running Linux directly?

This sounds more like a bug than a feature, especially given that Linux on battery runs at full clock speed and downclocks when plugged in, and Windows does the opposite, suggesting they could at least mitigate the problem with no changes to their software stack by inverting a conditional in the firmware and adding a Linux kernel parameter to control memory allocation as apparently already exists for Windows.

Also various versions of OS/2, or for something completely different, iRMX (PC versions that co-install with DOS or Windows).

For NeXT OSes, OPENSTEP 4.x also ran on x86.

I can vouch for this: whenever there's heavy snowfall (in Indianapolis, so less common and severe than Michigan, but same ballpark), nearly all cars I see stuck in snowbanks are AWD/4WD.

Personally, I'm comfortable driving anything but a RWD pick-up truck with no payload in snow and ice, try to drive at an appropriate speed for weather conditions, and rarely have trouble.