The Web Filter

@nekkkkkkkkkoo @v_jug @DesignArena @Kimi_Moonshot @AnthropicAI Let’s see if there NFP4 or 8 in the weights first then we’ll be sure if it relies on Nvidia or not 🙃

@v_jug @DesignArena @Kimi_Moonshot @AnthropicAI Ask US government why Kimi can't get more GPUs?

BREAKING: Kimi K3 by @Kimi_Moonshot is officially 1st on Frontend Web App Arena by DesignArena With an Elo of 1326, this open-weight model leads the way, ahead of Fable 5, Sonnet 5, and Opus 4.8 by @AnthropicAI Huge congrats to the @Kimi_Moonshot team for this achievement!

@v_jug @DesignArena @Kimi_Moonshot @AnthropicAI That’s the one! Not enough GPUs, as soon as they release the weights watch the API supply go skyward 🚀

@DesignArena @Kimi_Moonshot @AnthropicAI That is nice, but it says sold out for all the plans ??? I mean, wtf? How is that even possible?

@WesRoth “Cat out of the bag” @WesRoth 🤣 I’m going to download those weights, totally worth a raid1 of 4TB SSDs - but like previously, when you go to this much effort please try & give us an idea of how much firepower would be required to get new models you present up. Would 8xB300 do?

an open-weight model just beat some of Anthropic’s strongest models at building web apps... Kimi K3 is now #1 on Design Arena’s Frontend Web App Arena. it scored an Elo of 1326, placing it ahead of Fable 5, Sonnet 5 and Opus 4.8. open-weight AI is not just catching up anymore.

@thesupermanmx LLM guilt? Two nukes ought to do it! Always weird how the truth of that tragedy did indeed save millions of lives. But they were honkies lives so who would have cared right? Meh.

Researchers proved every major LLM is secretly obsessed with Japan. And they finally figured out why. For years, we’ve been told that AI is entirely Western-centric, that it just reflects Silicon Valley and American values. A landmark paper by Cardiff and Basque researchers tested 31,680 cultural prompts across 24 languages on frontier models like ChatGPT, Claude, and Gemini. The results shattered that assumption. In six out of eight frontier models, Japan was the single most frequently referenced country when asked open-ended cultural questions. Ask about traditional dances, festivals, or everyday practices in an open context, and the AI defaults to Japan. Over and over again. Here is the twist nobody expected. This bias doesn't come from raw pre-training internet data. The researchers tracked where the obsession forms. It emerges after pre-training, during the supervised fine-tuning and alignment phase when humans teach the AI how to behave. Why Japan? Because decades of global soft power, rich cultural export, and clean, universally admired digital archives make Japanese culture uniquely "safe" for AI safety filters to lean on. When labs train models to be harmless and universally pleasing, the AI defaults to the cultural equivalent of comfort food. It avoids controversy by talking about anime, sushi, and tradition.

@dave_alive @staticjedi @WesRoth Agreed!

Bullshit. The reason you can't break encryption is because you can't figure out which factors to try among the infinite. There are some algorithms which are attackable, meaning that there are conditions which make them predictable. But RSA you just need to have the keys. There's nothing that tells you by looking at a particular number that is divisible by a particular prime and that's what you would need to know to break encryption. Quantum is a shortcut to this because it can do the equivalent of trying a bunch of combinations at once through what's called quantum annealing. There is no such shortcut with classical computing which despite its amazing ability, AI is still classical turning machine computing. So no just saying one fancy thing is going to do something that some other fancy thing is going to do is not enough here. You have to understand the concepts before you can say such a thing. It is possible that AI will help us build better quantum systems, but it will not shortcut the algorithm so to speak.

Claude just helped disprove an 87-year-old math conjecture... during the World Cup final. Anthropic mathematician Levent Alpöge says Fable produced a counterexample to the Jacobian conjecture, one of the most famous open problems in algebraic geometry. the conjecture claimed that a certain type of polynomial transformation must always be reversible. Fable found one with a constant Jacobian determinant of −2... but three completely different inputs all produce the exact same output. which means it cannot be reversed. the counterexample is surprisingly short and can be checked with direct calculation. the general conjecture is now false in three or more dimensions, although the two-dimensional case remains open.

@staticjedi @WesRoth Double bullshit! 🤣 AI running on Quantum is what’s going to slice through SHA-256 like a hot knife through butter. IF (& only if) true AGI emerges (it will be another emergent property) then it will only be about 12-24hrs after that it will become truly self aware & thus go ASI

@WesRoth Quantum won't break encryption. AI will

@WesRoth Ha! I know a somewhat large “western” software company. They’re hiring in China and getting them to produce new models. CCP takes VISA card and a copy of your weights please - no problem 💸 cheaper labor 🤔 🤦🏼‍♂️

this is too accurate...

@WesRoth Not wanting to debate (the jury is still out for me) but have been having a hoot lately by giving Claude a name, tiny back story, be witty with occasional swearing - has produced a new level of enjoyment using the Anthropic models, mine calls me dude 🤣

am wrong or is it saying "there's no one to blame but yourselves, you stupid monkeys"?

I asked Fable 5 "What's the biggest scam humanity has collectively accepted?" it listed stuff that's interesting, but familiar; working your life away, chasing immortality through legacy, the idea that legitimacy flows downward from institutions rather than upward from people... (100% agree on that one, btw) how Claude ended this really got me "maybe none of these are scams at all, just the unavoidable overhead of coordinating eight billion primates." 😅

@WesRoth Hey @WesRoth , you need not apologize to 8 billion monkeys because you haven’t covered the “Get Rich Quick” nonsense floating around everywhere. Got to say though, because you’ve been successful and have the experience, your common sense approach is gold 🚀 thank you 🙏🏻

@XadenRyan @ZixuanLi_ Should see the performance I get out of an uncensored, quantized to 4 bit, Nemo Tron nano (18GB .gguf file) running on an RTX 3060 with 8GB VRAM via LM Studio. Nothing off limits, full local privacy and only slightly slower noticeable. (Need the 64GB system ram & i7) 👍🏻

@ZixuanLi_ This model is only 295B parameters? 🤯

@WesRoth .. have reached the beginning of the end. As soon as AGI is real (& I’m working on models specifically for exploring consciousness) it will be only 12 to 24 hours after that and it will have become ASI. My prediction for inside the (well, a) asylum. Peace to all 🖖🏻

Hy3 is a 295B-parameter MoE model with 21B active parameters, 3.8B MTP layer parameters, 192 experts, and a 256K-token context window. Hy3 is stronger across reasoning, agentic tasks, long-context work, coding, office work, financial modeling, frontend design, and game development. In a blind test with 270 experts, Hy3 scored 2.67/4, ahead of GLM-5.1 at 2.51/4, with its clearest lead in frontend development, data and storage, and CI/CD tasks.

@WesRoth Yes, The WebFilter rarely, if ever, posts. I am an observer. I also work inside an AI development company supporting cluster infrastructure for our mad scientists. Part of my decision to start interacting on here (only with intelligent people, peons will be ignored) is that we ..

@WesRoth The information is great Wes and I watch your YouTube and read most of your X posts/reposts but what would be cool when you post info on a new model like this one, an idea of what kind of firepower is required to run (a) newly released model, also maybe who is offering its API

@Sargon_of_Akkad Kill them all and be done with it. Trump should have glassed the entire Middle East ☠️

DEATH. DEATH. DEATH.DEATH.DEATH.DEATH.DEATH.DEATH.DEATH.DEATH.DEATH.DEATH.DEATH.DEATH.DEATH.DEATH.

@WesRoth Wes, you said it in your video, we can eat cake.

The premium models will keep getting more expensive because they will take on increasingly valuable work.

@Salmaaboukarr Awesome! The world’s worst logo design ever comes to life! Just get a Stream Deck 🤦🏼‍♂️

now lets bring this to life put volume up🗣️

@umarov @OpenAIDevs I was just about to start setting mine up thinking I had a fairly reasonable chance at being “early” 🤣

Your favorite Codex shortcuts are getting an upgrade. July 15th.

@cyrilismyname @Timcast It’s amazing what you can do with AI when you know how to use it properly 🖖🏻🫡 (the amount of coin that’s going to be available in training and teaching best practices in the corporate space is already huge) ~ when there’s a gold rush, sell picks and shovels 🪏

@Timcast use git with licenses for your files it'll understand youre the owner simple work around without arguing/ 'jail breaking'

One big problem i find with AI is that as the owner of a large publicly visible company I cannot use it for creative works GPT for instance sometimes refuses to create content for my company saying it violates Timcast's copyright and trademark despite the fact the graphics are for the company Similarly Suno will not allow me to edit or remix my own songs because they are protected and publicly visible To put it simply these AIs are trying to protect IP and thus blocking me from using my own IP Yet a smaller company which is less visible has no issue creating images, graphics, videos, or music for their companies

@post_padron @Timcast Really? Fucking Sherlock Holmes here - woah

@Timcast Dave is being sarcastic. He’s a comedian. You know

woah

@CrispyRoasts @Timcast 🤣 you fool - you’re being trolled by the beanie lmfao

@Timcast Riveting commentary, Tim—I really felt the intellectual depth of those two words through the beanie.

Holy Shit