The obsession of religious people with validating their beliefs through reinterpretation is frustrating. Their books are often filled with contradictions to science, yet if even one thing sounds similar, they use it as validation.

Terran R stage 1 spotted 👀

I don't care about an IPO or the stock market, but saying SpaceX hasn't created anything is peak ignorance. They are solving the some of the hardest problems, including full reusability, efficient thermal protection, and the infrastructure needed for high cadence flight.

If it were up to Anthropic, even GPT-2 would have been too dangerous to release; they would even demand a ban on certain math just because it can be used in research for AI models.

Open source is the best way forward for AI, and today I have been proven right once again.

Most of the problems listed in it will be solved one at a time, and not just suddenly become a thing, just like how Starlink happened.

A funny thing I've seen recently is that so many self-proclaimed progressive people instantly become just as intolerant as MAGA morons when they can't get away with their idiotic logic.

So, you are counting a rocket in its development phase as part of the success rate?

I have one suggestion for you. When I select text and right click to ask Gemini, the selected text should automatically appear in the Gemini panel.

I don't like Elon as much as the next person, but one thing I know is that SpaceX engineers are amazing; they are already working on some of the most challenging problems.

To me, open source is the way forward for technology and power, so it does not remain in the hands of a few chosen people.

Thank you for doing this. Fast local models are finally possible. First MTP, now this.

Its funny seeing CCP and USA propagandists fighting each other. It is kind of funny, and also sad, that the main driver is not "I will do better," but "I hope they fail." This is one of the things that can expose a propagandist mindset.

What’s new in Blender 5.2 LTS Beta? These features are worth trying first: 🔸 Cycles texture cache - Cut memory use and startup time on texture-heavy scenes. 🔸 Hair/Cloth dynamics - Node-based physics simulations powered by Geometry Nodes entered the chat. 🔸 Online Asset Library - Host libraries remotely, browse them inside Blender, and download assets on demand. 🔸 Loop Tools - Circle, space, and flatten from the beloved add-on, now built into Blender. 🔸 Rendering – new Thin Wall mode in Principled BSDF is here to render leaves, paper and other meshes with minimal thicknesses in a more photoreal way. 🔸 Geometry Nodes - New Bevel node, Geometry bundles, lists, a new Sample Sound node to pull frequency data from audio, Empty objects now support Geometry Nodes and so much more. 🔸 Grease Pencil - A brand new Fill tool and updated brush library. 📖 Check the full menu in the release notes: developer.blender.org/docs/r… Support future Blender development: fund.blender.org 🧡 #b3d

The Mesh Bevel node has been added to Blender 5.2 #b3d #geometrynodes

What if you could take three completely different model families… and distill them into one tiny model? 🤯 📜 Paper: arxiv.org/pdf/2605.21699 MOPD (Multi-Teacher On-Policy Distillation) has become a standard procedure in post-training. We already distill multiple specialized variants of the same model into a single set of weights. But what if we could go further - and distill models from entirely different families? Turns out, it is possible. Today we’re releasing a paper on cross-tokenizer distillation - our first steps in this exciting direction. 📄 We distilled Qwen3-4B, Phi-4-Mini, and Llama-3B into Llama-3.2-1B. MMLU jumped from 32.05 → 46.32 when using multiple teachers. 📈 The team is now working on Nemo-RL integration so the community can try this method in their own settings. Plus, we are scaling experiments up. 🚀

Jan now connects to any Anthropic-compatible endpoint, from LiteLLM to Bedrock proxies to self-hosted Claude gateways. Update your Jan to v0.8.1. jan.ai/