kimixa

Indeed, that seems unnecessarily complex for what is actually needed. I don't understand why the great grandparent comment seems to suggest it's an "unsolved" problem - as if grid-scale solar buildouts don't already have examples of things like motorized brushes on rails for exactly this already.

And it's always a numbers game - sure they're not /perfect/, but a few % efficiency loss is fine when it's competing against strapping every kilo of weight to tons of liquid hydrogen and oxygen and firing it into space. How much "extra" headroom to buffer those losses would that equivalent cost pay for?

And solar panels in space degrade over time too - between 1-5% per year depending on coatings/protections.

I feel that Cynics are often average.

Optimists are either rich, or destitute. And though you probably hear more about the richer parts, that doesn't mean they're more common.

Yup. We don't use fibre optics on earth rather than lasers because of some specific limitation of the earth's surface being in orbit would avoid.

We use them because they're many orders of magnitude cheaper and simpler for anywhere near the same bandwidth for the distances required.

Yup - my example of the Sahara wasn't really a specific suggestion, so much as an example of "The Most Inconvenient Inhospitable part of the earth's surface is still much better than space for these use cases". This isn't star trek, the world doesn't match sci-fi.

It's like his "Mars Colony" junk - and people lap it up, keeping him in the news (in a not explicitly negative light - unlike some recent stories....)

Output from radiating heat scales with area it can dissipate from. Lots of small satellites have a much higher ratio than fewer larger satellites. Cooling 10k separate objects is orders of magnitude easier than 10 objects at 1000x the power use, even if the total power output is the same.

Distributing useful work over so many small objects is a very hard problem, and not even shown to be possible at useful scales for many of the things AI datacenters are doing today. And that's with direct cables - using wireless communication means even less bandwidth between nodes, more noise as the number of nodes grows, and significantly higher power use and complexity for the communication in the first place.

Building data centres in the middle of the sahara desert is still much better in pretty much every metric than in space, be it price, performance, maintainance, efficiency, ease of cooling, pollution/"trash" disposal etc. Even things like communication network connectivity would be easier, as at the amounts of money this constellation mesh would cost you could lay new fibre optic cables to build an entire new global network to anywhere on earth and have new trunk connections to every major hub.

There are advantages to being in space - normally around increased visibility for wireless signals, allowing great distances to be covered at (relatively) low bandwidth. But that comes at an extreme cost. Paying that cost for a use case that simply doesn't get much advantages from those benefits is nonsense.

And profiting from it, though less directly than "$ for illegal images". Even if it wasn't behind a paywall (which it mostly is) driving more traffic for more ads for more income is still profiting from illegal imagery.

Yup - I've related it to working with Juniors, often smart and have good understandings and "book knowledge" of many of the languages and tools involved, but you often have to step back and correct things regularly - normally around local details and project specifics. But then the "junior" you work with every day changes, so you have to start again from scratch.

I think there needs to be a sea change in the current LLM tech to make that no longer the case - either massively increased context sizes, so they can contain near a career worth of learning (without the tendency to start ignoring that context, as the larger end of the current still-way-too-small-for-this context windows available today), or even allow continuous training passes to allow direct integration of these "learnings" into the weights themselves - which might be theoretically possible today, but is many orders of magnitude higher in compute requirements than available today even if you ignore cost.

He does (or at least a good proportion) if you want to use as precedent for delivering on these promises, though. Especially for the larger more extreme statements and not just buying himself into an existing business.

Considering Microsoft's significant (and vocal) investment in LLMs, I fear the current state of Windows 11 is related to a team trying to do exactly that.

That guy's style guide seems to conflict with the Cambridge editorial services guidelines - though that is for books rather than papers:

> Spaced en rules (or ‘en dashes’) must be used for parenthetical dashes. Hyphens or em rules (‘em dashes’) will not be accepted for either UK or US style books. En rules (–) are longer than hyphens (-) but shorter than em rules (—).

Section 2.1, "Editorial services style guide for academic books" https://www.cambridge.org/authorhub/resources/publishing-gui...

The Oxford style guide page 18 https://www.ox.ac.uk/public-affairs/style-guide

> m-dash (—)

> Do not use; use an n-dash instead.

> n-dash (–)

> Use in a pair in place of round brackets or commas, surrounded by spaces.

Remember I'm specifically speaking about british english.

In British English you'd be wrong for using an em-dash in those places, with most grammar recommendations being for an en-dash, often with spaces.

It's be just as wrong as using an apostrophe instead of a comma.

Grammar is often wooly in a widely used language with no single centralised authority. Many of the "Hard Rules" some people thing are fundamental truths are often more local style guides, and often a lot more recent than some people seem to believe.

As a brit I'd say we tend to use "en-dashes", slightly shorter versions - so more similar to a hyphen and so often typed like that - with spaces either side.

I never saw em-dashes—the longer version with no space—outside of published books and now AI.

True, you can increase bit width to guarantee never hit those issues, but right now saturating arithmetic on types that pretty commonly hit those values is the standard. Guaranteeing it would be a significant performance drop and/or memory use increase with current techniques to the level it would significantly affect availability and cost compared to what people expect.

Similarly you could not allow re-ordering of operations and similar - so the results are guaranteed to be deterministic (even if still "not correct" compared to infinite precision arithmetic) - but that would also have a big performance cost.

Fixed point arithmetic isn't truly associative unless they have infinite precision. The second you hit a limit or saturate/clamp a value the result very much depends on order of operations.

I worked on an extremely niche project revolving around an old DOS game. Code I worked on is often pretty much the only reference for some things.

It's trivially easy to get claude to scrape that and regurgitate it under any requested licence (some variable names changes, but exactly the same structure - though it got one of the lookup tables wrong, which is one of the few things you could argue aren't copyrighted there).

It'll even cheerfully tell you it's fetching the repository while "thinking". And it's clearly already in the training data - you can get it to detail specifics even disallowing that.

If I referenced copywritten code we didn't have the license for (as is the case for copyleft licenses if you don't follow the restrictions) while employed as a software engineer I'd be fired pretty quick from any corporation. And rightfully so.

People seem to have a strange idea with AI that "copyleft" code is free game to unilaterally re-license. Try doing that with leaked Microsoft code - you're breaking copyright just as much there, but a lot of people seem to perceive it very differently - and not just because of risk of enforcement but in moralizing about it too.

I mean technically there's nothing they can do that SMM couldn't - introduced in a revision of the 386. It's code running with system permissions invisible to the "parent" user code and OS.

You're already pretty much trusting the same people then as now, at least if they are "actively malicious".

Yup, it's their right, as it's their right to crank their seat back. Both are available, and expected. I see some take issue with people doing them at all, but I don't really mind much. Might be related to being 180cm or so - about 5"11, when most complaining seem to be larger.

My issue on my original post isn't really them moving the seat, but the lack of notice. It would have taken barely a second to lean over the seat and let me know. But I suspect they didn't even think about how that might affect anyone else.

It would be kinda crazy if someone didn't say they wanted to get out to the aisle and just started trying to climb over you with no warning.

I think the issue is that people limited compute time as a proxy for difficulty.

In that case you'll hit issues on any device that performs significantly differently from that which it was tuned in.

Though I am slightly amused by people using the apple chip as an example of "high performance" in a problem that scales very well with threading.

That happened to me when I had an ipad in a standing case and the seat in front cranked back - trapping then pinging the tablet across me and by neighbour's lap.

Though the ipad itself wasn't damaged, a couple of glasses didn't make it, and required the steward to try to brush up whatever fragments of glass they could.

I feel that airlines are a microcosm of "Do you care about who you actions might affect?" - similar to the "Do you return the cart to the corral" test at supermarkets - are you willing to put even the smallest bit of effort to significantly improve other people's experiences?

I feel to be consistent the output of that model will also be under that same open license.

I can see this being extremely limiting in training data, as only "compatible" licensed data would be possible to package together to train each model.

The psychology behind "creating it for the sake of creating it" can also be significantly changed by seeing someone then take it and monetize it without so much as a "thank you".

It's come up quite often even before AI when people released things under significantly looser licenses than they really intended and imagined them being used.

Let be honest, you'd only bother trawling if you already had significant suspicion.

The output of AIs that is "churned out" wouldn't exist without templates like this being used as an input to the training. But that isn't "Copyright Infringement", according to the AI companies.

Add in presumably ignoring the license of the scraped documents.

Zen1 was further behind in ST perf than Intel is today in it's desktop offerings. They really exploited their strength in MT and price, and showed that the market was already chafing under Intel's reluctance to go beyond 4 cores on their consumer line, presumably to avoid stepping on the toes of HEDT. But that just caused the competition to pretty much invalidate that entire line instead.

And I don't really see the situation being that obviously different if it was Nvidia who they merged with and Jensen was CEO.

The big issue was simply that AMD didn't have the cash at hand to both pay for ATI and maintain investment in R&D, at least without their next few products completely dominating the competition. I don't see a different CEO changing that. Unless Jensen was willing to value Nvidia significantly less than ATI at the time.

I kinda agree. We've been told for years it's a "massive productivity multiplier", and not just an iterative improvement.

So you expect to see the results of that. The AAA games being released faster, of higher quality, and at a lower cost to develop. You expect Microsoft (one of the major investors and proponents) to be releasing higher quality updates. You expect new AI-developed competitors for entrenched high-value software products.

If all that was true, it doesn't matter what people do or don't argue on the internet, it doesn't matter if people whine, you don't need to proselytize LLMs on the internet, in that world people not using is just an advantage to your own relative productivity in the market.

Surely by now the results will be visible anyway.

So where are they?

Not in RE, maybe, but directly viewing source isn't RE either.

It seems they're pretty directly admitting to referring to the LGPL library while implementing theirs under a different license.

I wonder if they'll have no issues with people directly reading their code while happening to implement the same functionality with a closed license? Or a GPL-style one?

I'm surprised they admitted to it - it's hardly "Clean Room"....

You could argue that microcode emulation is what they do now.