frabcus

Given a German court talked about not having the prompt logs making it hard in a specific case to prove it had sufficient human input, maybe someone could do an open source hosting service where the Claude Code logs were fully uploaded with each commit, so that could be later proven? It would be cool anyway to have such a service, so people could learn how to do LLM coding better from the examples.

Germany copyright law feels like the relevant thing here - Codeberg is a non-profit based in Germany.

Best quick English language overview of status that I could find: https://www.twobirds.com/en/insights/2026/germany/when-can-a...

It looks like Codeberg want only copyrighted material in their service, so it is reliable in the future that e.g. licenses must be followed (e.g. GPL), and copyright doesn't suddenly get declared as being of the model owner, and it isn't a copy of something else.

That is a cautious reasonable position - in early days of LLM coding (3 years ago!) indemnity from model companies was a major issue globally because of the lack of clarity of the law around this. The US specifically has settled on it being (effectively?) public domain. But I don't think that is fully settled, and it certainly isn't settled in international copyright law.

The goal of the vague "mostly" in the Codeberg change is to ensure there is enough human input to the code they host, to be reasonably sure under German copyright law it is copyright of the person sharing it.

Umm, Fable only really came out 2 weeks ago, and GPT-5.6 Sol only 1 week ago.

Yes, Kimi K3 appears a touch below them both, but above all other models. So I'd say a few weeks behind, not months now...

The advent of the internet was collaborative and based on introducing shared protocols for a couple of decades. It deserved criticism when globalised capitalism got involved, and monopolies started forming, leading to rent-seeking, excessive centralisation, and enshittification.

The impact is that the internet has a fraction of the value to improve people's lives as it should have. It is a very poor free market, incredibly poor competition because of lack of standards and protocols and interoperability. People's minds are ground down by social media, search engines don't work well any more and so on.

So yes - every new technology deserves many criticisms, so they can be addressed, and as a society we can gain the benefits of that technology and minimise the disadvantages.

The printing press lead to copyright, public libraries, universal literacy... All things which are now widely celebrated. They took centuries to work out, and are all government and regulatory intervention to fix problems critics noticed and campaigned about.

AI is the same, only it is at risk of moving much faster and having a much large negative impact before society reacts.

So no, most of the criticism of LLMs are not wrong - they are correct, as are the people saying the technology of LLMs is useful to people and the economy.

Critics are friends of a new technology - without responding to every criticism in a significant way, AI will rapidly lead to a Butlerian jihad. If you like AI, you should love criticism of AI even more.

I'm trying sourcehut at the moment https://sourcehut.org/ and it seems really good - very simple and fast. And does seem to be free for hosting open source projects.

Anyone else used it and have thoughts on it?

Better and easier to understand and use UX.

This being a news site for hackers, I should point out you can use browser plugins to change the default feed. And (if you use an Android phone, but maybe there's a way with Safari?) you can run those plugins (at least on Firefox), so you can have the experience you want on mobile too!

I made one of these (called Instalamb for Instagram), but haven't maintained it recently as there wasn't much interest. There are plenty of others though.

I think my biggest disappointment with social media is not that capitalism made it harmful and addictive (that was inevitable), but that most people don't seem to care enough to even install an advert blocker, never mind something to make their feed cleaner. Despite having had a better experience before, and it being much easier to do than many things people do all the time in their daily lives.

I've been wondering this - I don't have an intuition for Anthropic's gaming around military applications, or how this stage could play out in terms of relationship to Government controlling AI.

Are there some Less Wrong posts or similar I should read that probably explain it?

Sounds good from an x-risk point of view then. Maybe that's their deliberate plan!

I only did undergraduate level in Maths, and to me there is a key aesthetic element which makes it created. The choice of axioms to use, the choice of with theorems are interesting.

Yes the "truth" (doesn't exist, see Gödels theorem) is discovered in a vast, wild landscape that Mathematicians explore.

But which areas are worth exploring is a critical question. Partly driven by application, partly aesthetic. It's a quest for simple things that are a bit surprising, or that were hard to make the statements so simple.

There have been some US cases about this, but it isn't generally settled internationally. "Fair use" is a US specific thing. Even in the US there are ongoing cases.

Paper about how weights are a derivative work of the training data: https://arxiv.org/abs/2407.13493

Currently in progress law suits about AI copyright: https://informationisbeautiful.net/visualizations/the-rise-o...

Long term, getting locked into proprietary software development tools is a bad idea. And these models are extremely proprietary. The ability of the US Government to cancel them at any time is one real recent example of one category of problem.

Back in the 1990s the good C++ compilers were proprietary, eventually GCC and LLVM caught up, and now dominate. The pattern repeats in software development, and there's no reason to believe it won't continue.

Yes, right now it makes sense to use Opus 4.8, but it is good that a significant number of people are using other options, and making sure they work and are ready for when you need them.

Plus it is extremely fun and connecting and hackerish to do local coding with a local model. Try it.

However true that is, it now has only to compete with the US, where any model could be shut down by the Government on a whim with no clear rules at any time.

It's happened once, could happen any time.

Not good for business!

LLMs are reducing n-day exploit time rapidly.

https://red.anthropic.com/2026/n-days/

So that is a poor bandaid to use now. Maybe instead validate things before, and have more of a cathedral and human reputation system.

Have any kind of provenance. eg like Debian has for 30 years. Key signing in person etc

They can't under GDPR. The DMA is for market access - there are other laws for privacy. Those require use commensurate with what is needed for the service, so anyone who e.g. scraped all of a user's local info and stolen it would be breaking EU privacy laws themselves.

This is not complicated. Even in the US, every other industry is regulated to your benefit, you're just used to it and haven't realised. Digital technology obviously needs to be too. And yes, you have to do it properly.

AISI in the UK has been doing this for years - there are lots of papers https://www.aisi.gov.uk/category/safeguards and specific reports, e.g. this on GPT 5.5 https://www.aisi.gov.uk/blog/our-evaluation-of-openais-gpt-5...

This old post goes into lots of detail about what they do to red team and why: https://www.aisi.gov.uk/blog/early-lessons-from-evaluating-f...

NIST's similar unit in the US is now called CAISI https://www.nist.gov/caisi - interesting that the most recent post is an evaluation of DeepSeek capabilities, which sound more like watching China. But presumably this executive order alters the emphasis?

Right the original article says "Do you think macOS will get better or worse in the next 2 years?" (rhetorically implying "worse").

That could easily be true and Apple "will use even more tokens and spend even more money".

Well, or not spawn any external commands, and actually have tools made of code written by someone who thought about what the agents at each level should be limited to doing.

Make harness independent of model, so when pricing or quality changes you can switch.

Avoid lock in to stack from one provider (things like a harness that only works with models from one provider and so on).

Use local models (a couple of them do work a bit now, if you have 20Gb video RAM), which saves money and is more private, and works offline.

Can improve the harness, fix bugs in it, make it compatible with different systems and techniques.

This game happens every time in new cycles of developer technology. The good bet historically has always been to use open source - there's a reason most developer tooling just pre-AI revolution was open source (even things like Java and .NET which used to be proprietary).

Reporters without Borders recently released Press Freedom Index 2026 puts Malta 67th, and the UK at 18. So no, certainly not much better - although looking at some of the historic data, it was better e.g. in 2010.

https://rsf.org/en/index

Agreed. All I see is a grok summary of a lot of X posts. The original link is not suitable. Anyone have a link to a proper announcement?

I was finding this really interesting, that maybe a human had written it and it really reflected a vision for how we build software in this new world. I want to know the way, I'm curious!

Until I got to "One platform, three modes." and my brain just pattern matched "AI slop" and the entire post dissolved into meaningless for me.

I don't know if I can stop my mind reaching this conclusion. I'm sure someone at GitLab made some effort to carefully edit the post... But that it wasn't entirely rooted in a human who'd worked out how this stuff goes, but clearly had lots of AI writing it out... Just made my instinct go "this isn't worth paying attention to after all".

I've tested this extensively in a workflow (not agentic) context, and you're right, the underlying models are both good at full rewrite of code files, and at doing search/replace.

They've been decent at full rewrite for 2 years. I don't think they were good at search/replace until a year ago, but I'm not so sure.

It's true that the models 2 years ago would sometimes make errors in whole rewrite - e.g removing comments was fairly common. But I've never seen one randomly remove one character or anything like that. These days they're really good.

Main reason agentic harnesses use search/replace is speed and cost, surely! Whole file output is expensive for small changes.

Qwen 3.6 is out now and a touch better than 3.5.

I'm finding Google's Gemma 4 even better though - seems to hold up the agentic loop better than Qwen.

All will load into 20Gb of VRAM. None are amazing, but they do just about work.

Block post - they contributed Goose: https://block.xyz/inside/block-anthropic-and-openai-launch-t...

The example usually given by pro-sanctions campaigners is South Africa (https://en.wikipedia.org/wiki/International_sanctions_during...)

Looks pretty real:

https://github.com/twitter/the-algorithm/blob/7f90d0ca342b92...

When this started it really put me off X - I'd have tolerated, and almost liked the idea, of a freedom of speeech place. But a place that boosts its owners posts... Nope.

I'm out - it's such a big personal diss of me, I'm not interested any more.

I tried Mistral for a bit, and it is so fast everything else feels bad now by comparison. I think there's lots of opportunity for OpenAI, Anthropic to stumble on features and performance.

The spikes in the last 2 years have happened for very short amounts of time. If renewables are working, you don't get a spike, and save loads on this tariff. The small amount of time they're not, you sometimes have to pay more, but not for long enough to matter. It's fundamentally more effective for everyone than the default of buying the insurance of fixed prices.