Clueed

in any corporate or half compliance-relevant setting switching isn't trivial. new DPA, subprocessor notifications, TIA, procurement review, security questionnaires, plus re-running your evals because prompts don't transfer 1:1. token cost is just one of the line items.

I tried it with minimax 2.7 and it really didn’t like the editing tool; collapsing rather quickly to using sed to edit files.

I guess it makes sense that models don’t generalize perfectly to arbitrary tools but are biased to those in its training data, especially for a common operation like editing files.

The Gemini family might be a good pick here since it generally underperforms in agentic tasks (due to lack of training data or other reasons) and thus might not have this inherent bias towards specific tools.

Really interesting. The crazy changes in opus 4.6 really make me think that Anthropic is doing library-level RL. I think that is also the way forward to have 'llm-native' frameworks as a way to not get stuck in current coding practices forever. Instead of learning python 3.15, one would license a proprietary model that has been trained on python 3.15 (and the migrations) and gain the ability to generate python 3.15 code.

I found DeepSeek R1 (better for questions) and V3 (better for prose) to be very willing to discuss sex with a simple system prompt, as well as being very pleasant in articulation. I guess, I prefer them because they are almost SOTA and very large.

Not through the official interface though. Needs to be hosted by a third party. OpenRouter has a generous free tier for both.

I just saw that there is an abliterated version as well. Not sure how to try it though.

I love his videos but all the sound demos are a a bit too much sometimes :D

Not sure about the specific implementation here but the very definition of cosine similarity includes normalization. [0]

[0] https://en.wikipedia.org/wiki/Cosine_similarity

I agree, if you don't like a key feature of a library, like tailwind-based styling which is meant to be customized, then that library is probably not for you.

The form component is more the exception for its complexity but forms are also very complex on the web.

Being a wrapper on top of Radix is kind of the point. You can't build a nice-looking MVP in an afternoon with Radix - you need one afternoon just to style every state of button - but with shadcn you can. Their experiment is to give you the ability to still spend an afternoon styling the button later without having to hack around a complex library. Your ability to rewrite it even away from tailwind is exactly the point. Ever tried rewriting MUI to use tailwind?

I've looked into the available options of parsing PDFs, including pypdf, which is what is being used here, a while ago and it's not good. While I haven't testing equations specifically, it think it's fair so assume that the results will be subpar especially complex ones.

I guess, this could be an application of the agent model. I've seen multiple LLMs recently trained specifically on LateX parsing. One model would recognize from the parsed PDF garbage that there is probably an equation there and call a different want to parse it.

Another aspect is css class reuse. Inline styles are not preprocessed so each inline style is a 1:1 increase in bundle size. In large applications this can be significant. Imagine having hundreds of popups each with 50 lines of css. It adds up.

I've read it like this: They differentiate themselves from OpenAI/Anthropic/other major players by being in the EU and (more) open-source; they differentiate themselves from other European competitors by being as good as OpenAI. I think it's a fine proposition given the early market and their team, as long as this hypothesis holds true for years to come,

> we believe that most of the value in the emerging generative AI market will be located in the hard-to-make technology, i.e. the generative models themselves,

and they truly can compete with market leaders.

I don't get why nobody is talking about this. I'm starting to suspect it only effects a subset of all AirPods Pro and is covered by this replacement program even though it's not really tasted clearly.

AirPods Pro even compared to the the non-Pros have awful microphones.

https://discussions.apple.com/thread/250813242

I agree. You can make a better case for “writing in a corporate context requires humility,“ e.g., someone choosing to post in public slack channel instead of talking to everyone individually.