webcoon

I see a lot of benchmarks which have not data for DS4.1 and a lot of other benchmarks, where DS4.1 beats this new model. I guess I can finally stick to my current AI coding setup for longer than a month.

"Across all attempted problems, the agents sent 4.9 million messages and used about 300 billion output tokens."

At a conservative estimate of GPT 6 Astra pricing, this would have cost upwards of 15 Million dollars for anyone using the OpenAI API!

To me this is the one silver lining. Yes, they can solve millennium prize problems, but it still costs a fortune.

Are these animated visualizations in the sky-explainer all original? If so, I am blown away by the visual consistency and meaningfulness of these throughout the article.

And it STILL uses text-based Go templates instead of a proper language based on structured input and output? This was always my main pain point with Helm and also of many others I talked to. This major upgrade was years in the making and they couldn't add support for a single of many available options like CUE, JSONNET, or KCL? What an utter waste.

Docs mention absolutely nothing about offline use, service workers, or PWAs. How am I supposed to build a modern web app with that? Rely on always having internet?

Awesome! This speaks to something, which I've been thinking (and wishing) for a long time. I've already done probabilistic programming in a scientific context (Python) and classical software engineering for web development (TypeScript, Python, Rust), and I've always wondered why I couldn't have the real-world modelling capacity of the former with the static type assurances of the latter. Love that you (and Microsoft) are thinking along the same lines! Do you perhaps know of any Python implementations for this? There are plenty of dynamic stats programming libraries, but none offer typing solutions AFAIK.

Not doing it for me here in Germany

If I only had a penny for every promising GitHub repo, which was immediately abandoned by the authors after their paper was finished...

This looks awesome! I use a paid VPN mesh service called Tailscale to build k8s clusters which can span across cloud and home setups, but was always frustrated that the Tailscale server component is closed source. Netbird seems to do pretty much the same thing, so maybe I can switch over to it once it's stable enough.