This hasn't yet been decided, as I pointed out above. The courts have indicated that push button LLM output is not copyrightable, but haven't said anything about larger orchestrations of multiple agents operating over hours or days to produce something with direction from a human.
Given that photographs are copyrightable, I wouldn't hold my breath on projects like this being deemed ineligible for copyright.
> In the copyright law in the United States, such copyright may exist when the materials in the compilation (or "collective work") are selected, coordinated, or arranged creatively such that a new work is produced. Copyright does not exist when content is compiled without creativity, such as in the production of a telephone directory. In the case of compilation copyright, the compiler does not receive copyright in the underlying material, but only in the selection, coordination, or arrangement of that material.
All my experience seeing orgs ship real production code with the assistance of AI over months/many-thousands-of-commits shows that the skill not only lies in what context you choose to provide in the first place (product and technical specifications, design guidelines, prompting process with the model(s)), but also in how to effectively review and test all of it, which is still done with human direction throughout. This is why people with no experience in coding can't do with AI what an industry veteran with 20 years shipping to production can.
In any case, I think existing reading of copyright law will be sufficient to say that large AI-driven software projects are a process of orchestration and also of curation in selecting what actually gets committed (or, if commit-by-default is enabled, then what to revert). This may not always be true in the future, though, as capabilities increase...one part of me thinks you could just do random stuff like "Make me an excel except every cell is a powerpoint presentation" and a good enough AI would reason through and produce something without further input. But I think in most cases, folks have a real set of requirements in mind, and get all of them into the AI through a process of iteration in various ways. I think that's real work that requires skill and, in many cases, creativity, at least in the sense that someone else walking up to the same tool and prompting on their own would produce a very different output than you'd see if you asked the same person to retry the same project.
I don't think we disagree that much. But the repo in question had basically no human input, at least no substantial one. It will be interesting where the courts will decide the boundary lies though.
I'd assume a company like that would have insurance for this. They'd be able to reimburse the customer knowing they'd get paid back from an insurance claim. That is, if they are actually insured for these kinds of losses.
A few years ago, the Canon camera battery supply chain was intermingled with knock offs. The seller I bought from was exchanging for legit batteries or reimbursement when a customer made a claim.
Then they have to be insured. Or the limit of their liability must be capped at the cost of the product or something. Otherwise they would have to reverse engineer every product they carry and just about every retailer would run the risk of being out of business any time someone pulls something like this. That’s obviously not happening.
From what I've read the seller's responsibility is just to fix or replace the product. Damages caused by the defect are the manufacturer's responsibility.
This implies that you know which question to ask. If people don't know about accessibility, security etc. How are they able to learn? Also, most people don't use AI to learn but to do it for them.
That's a good point, but if you have an always-on agent focused solely on helping you improve your knowledge/skills, it can ask you questions every day and gradually identify gaps in your knowledge and it gets really good at knowing where to push you.
That's where books come in. Read them, explain what you're learning to the AI, and what interests you about it, ask for it to check your understanding. Explain your other interests, ask it to suggest other things to read which will bridge the gaps between what you're learning now and the other things you already know.
I mean, the frontier models are pretty damn good at giving you all the standard conventions on things... The first question I usually ask is "give me the all the modern day conventions for this use-case", and then I start the context building from there. This applies to a lot of things, but I mainly use it when I am coding.
With an open set of weights, fixed seeds and parameters, recorded inputs and feedback loops, the entire AI decision making process is auditable.
It also lends itself nicely for the ability to adjust the AI systems when deficiencies are found. You can tweak the AI systems to align itself with the desired output, and check for regressions, much like tests in a codebase.
Human decision-making lacks this thoroughness. I’ve scarcely come across rigorous documentation of decisions and how they were reached professionally.
I have plenty of fire alarms, both local and remote, and extinguishers of different types around, I'm not too worried :) But thanks for the concern kind stranger!
The other thing to do is discharge it, so it has little energy stored to maintain/enlarge a fire
A drone engineer I worked with would immediately put any suspect battery into a barrel of water for several days, then get them disposed of properly.
The risk really seems trivial, probably because of our extensive experience with ordinary alkaline and NiCad batteries with a very benign degradation pattern, but the risk with Li-Ion is no joke.
All your fire extinguishers will do you no good if it goes off while you are out of the house. Also, the fire extinguishers will only suppress the surrounding fire, and do nothing about the core Li-Ion fire itself.
It's trivial until they go, and then it is instant misery. I've done it on purpose just to see how bad it would be and even knowing it will happen it will still surprise you with how sudden, total and hot those things burn. You have < 2 seconds from first signs of ignition to all-out fire with the battery moving all over the place due to outgassing. It's a thing to behold.
How am I supposed to be able to observe it slowly swell more over time then? I think I might have overstated how big this battery is, or understated how small it is. It's fiiine :)
reply