What exactly is the difference between "a machine-readable contract for what the output has to be" and "source code"?
What is the difference between an "agent" and a "compiler"?
For that matter, what is the difference between "I got an agent to provide a high level description" and a decompiler?
What is the difference between ["decompiling" a binary, editing the resulting source, recompiling, and redistributing] and [analyzing the behavior of a binary, feeding that description into an LLM, generating source code that replicates that behavior, editing that, recompiling and redistributing]?
Takeaway: we are now in a world where software tools can climb up and down the abstraction stack willy nilly and independently of human effort. Legal tools that attempt to track the "provenance" of "source code" were already shaky but are now crumbling entirely.
That's funny, because the "network effect" of the GPL was always extremely obvious to me as part of the point. The idea being that the more that GPL software forms a cohesive ecosystem, the less financially viable it becomes to operate outside of that ecosystem. It's viral, right? And software builds on software, we're about 15 layers deep at this point. The hope is that eventually, so much software is GPL that you effectively have two choices when writing something new: 1) join the GPL borg, or 2) boil the ocean reimplementing the entire stack from scratch.
Unfortunately, free software was successful enough that the latest generation takes it for granted, and has forgotten why radical software politics is necessary. They do not understand, if they even think to ask, why so many nerds ran GNU/Linux even when it was objectively kind of terrible - why so many people were motivated to pour time into a half-broken thing. I hope by the time they do understand, it will not be too late.
My controversial opinion: the problem with the GPL is that it isn't viral enough. It was written by nerds with a good understanding of computers, and poor understanding of people. As such, it focuses too much on technicalities like what it means "link" programs together, in an attempt to rigorously specify definitions that permit running proprietary programs on free operating systems, and vice versa.
But it doesn't work. Every time your Android phone downloads a firmware update, by rights that should be a GPL violation, as it's a single giant executable that mixes together GPL and proprietary code, deliberately made in such a way that separating them out after the fact is impossible - in fact, the program is explicitly designed to fail to run if you so much as tamper with a single bit (signed images). It is hard to imagine something further from Stallman's vision - hard to imagine something less respectful of user freedom. And yet this is permitted on technicalities, because this functionally unmodifiable binary blob happens to be structured in a particular way that computer nerds recognize as a type of database called a "filesystem", and the GPL parts are neatly organized into database entries called "files". And they all agree that that's okay, whereas if you mix the code in a different type of database called a "link table", well that's bad and wrong.
Which "individual freedom" do you feel the GPL denies you? As far as I can tell, it only prevents you from piggybacking on other people's work, and adding unfair stipulations to the resulting product. It is a very symmetrical, "do-unto-others" type license.
In the scenario you’re describing, when I write my own code, I am limited in what license I can pick for that code because of licensing choices other people made.
But you're not actually restricted from doing anything, are you? What is it exactly you want to do that other people's choice of GPL prevents? Steal their work and sell it? Oh how unfair!
We are talking about a hypothetical universe in which nearly all software is GPL, such that it is almost impossible to write useful software without building upon other GPL code. In such a universe, licensing "your" code as MIT would indeed be unfair, because you would be taking the work of others, illegally stripping the label, and making it available to profitable interests to use without compensation to the original developers against their express wishes - said compensation merely being the extremely reasonable request to share back, as you were shared to.
You still haven't really explained why you're so keen on doing that sort of thing.
> licensing "your" code as MIT would indeed be unfair, because you would be taking the work of others, illegally stripping the label, and making it available to profitable interests to use without compensation to the original developers against their express wishes
I'm not sure why there are quotes around "your".
If I write code and license it MIT, but it includes code that has a different non-GPL license (lets say Apache), my code is MIT-licensed, and the included code is still Apache-licensed.
I haven't illegally (or legally) stripped any licenses, or changed how it's available to others. I've picked a license for code I wrote, and the developers of code I took a dependency on picked a license for their code. People who want to use my code have to consider the license of my code and also the dependencies I used.
The GPL is largely unique in its desire to control what license I can pick for my own code.
I'm keen on picking my own license for my own code because I personally don't want to block my code from being used by anybody, commercially or otherwise. I've got no issue with developers who do want to prevent closed-source, commercial, or any other kind of downstream usage. And I'm happy to comply with the licenses of code that I leverage as part of my code. I do take issue with developers who want to impose their licensing preferences on my code.
Libraries or executables, it makes no difference. You are incentivized to use GPL because you wish to build on top of work that is GPL.
Obviously in large part that didn't happen, because of a cultural tendency to use more permissive licence variants (such as AGPL) for libraries, in the pragmatic hope that this would encourage their use even in proprietary programs, and therefore incentivize back-contributions from a wider audience. But this indeed halts the "virality" of the GPL, and so one is once again forced to conclude - incredibly - that Stallman was not radical enough...
Couldn't make it work with a friend sitting next to me. The QR codes led to the website but didn't add a contact. Manually adding the ID resulted in the number abruptly disappearing with no feedback and no effect.
This shouldn't be surprising. They are ultimately trained on human generated text! Or on text that was generated by something trained on human-generated text, or some even deeper recursion. In the end, all "intelligence" is an emergent consequence of emulating humans. The less human-like you make them, the further they get from the "source" of their intelligence. It wouldn't be problem if we knew how to teach programs "how to think" - but we don't!! That is why, in 2026, we train language transformers on huge corpuses instead of symbolically programming expert systems in Lisp.
Something I'm kind of surprised by is the lack of interest in bootstrapping language models into something like a "person". Not a butler, assistant, programming tool, doctor, therapist, sycophant, whatever - a convincingly independent person with thoughts and feelings, moods, flaws and all. Maybe there isn't economic demand for it.
I'm not following - are you implying that handing a contact card to someone is a sexual pass? Or is it only considered sexual when the recipient is underage?
On the basis of what I've seen so far I find it unlikely that Epstein actually cared much about establishing a "behavioral engineering institute". Baiting "thought leaders" by emailing them about big ideas and grand plans was his standard M/O, it seems. He must have sent hundreds if not thousands of such emails - Scott Aaronson got one[0], and with no disrespect intended to him he must be pretty far down the list of powerful people you'd want to collect.
Almost everyone voting Brexit was uneducated on the issues, a fact rapidly borne out by actually speaking to them. Actually "uneducated" is quite charitable as quite a lot are racist also. I don't know how anyone can claim otherwise with a straight face.
Which issues were they uneducated on? How many did you speak to, and I'm not talking about the engagement-bait bots on places like the Trending side of Twitter
Is it possible that software is not like anything else, that it is meant to be discarded: that the whole point is to always see it as a soap bubble?
--Alan Perlis
This is an excellent observation that probabilistic phrasing is inherently contextual, but I don't know if I'd agree that something being nuanced and context dependent inherently means "alignment is impossible". Certainly it implies that the design of the study was simplistic.
What is the difference between an "agent" and a "compiler"?
For that matter, what is the difference between "I got an agent to provide a high level description" and a decompiler?
What is the difference between ["decompiling" a binary, editing the resulting source, recompiling, and redistributing] and [analyzing the behavior of a binary, feeding that description into an LLM, generating source code that replicates that behavior, editing that, recompiling and redistributing]?
Takeaway: we are now in a world where software tools can climb up and down the abstraction stack willy nilly and independently of human effort. Legal tools that attempt to track the "provenance" of "source code" were already shaky but are now crumbling entirely.
reply