People are really misunderstanding Skills, in my opinion. It's not really about the .md file. It's about the bundling of code and instructions. Skills assume a code execution environment.
We have to find a way to punish Google if they move forward with this. We need the Gemini folks to be worried that this distraction will jeopardize their competitiveness in AI.
It's not either/or. Generally you finetune when optimized many-shot still doesn't hit your desired quality bar. And it turns out with RL, things like system prompts matter a lot, so searching over prompts is a good idea even when reinforcing the desirable circuits.
I just read that document and, I'm sorry but there's no way it's written in good faith. You support open weights, as long as they pass impossible tests that no open weights models could pass. I hope you are unsuccessful in stopping open weights from proliferating.
If you read this article and it gave you pleasure, examine those feelings. I find the generally schadenfreudic reaction to be disgusting. These people just wanted to get together to talk about something they enjoy and you don't.
I think this game is one of the best pieces of storytelling ever created, and it's so weird and arthouse that I can hardly believe there's an audience for it.
Meanwhile, the other rumor is that Nvidia stopped production on RTX 40 series in order to shift all manufacturing towards ML chips like H100. Seems like they're both in an all-out sprint to produce as many H100 and MI300X as possible. Which makes total sense to me.
No, do not do this. Use Matrix, or...anything else really. I tried building on top of IRC a decade ago and it was a more forgivable mistake back then, but a mistake nonetheless.
I would also check out their 3B model. I tested it on launch with LoRA fine-tuning and found it to be surprisingly capable despite its size. I think a lot of people are skipping past testing it because it only has 3B params.
Horizon: Call of the Mountain is a recent example of a UX interaction that I really like with eye tracking + a peripheral. It uses a button press, so it's an intentional action, but the eye gaze determines which action to take.
I think in the world of ML, contributing that much compute should count as co-creation. There is a lot of code published by academia that just needs compute and data, but they don't receive it. Stability deserves credit for Doing the Thing. IP rights are another thing but that's a whole subtree of legal questions that society is barreling towards.