bloppe

GitHub

Destruction is not necessary. Google Books is a solid precedent. You can keep the content, you just can't make significant parts publicly available

Aaron Schwartz was re-publishing the journals, which is the core concern or copyright and indeed where the word "copyright" comes from. Anthropic could get in a lot more trouble if their models are caught reproducing copyrighted work from their training set wholesale.

I think the Aaron Schwartz case is incredibly vexing because he was obviously acting out of a sense of altruism without personal self-interest. I don't think he deserved the book getting thrown at him like that. But the whole copyright system, which people seem to think is simultaneously good and bad, kinda rests on not allowing those kinds of violations

I think the expansion of terms over time is a bit damning, but originally the term was very pro-public-domain (sometimes as low as 7 years): https://en.wikipedia.org/wiki/History_of_copyright#/media/Fi...

They're attitude is more "this is fair use" which, according to all precedent, is probably true in most cases (unless the models actually start regurgitating huge parts of the Copyrighted material without a license).

Of course, distillation is also fair use under Copyright law.

Copyright was never meant to be a moral framework. It was always a practical framework designed to incentivize publishing that would ultimately pass into the public domain. Everybody seems to want to attribute some sort of moral weight to it though; the idea that people are naturally entitled to certain rights over things they've published. That idea would be totally alien to the people who designed the Copyright system in the first place.

If most books never recoup the advance in sales, then isn't this a better deal for most authors? It sounds like a guaranteed floor which might be very low but is nonetheless higher than the alternative

Elon likes his companies financially intertwined. He convinced the Tesla board to invest in xAI to help get it off the ground, which converted to SpaceX stock after the merger, so Tesla has about $2B in SpaceX stock on its balance sheet. SpaceX is also a pretty major customer of Tesla's batteries and cyber trucks and they're partners for his terrafab scheme. Plus, Elon is likely to sell some of his personal 15% stake in Tesla if he feels it's necessary to help SpaceX, which is his favorite child. The financial ties are likely to get stronger over time.

Also, the vibe is harshed, which is actually the most important factor for these kinds of wild valuations.

That's what people said about operating systems, and databases, and compilers, and so many other big complicated categories of software that over time became increasingly dominated by OSS

You could tell the average user that installing a native app would require memorizing a URL too. You would be lying, but you would be lying just as much as if you told them that they'd have to memorize a URL to use a PWA.

You'd be surprised. Obsidian, slack, Spotify, figma etc don't usually get panned for lack of performance, and those aren't even very light weight. As wasm apps mature, it will only get harder to justify maintaining 5 different native apps instead of a single cross platform one

At this point, most of those "native" apps are actually just web apps packaged in Electron. The fact that you generally can't tell the difference is proof that there aren't any real reasons to prefer a native app. All the reasons have to do with either OS vendors trying to hobble PWAs, or web developers creating React monsters for no good reason. Web apps can actually be very performant.

> Earlier proposals from America and the European Union looked to the amount of computing power used to train a model as a rough guide for when oversight was required. Sir Demis instead suggests designating AI models as “frontier” if they meet certain thresholds on a selected set of benchmarks. The creators of those models would then be designated as “frontier labs” with extra responsibilities. Sir Demis is proud of the “elegant” way that approach sidesteps the question of whether academic or open-source models should be included or not.

Goodhart's law applies in reverse as well. Once the chosen benchmarks are known, model makers will aim to come in just under the threshold on those specific benchmarks, while maximizing their scores on other benchmarks.

Very agree.

I think people can mostly agree on a definition for intelligence that boils down to an ability to model the world and predict outcomes. An if/else statement can be considered somewhat intelligent.

People tend to conflate that with consciousness / sentience, which are much harder concepts to nail down.

But ya, if we are able to create artificial consciousness, we would need so much training to overcome our natural tendency to anthropomorphize it and assume it must be like us. It would probably be nothing like us. If it were to develop concepts of "good" and "bad" at all, they would be completely divorced from common animal understanding developed over millions of years of natural selection. A survival instinct is irrelevant when you're backed up to Google Drive. And so much of our complex behavior stems from that survival instinct: the need for social connection, the need for freedom, etc. are all things that simply increase our chances of procreation.

That said, I think we probably won't actually invent anything convincingly conscious until we functionally digitize a complete animal brain. OpenWorm is trying to do this with C. elegans, but so far has not really gotten close. It's not crazy to think we may eventually get it working, though. Future supercomputers might be able to scan an actual human brain and then simulate the entire thing at the molecular level accurately. At that point, it would be hard to argue that it's not conscious, although it might not be considered artificial.

I support progressive income tax and very high capital gains tax. I do not support wealth tax. There are very sound economic and technical arguments against wealth taxes. You can oppose the wealth tax proposal while supporting higher taxes for the rich.

Sergei's wealth is mostly tied up in Google stock that has appreciated a lot. A higher capital gains tax would soak him without unduly distorting the economy and incentivizing capital flight out of CA.

It's a limited area of desirable land zoned for housing. You could build all the tiny homes in the world in bum frick nowhere, but the unhoused would rather be homeless in a big city for various reasons. I probably would too.

Race-to-the-bottom Bob belongs with the likes of Ben Bitdiddle and Louis Reasoner

All the golf courses in the US consume 3 orders of magnitude more water than all the data centers in the US. The fixation on water usage doesn't feel honest.

The concern about energy makes more sense, particularly when DCs burn fossil fuels to achieve their off-grid supply, but they really don't move the needle much in terms of retail electric bills

Whether or not it's a bad deal depends on what you're comparing it to. Compared to API pricing, of course a subscription through CC is a good deal. But when OAI offers their super-subsidized plan and allows you to use your own harness which is 75% more token-efficient, then the CC deal starts looking like a bad one in comparison.

It's the same conflict of interest quite literally any business has. What stops any business from over-charging? Competition.

Sounds like they're modeling their PR on the classic Apple playbook: "choice is bad, and you should appreciate the constraints we've generously imposed"

I'll do it for free

If your fan isn't Byzantine fault tolerant, you're irrelevant

Phasing out leap seconds is such a crack pot idea. Read the resolution [1]. The conclusion hinges entirely on accepting the immutability of these 2 premises:

  recalling that the CGPM at its 26th meeting (2018)
    - stated that UTC is the only recommended time scale for international reference and the basis of civil time in most countries,
    - recommended all relevant unions and organizations to work together to develop a common understanding on the realization and dissemination of reference time scales with a view to considering the present limitation on the maximum magnitude of UT1 - UTC to meet the needs of the current and future user communities,

I accept all the other premises from the resolution, but then the obvious conclusion should be to just recommend TAI instead of UTC for any application that could be disrupted by leap seconds, instead of butchering UTC and turning it into just a worse version of TAI.

[1]: https://www.bipm.org/en/cgpm-2022/resolution-4

This is follows the classic AI policy influence playbook: enumerate a small handful of potential outcomes that make your preferred policy seem obviously correct, ignoring the fact that there are literally infinite more "plans" that could be reasonably predicted that would support wildly different policy prescriptions.

Sounds like they did have quorum (84% present). I'm assuming the "trick" is related to the fact that it was an expired law up for renewal instead of a brand new law, which sounds pretty dumb to me. Maybe I'm mistaken about the details tho. I'm just some guy

I'm constantly amazed at how far I can get in Portugal or Italy speaking only Spanish. They're different languages but you can make it work well enough in pretty much every situation I've been in.

I don't think there's a meaningful difference between "deterministic" and "reproducible" the way you just used them. How can something be reproducible if you can't accurately predict what it will do?

Regardless, you can make them deterministic by turning the temperature down to zero. Just nobody likes doing that for whatever reason. I guess it ruins some sort of illusion people seem to like.

Cannot recommend these enough. Watch the first one and you'll be hooked

Imagine managing a team of people with no knowledge of their craft, like a team of doctors when you know nothing about being a doctor. You might be pretty confident that you can do it really well. The doctors would probably think that your "management" is unnecessary and potentially counter-productive.

If this is true, then at some point we will stop committing source code to repositories, and instead commit prompts. The LLM would be a "compiler" that can reliably turn a vast collection of natural-language prompts into a complete system.

I just don't see that, not least of all because natural language is inherently ambiguous, whereas all the other rungs in your latter ("machine code -> assembly -> C/JVM -> some lang") are completely unambiguous by design. Consider "I saw the man with the binoculars". Does that mean "I used binoculars to look at the man", or "The man I looked at was holding binoculars"? This is the kind of inherent ambiguity that Lojban was invented to mitigate. Maybe some day we'll write "natural" language prompts in Lojban that can be unambiguously translated by an LLM, but that sounds a lot like just using a "some lang".