the most generous reading of your arguments is that you are philosophically defeatist

That’s probably a fair assessment.

But I feel like the core of my argument remains: I’m not disputing that MS or Google or Amazon or Apple services are sold to people and orgs who use them to commit evil. Of course they are.

But these aren’t munitions. They are general-purpose computing products being turned to evil outcomes by bad actors. The article, for example, cites Microsoft’s open-source LAVENDER, which is a general purpose image and video analysis tool for AI. Describing it as:

‘Lavender’, an AI-powered system designed to identify bombing targets

This simply isn’t true. Somebody in the Israeli military used LAVENDER to process video data to identify bombing targets, like somebody might use a hammer to smash someone’s head in. The articles you cite are full of rhetorical tricks to imply that Microsoft corporate had some hand in the decision making, but it’s genuinely all “well the Israeli military has some Azure servers, therefore Microsoft killed people”.

Which militaries should Microsoft (or Google or Apple or Amazon, etc) be allowed to sell products to? Who makes that determination? A cohort of employees or consumers? NGOs?

If government makes the call – distilling a public consensus on the matter, one hopes – then I can see some reasonable way to approach this question.

EDIT: Details on LAVENDER:

https://www.microsoft.com/en-us/research/publication/lavender-unifying-video-language-understanding-as-masked-language-modeling/

I just don’t see it doing any good. Why would Israel’s military, supplied with US military hardware, care about Microsoft? Or Apple or Google or Amazon or… I’m sure none of their critical military infrastructure is in danger if one or several of these companies turn on them.

And how does Microsoft even enforce this ban? Turn off Windows remotely? It’s not even clear how such a ban on Israel-linked business would work.

If world governments want to put sanctions on Israel and Gaza to try and make the two governments come to the table, I think that’s a much better strategy.

Honestly, I struggle to draw a connection between world conflict and non-military technology like Windows or cell phones or whatever.

Is every single Israeli resident complicit in what their government is doing? None of them should be allowed to use Windows? What about Israelis outside of Israel? What about people who support Israel? What about (gasp) Jews? How do you even enforce any of this without massive overreach by the companies?

Call on Microsoft or Apple all you want, ultimately I don’t think a company should ban sales to customers on the argument that those customers might not have morals aligned to the company. Not that it’s even possible, with world supply chains being what they are.

Capitalism does an extremely poor job of planning beyond the next accounting period.

With respect to the article, it’s wrong. AI help desk is already a thing. Yes, it’s terrible, but human help desk was already terrible. Businesses are ABSOLUTELY cutting out tier 1 call center positions.

LLMs are exceptionally good at language translation, which should be no surprise as that kind of statistical chaining is right up their alley. Translators are losing jobs. AI Contract analysis & legal blacklining are going to put a lot of junior employees and paralegals out of business.

I am very much an AI skeptic, but I also recognize that people who do the things LLMs are already pretty good at are in real trouble. As AI tools get better at more stuff, that target list of jobs will grow.

Took them 30 seconds to throw animators under the bus to make their point.

It’s hopeless. We’re all just gonna eat each other so the billionaire class can go live in a giant space station.

It was the style at the time.

Perhaps, but I don’t read anything on Substack unless I’m subscribed. Reputation is the entire point on Substack, without it, the content will get no traffic.

AI with dedicated nuclear power? I can’t imagine anything that could possiblye go wrong in this scenario.

Oh good, now when I search I’ll have to wade through the effluent of AI-produced pablum to find an actual human journalism product.

Remember when Substack, the home of many excellent journalists, started to defend fascist and white supremacist content on their platform?

Oh, wait, that’s happening right now.

It’s not “inexplicable”.

DIMM mounting brackets introduce significant limitations to maximum bandwidth. SOC RAM offers huge benefits in bandwidth improvement and latency reduction. Memory bandwidth on the M2 Max is 400GB/second, compared to a max of 64GB/sec for DDR5 DIMMs.

It may not be optimizing for the compute problem that you have, and that’s fine. But it’s definitely optimizing for compute problems that Apple believes to be high priority for its customers.

Before reading the article, I just assumed that N. Korea had hacked a game with loot boxes.

And it won’t need to exist locally on the phone anyway. Higher bandwidth cell and wifi signals mean more and more exotic AI processing can be offloaded onto cloud resources.

It’s great when you have an app that works well when not connected to a network, of course. But most phone buyers don’t really care.

Do people still use emacs to code, for example?

Umm. Yes.

RiffTrax Friends, The Gizmoplex, and Twitch Turbo give me everything I need.

it’s basically impossible to tell where parts of the model came from

AIs are deterministic.

1. Train the AI on data without the copyrighted work.

2. Train the same AI on data with the copyrighted work.

3. Ask the two instances the same question.

4. The difference is the contribution of the copyrighted work.

There may be larger questions of precisely how an AI produces one answer when trained with a copyrighted work, and another answer when not trained with the copyrighted work. But we know why the answers are different, and we can show precisely what contribution the copyrighted work makes to the response to any prompt, just by running the AI twice.

It’s literally not possible to be exposed to the history of art and not have everything you output be derivative in some manner.

I respectfully disagree. You may learn methods from prior art, but there are plenty of ways to insure that content is generated only from new information. If you mean to argue that a rendering of landscape that a human is actually looking at is meaningfully derivative of someone else’s art, then I think you need to make a more compelling argument than “it just is”.

There is literally not one single piece of art that is not derived from prior art in the past thousand years.

This is false. Somebody who looks at a landscape, for example, and renders that scene in visual media is not deriving anything important from prior art. Taking a video of a cat is an original creation. This kind of creation happens every day.

Their output may seem similar to prior art, perhaps their methods were developed previously. But the inputs are original and clean. They’re not using some existing art as the sole inputs.

AI only uses existing art as sole inputs. This is a crucial distinction. I would have no problem at all with AI that worked exclusively from verified public domain/copyright not enforced and original inputs, although I don’t know if I’d consider the outputs themselves to be copyrightable (as that is a right attached to a human author).

Straight up copying someone else’s work directly

And that’s what the training set is. Verbatim copies, often including copyrighted works.

That’s ultimately the question that we’re faced with. If there is no useful output without the copyrighted inputs, how can the output be non-infringing? Copyright defines transformative work as the product of human creativity, so we have to make some decisions about AI.

This issue is easily resolved. Create the AI that produces useful output without using copyrighted works, and we don’t have a problem.

If you take the copyrighted work out of the input training set, and the algorithm can no longer produce the output, then I’m confident saying that the output was derived from the inputs.