The Global Foundries split was probably a way to get AMD out of the hyper-capital-intensive fab business. And without a tier-1 customer, Global had less reason to pursue smaller nodes.

Intel has that national-champion thing to keep it afloat. I can imagine there are defence contracts that will never go to a “TSMC Arizona Division” and they’ll pay whatever it takes to keep that going.

It’s easier just to price in the fee than having to shut down or retool a project.

The problem with attribution is the difficulty of 1000% accurate compliance.

If you grab 100 lines of code from a repository, or five paragraphs from a story, there’s probably a claim there. If you grab a single word, there’s probably not. But in the middle, there’s a paralysis of uncertainty-- is n lines similar enough to create liability? Can you remember where you saw what reliably? You end up with a bias towards “over-attribution” and it becomes difficult to pare it back. Does everything need a full Git-style commit history? Are we forever stuck keeping a credit on a project because it’s difficult to prove you’ve fully scrubbed their contributions?

Focus on how we pay artists (ideally lush grants) and forget about credit. Maybe establish a culture where it’s voluntary and acceptable-- that people feel that they’re allowed to cite their raw materials, and reuse doesn’t make the work lesser-- but don’t try to use the courts to force people to try to remember and track where they saw something when they just want to create, or it creates a hostile environment.

Hey, the broken clock’s right!

IP law always had a built-in scale pronlem. Without a registration-required copyright model, and probably some sort of mandatory licensing rate system, the sheer logistics of finding and arranging rights made a lot of business models inpractical. (For example, why aren’t modern bookstores just print-on-demand kiosks, or streaming services have All The Content? In large part because it would cost thousands to track down owners and negotiate terms for $1.87 in royalties multiplied by every item in the catalog.)

This was ignorable for a long time, or even a commercial advantage for firms with access to large, pre-negotiated catalogs. The AI boom created a surprise market of non-incumbents who need to get access to a lot of IP in a streamlined manner.

If we open the door for bulk IP clearance to grant the AI bubble a stro ger legal footing, it can also allow other, potentially more interesting business ideas to slip through.

But what data would it be?

Part of the “gobble all the data” perspective is that you need a broad corpus to be meaningfully useful. Not many people are going to give a $892 billion market cap when your model is a genius about a handful of narrow subjects that you could get deep volunteer support on.

OTOH maybe there’s probably a sane business in narrow siloed (cheap and efficient and more bounded expectations) AI products: the reinvention of the “expert system” with clear guardrails, the image generator that only does seaside background landscapes but can’t generate a cat to save its life, the LLM that’s a prettified version of a knowledgebase search and NOTHING MORE