RE: https://cosocial.ca/@mhoye/116325532654671149

OK, this is evil. #Anthropic specifically coding #Claude to be nontransparent / opaque. Should we stop talking about bringing them to the EU now? Fully against the spirit, intent, and law of the #AIAct. Unless those guys are REALLY convincing that they never wanted to do this and want raw money less than to be working within functioning democracies and the rule of law. #AIEthics #devops

@j2bryson that's not necessarily evil. I've actually been using Ollama and Claude to write commit messages for me and it's 10x better than what I would do. However, I have a similar function because I don't want my clients to see the repo and just assume the entire thing was vibe-coded.

AI isn't going away, but it should be used for handling the skut work and some analysis, not real design work. It sucks at that.

For this, however, with the shit I see from my developers any even myself, I really prefer it.

@coldfish @j2bryson Transparent disclosure of any kind of AI-assistance shouldn't be a controversial topic for anyone who's choosing to use it. The possibility of your clients distrusting you about the extent of the AI-assistance in your work arises precisely because there's a lack of transparency about it in the first place. And if your clients have an issue with you using genAI in your work in general - well, I suppose you change your target audience then. Because consumers have a right to know. And there are plenty of reasons besides accuracy or security for why someone would choose to opt out of it. Maybe they'd rather not contribute to the climate impact of it or to the abuse of people in the Global South in order to drive the AI-adoption forward. Similarly to how vegetarians choose to not consume meat. That doesn't give manufacturers freedom to not mention that there's meat in their products. I suppose you can see how that would be unacceptable.
What is the problem with prompting a coding assistant to not include internal codenames and versions in commit messages when the user type is an anthropic internal user who may be working with unreleased versions of Claude and Claude code? I fail to see how this is lying by any definition.
@jonathan.boxcake.net It's asking the system to actively supress who has written the code how and why, in EVERY context. It's a digital artefact, it should be trivial to actually make it succinctly record how it's being used.