lispyrabbit

Codex hit 4M active users, less than two weeks after hitting 3M. We will reset rate limits today!

やばいな Github Copilotが新規サブスク受付停止 すでに課金してるユーザーだけ継続利用可能 推論用の計算資源の枯渇がいよいよリアリティを帯びて来た Gista.jsでCodexとClaude Codeの次に推してたのに うーん、どうしよう… Cursorは初心者には勧めにくいし…

Kimi 2.6 looks like it could be the new contender for best open agentic model for Hermes Agent, stacking up strongly against even opus 4.6 in these agentic benchmarks 😲

Meet Kimi K2.6: Advancing Open-Source Coding 🔹Open-source SOTA on HLE w/ tools (54.0), SWE-Bench Pro (58.6), SWE-bench Multilingual (76.7), BrowseComp (83.2), Toolathlon (50.0), Charxiv w/ python(86.7), Math Vision w/ python (93.2) What's new: 🔹Long-horizon coding - 4,000+ tool calls, over 12 hours of continuous execution, with generalization across languages (Rust, Go, Python) and tasks (frontend, devops, perf optimization). 🔹Motion-rich frontend - Videos in hero sections, WebGL shaders, GSAP + Framer Motion, Three.js 3D. 🔹Agent Swarms, elevated - 300 parallel sub-agents × 4,000 steps per run (up from K2.5's 100 / 1,500). One prompt, 100+ files. 🔹Proactive Agents - K2.6 model powers OpenClaw, Hermes Agent, etc for 24/7 autonomous ops. 🔹Claw Groups (research preview) - bring your own agents, command your friends', bots & humans in the loop. - K2.6 is now live on kimi.com in chat mode and agent mode. For production-grade coding, pair K2.6 with Kimi Code: kimi.com/code - 🔗 API: platform.moonshot.ai 🔗 Tech blog: kimi.com/blog/kimi-k2-6 🔗 Weights & code: huggingface.co/moonshotai/Kim…

This is the moment NVIDIA should be seriously worried. In the next couple of weeks DeepSeek V4 will be launched. It’s a direct attack on the entire AI stack that American companies have spent years locking down. Full “de-NVIDIA-ization”, a complete shift away from CUDA into Huawei’s CANN ecosystem, running on Huawei Ascend chips. That means one thing, breaking the dependency that made NVIDIA untouchable. 35x faster inference vs early versions. Nearly 3x the performance of NVIDIA’s H20 on a single card. 40% less energy consumption. Over 95% CUDA compatibility with migration times collapsing from months to hours. Even Jensen Huang has already admitted it. If this works at scale, it’s a “terrifying outcome” for US companies. Because here’s the real problem, this isn’t happening in isolation. Chinese tech giants like Alibaba, ByteDance, and Tencent are already ordering hundreds of thousands of Ascend chips. Market share is shifting fast, domestic chips now at 41%, NVIDIA slipping to 55% in China’s AI server market. Additionally DeepSeek V4 is reportedly offering API costs at a fraction of US competitors. $300 for massive workloads that would cost $2,500+ on OpenAI models, or even $5,000 on Anthropic. So this isn’t just about one model. It’s about China building a fully independent AI stack, chips, frameworks, models, and applications. Completely outside of US control. NVIDIA doesn’t just lose sales. It loses its grip on the global AI standard.

i am not being able to recover from what grok 4.3 is doing. been pushing it through autonomous agent work overnight and its operating at a level other grok versions was not capable of. the way it handles multi step reasoning, the way it doesnt bail halfway, different model energy entirely. everyone sleeping on this needs to wake up

Jensen Huang just made the case for American empire. Said it plain. Didn’t flinch. Didn’t walk it back. And almost nobody caught what he actually admitted. Jensen Huang: “The amount of compute in the United States is a hundred times more than anywhere else in the world.” One hundred times. That is not a market lead. That is a monopoly on the future of intelligence. The kind that compounds every six months until no one else can close the distance. Jensen Huang: “We make sure that the US labs are the first to hear about it and the first chance to buy it.” Every chip Nvidia designs. Every architecture they ship. America gets first access. Everyone else gets what is left. That is not a sales strategy. That is arms distribution with a quarterly earnings call. Jensen Huang: “And if they don’t have enough money, we even invest in them.” The company building the weapons is bankrolling the people who fire them. Nvidia is no longer a public company. It is a state instrument with a stock ticker. Jensen Huang: “Why would you want the United States to give up the world?” The CEO of the most valuable hardware company on earth did not hedge that. Did not qualify it. He said it like it was obvious. Because to him, it is. Nations used to be measured by steel output. Then oil reserves. Then warhead count. Now it is how much intelligence they can produce per second. Compute is no longer a commodity. It is a strategic resource. Like uranium in 1944. Except this one doubles faster than anyone can respond. Europe understands none of this. They are drafting AI regulations. Compliance frameworks. Ethics panels. Risk tiers. They are bringing paperwork to a physics war. You cannot govern intelligence you do not have the silicon to produce. China gets it. That is why they are building fabs, not filing comment periods. Nvidia already made sure the gap is not annual. It is generational. Silicon Valley still thinks it is building consumer software. Huang just told them they are building American infrastructure. Every model trained here runs on machines that exist nowhere else. Every company that scales here scales on silicon no rival can touch. The world thinks Nvidia sells chips. Nvidia sells the ability to think. And they only sell it under one flag.

Interesting insights, especially this: Hermes starts off as any other agent does, inefficient and often not sure how to complete a task that is training didnt have priors for. However, solve it once and you unlock huge efficiency gains. I sometimes call this linearized RL, coming from a post training backround, where you would run a task dozens of times, and determine which runs won, and then train on it, this is strong parallels, but in one agent loop, and obviously, learned at test time, and locked in through skills and in context learning, rather than through weight updates.

Larry Ellison just asked the one question no journalist on Earth can answer. A Wall Street Journal writer told Ellison to his face that Elon Musk doesn’t know what he’s doing. Ellison didn’t argue. Didn’t get emotional. He just asked a question. Ellison: “This guy is landing rockets on robot drone rafts in the ocean, and you’re saying he doesn’t know what he’s doing. You ever land a rocket?” One question. No recovery. Ellison: “Who are you? Why should I believe you as opposed to my friend Elon?” This is the question the entire media class has been dodging for a decade. Who are you to judge? What have you built? What have you shipped? What problem have you solved that didn’t involve a keyboard and a deadline? Ellison: “You’re there in front of your Apple Macintosh typing up an article saying Elon’s an idiot.” They sit behind a laptop they did not engineer. Using a network they did not build. Running on silicon they cannot explain. To tell the world that the man sending humans to space doesn’t know what he’s doing. They have never built anything heavier than a Word document. And they publish it with absolute certainty. That’s the part that should disturb you. Not the criticism. The confidence behind it. The total absence of self-awareness it takes to judge disciplines you wouldn’t last a single semester in. Musk does not operate in opinion. He operates in the physical layer of the universe where the math closes or the rocket does not come home. His critics operate in a text editor. He built the vehicle that carries NASA astronauts to the International Space Station. The satellite constellation delivering internet to active war zones. The EV that forced every automaker on Earth to abandon their combustion roadmap. His loudest critics built a byline. So why the coordinated hatred? Because they lost the leash. The attacks didn’t escalate because Musk got worse at engineering. They escalated because he bought X. He cracked open the algorithm. He handed the public square back to the people. And he shattered their ability to control what you’re allowed to think. They don’t hate the engineer. They hate that the engineer took their monopoly. You cannot cancel a rocket. You cannot publish a hit piece on gravity. You cannot edit the laws of physics. They own the syntax. He owns the physics. One of them is going to Mars.

@yao_ej24569 @WesEarnest Have you tried v2 TUI, finalizing next week? if your hermes is updated run `hermes --tui`

omg we have the same biggest annoyance! In our TUI v2 it will keep your original prompt for that agent loop right beneath the chat bar so you never forget. It was my biggest ask to our front end devs with the next TUI. I'll run 12 in parallel, it'll work for 20 mins, I come back, and then I forget what I even asked it to do!! I feel your pain!

これはYann LeCunが正しい。「生成AIが人々の仕事を奪うかどうか」は労働市場における技術革新が及ぼす効果をきちんと研究している研究者や専門家たちの分析を待つべきであって、生成AI提供サイドのポジショントークを鵜呑みにするべきではない

Hermes Agent is model & tool backend agnostic for a reason, everyone should have access to AI. We don't dictate the rules of use for your agent, YOU do

Jensen Huang just told Silicon Valley it’s fighting on the wrong floor. Every boardroom in tech is locked on the same question. Which model wins. OpenAI or xAI. GPT or Claude. Grok or Gemini. Trillions moving on that bet alone. Huang zoomed out and showed them the whole building. Huang: “AI is actually essentially a five-layer cake.” Energy at the bottom. Chips above it. Cloud above that. Models next. Applications on top. Five layers. One war. Everyone crowded onto the fourth floor. Huang: “This is where most people think AI is.” He was pointing at the model layer. Every pitch deck. Every valuation. Every founder story. All packed onto one floor. One floor below the finish line. Three above the foundation. The middle of the building. Huang: “At the bottom is energy.” Not data. Not parameters. Not talent. Power. You cannot out-code the grid. You cannot train a frontier model with a press release. The smartest model on Earth still needs a dumb turbine spinning somewhere. The smartest engineers alive are building on top of someone else’s silicon, inside someone else’s cloud, powered by someone else’s electricity. They own nothing beneath them. Huang: “This layer on top ultimately is where economic benefit will happen.” Healthcare. Finance. Manufacturing. The only floors where AI actually meets money. Every dollar of real value lives at the top. Every physical constraint that decides who gets to play lives at the bottom. The model sits in between. Squeezed from above and below and owning neither end. Silicon Valley is burning hundreds of billions to build plumbing for somebody else’s economy. The basement decides if it runs. The penthouse decides if it pays. The companies building models think they are building the future. Huang just told them they are the middle layer in someone else’s cake.

Hermes Agent just hit 100,000 stars on GitHub!!! Thank you everyone!!

Hmm just noticed that there’s a new 100B stealth model called "elephant-alpha" sitting at No.1 on OpenRouter trending list . Seems really good for agentic tasks and coding in my tests. The "no-yap" energy is the real standout. So far it works very well, even for browser automation. It just executes without the usual conversational filler. And the speed is impressive given the size, snappy, with no wasted fluff. No idea which company or lab it came from.

この人、その後のみんなのスレでも言われてたけど、会社を一つのプラットフォームに頼るのってやっぱり危険性があるよねっていうことをみんなから出てたような気がするし、その通りだと思う。特にAnthropicはクセもんだよな。これだけいろんなものを潰す、そして依存させるものを作っておきながら、小手先一つでBanできるんだから怖いもんだ。

個人的には、なんて言うんだろう。AIエージェントにGUIをコントロールさせるとはあまり好きじゃないんだよね。なんというか、本質からずれてる気がする。 全てのものはAIエージェントがコントロールしやすいように作り直されるべきだし、CLIの方がエージェントにとってコントロールしやすい対象だというふうに思っている。GUIはなんというか、ワンレイヤー余計なものがかぶさっている感じがするんだよね。あれは人間が操作するもののためで、AIが操作するもの、効率的に操作するためのものじゃないよね。そこが気になっているところなんだ。 もちろん緊急回避用にはいいんだけどね。CLIもない、APIもない、どうしようもないアクセスができないもののためには、最後の砦としてコンピューターユースはあると思うんだけども。

codex is becoming a full agentic IDE

Thanks to our incredibly community and Viviennn in particular, we have greatly improved session-based recall for Chinese users! Thanks for the contribution! github.com/NousResearch/h…