FOSS came up around the core idea of liberating software for hardware, and later on was sustained by the idea of a commodity of commons we can build on. But with LLMs we have alternative pathways/enablement for the freedoms:
Freedom 0 (Run): LLMs troubleshoot environments and guide installations, making software executable for anyone.
Freedom 1 (Study/Change): make modifications, including lowering bar of technical knowledge.
Freedom 2 (Redistribute): LLMs force redistribution by building specs and reimplementing if needed.
Freedom 3 (Improve/Distribute): Everyone gets the improvement they want.
As we can see LLM makes these freedoms more democratic, beyond pure technical capability.
For those that cared only about these 4 freedoms, LLMs enable these in spades. But those who looked additionally for business, signalling and community values of free software (I include myself in this), these were not guaranteed by FOSS, and we find ourselves figuring out how to make up for these losses.
Free software apply for software made for humans, by definitions any LLM code can only have the license inherited from the previous work as it's a transformation work. In the end LLM's are just a source of legal troubles and propietary software companies will end suing each other while every pre AI libre software will be 100% legal to use.
I have to disagree. LLMs have shown that the only way to participate in the new software ecosystem are through leveraging an extremely powerful position that is create, backed, and maintained through the exploitation of capital, labor, and power (political, legal, corpotate) at levels never really seen before. The model of the Cathedral and the Bazaar was not broken by LLMs but instead the entire ecosystem was changed.
Now the software doesn't matter. The code doesn't matter. The hardware doesn't matter. Anyone can generate anything for anything, as long as they pay the fee. I think it can likely be argued that participation is now gated more than ever and will require usage of an LLM to keep up and maintain some kind of competition or even meager parity. Open weight models are not really a means of crossing the moat; none of the open weight models come close to the functionality, and all of them come from the same types of corporations that are releasing their models for unspecified reasons. The fact remains that the moat created by LLMs for open source software has never been larger.
IMO we need more sovereign systems like this (this is too simple IMO). Other sovereign systems are complex to deploy. if good FOSS commodity options come up, then we can expect a hosting/deployment infra and companies to setup and offer it for non self-hosters as well - ala WordPress.
That would simplify things, but in my opinion that is still too high a hurdle. I'm all about privacy and FOSS, but I don't self-host anything (except for my personal website).
I wish that more apps would instead move the logic to the client and use files on file syncing services as databases. Taking tasks as an example, if a task board were just a file, I could share it with you on Dropbox / Drive / whatever we both use, and we wouldn't need a dedicated backend at all.
The approach has limitations (conflict resolution, authorization, and latency are the big ones), but it is feasible and actually completely fine for lots of apps.
To the folks and Kitten team: I'm working on TTS as a problem statement (for an application), and what is the best model at the latency/cost inference. I'm currently settling for gemini TTS, which allows for a lot of expressiveness, but a word at 150ms starts to hurt when the content is a few sentences.
my current best approach is wrapping around gemini-flash native, and the model speaking the text i send it, which allows me end to end latency under a second.
are there other models at this or better pricing i can be looking at.
Claude models have made very good progress (see BS benchmark), and that probably explains why they're leading now. others will follow this precedent shortly, no doubt.
It's orthogonal IMO. YOLO or not is simply a sign of trust for the harness or not. Trust slightly affects cognition, but not much. My working hypothesis: exhaustion is the residue of use of cognition.
What impacts cognition for me, and IMO for a lot of folks, is how well we end up defining our outcomes. Agents are tremendous at working towards the outcome (hence by TDD red-green works wonderfully), but if you point them to a goal slightly off, then you'll have to do the work of getting them on track, demanding cognition.
So the better you're at your initial research/plan phase, where you document all of your direction and constraints, the lesser effort is needed in the review.
The other thing impacting cognition is how many parallel threads you're running. I have defaulted to major/minor system - at any time I have 1 major project (higher cognition) and 1 minor agent (lower cognition) going. It's where managing this is comfortable.
Has anyone implemented a system of Pi for a team? Basically consolidate all shared knowledge and skills, and work on things that the team together is working on through this?
Basically a pi with SSO frontend, and data separation.
If no one has - I have a good mind to go after this over a weekend.
And others pull regularly from the pool? how are knowledge and skills continuously updated? I was thinking these necessarily need to be server side (like the main project under discussion) for it to be non-clunky for many users, but potentially git could work?
Like, let's take a company example - gitlab. If an agent had the whole gitlab handbook, then it'll be very useful to just ask the agent what and how to do in a situation. The modern pi agents can help build such a handbook with data fed in all across the company.
>If you ask me, no court should have ever rendered a judgement on whether AI output as a category is legal or copyrightable, because none of it is sourced. The judgement simply cannot be made, and AI output should be treated like a forgery unless and until proven otherwise.
Guilty until proven innocent will satisfy the author's LLM-specific point of contention, but it is hardly a good principle.
You are missing the point of the author. He literally said no court should have rendered a judgement, that's the exact opposite of guilty until proven innocent. Guilty means a court has made a judgement.
He is proposing to not make a judgement at all. If the AI company CLAIMS something they have to prove it. Like they do in science or something. Any claim is treated as such, a claim. The trick is to not even claim anything, let the users all on their own come to the conclusion that it's magic. And it's true that LLMs by design cannot cite sources. Thus they cannot by design tell you if they made something up with disregard to it making sense or working, if they just copy and pasted it, something that either works or is crap, or if they somehow created something new that is fantastic.
All we ever see are the success stories. The success after the n-th try and tweaking of the prompt and the process of handling your agents the right way. The hidden cost is out there, barely hidden.
This ambiguity is benefitting the AI companies and they are exploiting it to the maximum. Going even as far as illegally obtaining pirated intellectual property from an entity that is banned in many countries on one end of their utilization pipeline and selling it as the biggest thing ever at the other end. And yes, all the doomsday stories of AI taking over the world are part of the marketing hype.
A forgery isn't a subjective assessment. A forgery is intentionally made inaccurate claim of the origin of something. If the by-line is claiming it was made by someone who didn't make it, it's a forgery no matter how good of a copy it is judged to be.
This is precedent setting. In this case the rewrite was in same language, but if there's a python GPL project, and it's tests (spec) were used to rewrite specs in rust, and then an implementation in rust, can the second project be legally MIT, or any other?
If yes, this in a sense allows a path around GPL requirements. Linux's MIT version would be out in the next 1-2 years.
Its very important to understand the "how" it was done. The GPL hands the "compile" step, and the result is still GPL. The clean Room process uses 2 teams, separated by a specification. So you would have to
1. Generate specification on what the system does.
2. Pass to another "clean" system
3. Second clean system implements based just on the specification, without any information on the original.
That 3rd step is the hardest, especially for well known projects.
So what if a frontier model company trains two models, one including 50% of the world's open source project and the second model the other 50% (or ten models with 90-10)?
Then the model that is familiar with the code can write specs. The model that does not have knowledge of the project can implement them.
Would that be a proper clean room implementation?
Seems like a pretty evil, profitable product "rewrite any code base with an inconvenient license to your proprietary version, legally".
3. claude-code that converts this to tests in the target language, and implements the app that passes the tests.
3 is no longer hard - look at all the reimplementations from ccc, to rewrites popping up. They all have a well defined test suite as common theme. So much so that tldraw author raised a (joke) issue to remove tests from the project.
> but if there's a python GPL project, and it's tests (spec) were used to rewrite specs in rust, and then an implementation in rust, can the second project be legally MIT, or any other?
Isn't that what https://github.com/uutils/coreutils is? GNU coreutils spec and test suite, used to produce a rust MIT implementation. (Granted, by humans AFAIK)
Treating an AI-assisted rewrite as a legal bypass for the GPL is wishful thinking. A defensible path is a documented clean-room reimplementation where a team that never saw the GPL source writes independent specs and tests, and a separate team implements from those specs using black-box characterization and differential testing while you document the chain of custody.
AI muddies the water because large models trained on public repos can reproduce GPL snippets verbatim, so prompting with tests that mirror the original risks contamination and a court could find substantial similarity. To reduce risk use black-box fuzzing and property-based tools, have humans review and scrub model outputs, run similarity scans, and budget for legal review before calling anything MIT.
I'm somewhat confused on how it actually muddies the waters - any person could have read the source code before hand and then either lied about it or forgot.
Our knowledge of what the person or the model actually contains regarding the original source is entirely incomplete when the entire premise requires there be full knowledge that nothing remains.
That why I carved it out to just the specs. If they can be read as "facts", then the new code is not derived but arrived at with TTD.
The thesis I propose is that tests are more akin to facts, or can be stated as facts, and facts are not copyright-able. That's what makes this case interesting.
FOSS came up around the core idea of liberating software for hardware, and later on was sustained by the idea of a commodity of commons we can build on. But with LLMs we have alternative pathways/enablement for the freedoms:
Freedom 0 (Run): LLMs troubleshoot environments and guide installations, making software executable for anyone.
Freedom 1 (Study/Change): make modifications, including lowering bar of technical knowledge.
Freedom 2 (Redistribute): LLMs force redistribution by building specs and reimplementing if needed.
Freedom 3 (Improve/Distribute): Everyone gets the improvement they want.
As we can see LLM makes these freedoms more democratic, beyond pure technical capability.
For those that cared only about these 4 freedoms, LLMs enable these in spades. But those who looked additionally for business, signalling and community values of free software (I include myself in this), these were not guaranteed by FOSS, and we find ourselves figuring out how to make up for these losses.
reply