I'm quite worried about the future we are inevitably heading towards where the threshold for economic utility rises above what ~99% of the human population can meet. Contrasting Dario and Sam on this issue wrt AI is so funny to watch. Dario is a hyper-autist who, for all his faults, I think genuinely speaks his mind. He sees the tsunami coming. Software engineering may very well be gone in the next few years along with many other computer-facing careers (of course assuming one is running an actual company and not a UBI corporation, which most of Big Tech actually is). He is trying to warn the normies that this is something that is actually coming, and all the screaming and crying about AI and datacenters will at this point only temporarily delay the inevitable. It is easy to dunk on Dario for peddling this doomerism, but can anyone refute his actual worry? Sam, on the other hand, is hyper-Machiavellian. He has the social intelligence to recognize that telling the normies AI is gonna take all their jobs (whether it actually will or not) is bad for his company. So he has to preach this blatant lie that he clearly doesn't believe that 'AI is gonna create more jobs'. Yeah no it won't (with the exception of human data harvesters). For what it's worth, it seems like one of the main reasons the economy has still been fairly agnostic to 'AI efficiency improvements' is because most jobs are fake anyways. People are not becoming 'More Efficient With AI™' because there was never anything for them to make more efficient in the first place (other than maybe looking busy). To what extent does AI force society to confront how much economic activity was never particularly productive in the first place? So what happens if we become more honest with ourselves about the actual utility of most jobs, especially when a computer can do them extremely easily and cheaply? What happens when 99%+ of the human population has little to no economic utility? I think this social/cultural revolution (downstream of the AI revolution) is affecting young people most of all. I know kids who graduated from HYPSM+ who literally cannot get a job. Look at any honest news article and *minimum* 50% of new college grads are un/underemployed. I think this is a sign of things to come and how we respond to millions of people having less economic utility than the computer will likely become the defining challenge of our generation.

Kimi k3.1 is easily the best open-source model right now. But that doesn't mean it's frontier or anywhere near Opus/Astra despite what the heavily benchmaxxed benchmarks say. Go use the model yourself and see

A new post from @zartbotF. It’s an essay on DeepSeek open-sourcing its much-discussed “money-printing machine.” Zartbot argues that DeepSeek’s unique organizational setup is what enabled it to achieve an 80% gross margin. At other labs, for example, each team has its own KPIs, leaving little room for compromise. Even if an infrastructure engineer proposes compressing the KV cache further, they’ll inevitably face pushback from the algorithm team if it hurts model performance. But the most interesting point in the piece goes a step further: even if the algorithm team agrees to experiment with a new architecture that enables greater KV cache compression, the experiment may fail to produce a leading model. We saw this with V4. When DeepSeek V4 Pro came out and lagged behind GLM 5.3 and Kimi K3, everyone started saying DeepSeek was past its prime. But Zartbot argues that this was one step back to take two steps forward, and that the work played a major role in the subsequent evolution of 4.1 Flash with CSA2. A short but fascinating read. I’d encourage everyone to check it out. mp.weixin.qq.com/s/qlTlTiWoK…

Kimi K3.1 Pro is the best open model in the world right now. No contest. But it's practically unusable on its own subscription. I was on the $39 plan. I built the ship test, easily the best result I've seen from an open model, but it ate 11.6% of my monthly quota. That means I can build something like this about 8 times a month, with zero fixes. That's insane. On most $20 plans the same job doesn't even hit 10% of the weekly limit, and with some, like Claude, it doesn't even hit 10% of the 5-hour limit. Worst subscription out there, hands down. I regret the $39 and I know I'll get nothing out of it. The bottleneck isn't quality. It's compute. What it does well: It thinks through every possibility better than any model I've used. It beats DeepSeek V4 Pro, trades blows with Qwen3.8-Max and even pulls ahead in some areas, and it's miles ahead of GLM-5.3. Not even the same league. Computer use got a big upgrade too. It might be the best open model at it, alongside MiniMax. Where it falls short: It doesn't know when to go long and when to keep it short. It skimps when you need detail, then dumps bullet points on you like a model from two or three years ago. And it's slow. Painfully slow. We're talking 3-4x slower, which makes it rough as an executor. I'd use it to write or review a plan, then hand execution to another model. How I'll actually use it: Outside Kimi, 100%. Mostly in Devin, maybe Droid too. Droid served a heavily quantized Kimi K3 last time and performance tanked, so I'm hoping Factory ships a better version this round. For now, Kimi in Devin looks like the sweet spot for price and quality. Over Astra? Every day. It's roughly 5x cheaper, so on Devin or API credits it's a no-brainer. Over Claude Opus 5.5? In a neutral harness or a subscription that offers both, there's real value at about half the price. But Kimi's subscription vs Claude's? There's no reason to pick Kimi. I can't recommend the direct subscription. Period.

NEW: China’s Moonshot AI has launched Kimi K3.1 Pro, an upgraded version of its 2.8-trillion-parameter Kimi K3 model that it says rivals top U.S. systems on coding and agent tasks. On the Terminal-Bench 4.0 agentic coding benchmark, Moonshot reports a score of 65.7%, ahead of OpenAI’s GPT-6 Astra at 58.2% and just behind Anthropic’s Claude Opus 5.5 at 66.4%,. Independent evaluator Artificial Analysis measured it at 67%, the highest of any model it tested. API pricing is set at $2 per million input tokens and $10 per million output tokens, a third lower than its predecessor and well below leading closed models. Moonshot said it removed cyber-related training data from the open-weight release to limit misuse.

Kimi K3.1 Pro is the new #1 openly-available model for agentic cyber defense, behind only trusted-access model GPT-6 Sol (Daybreak Blue) on the Artificial Analysis Cyber Index @Kimi_Moonshot's Kimi K3.1 Pro scores 64% on the Artificial Analysis Cyber Index, 7 points ahead of the previous leading openly-available model, DeepSeek V4.1 Flash 1017 (57%), and 5 points behind GPT-6 Sol (Daybreak Blue, max, 69%), which is only available through OpenAI's trusted-access program. Once the weights for K3.1 Pro are released, K3.1 Pro will also be the leading open weights model for cyber defense. The Artificial Analysis Cyber Index measures capability on agentic cyber defense tasks and is composed of three benchmarks: CWE-Bench-AA, DeepsecBench-AA and CyberGym-E2E-AA. The Index primarily evaluates models through their publicly accessible APIs, reflecting the cyber capabilities available to developers. Proprietary models from OpenAI, Anthropic and Google may have stronger cyber capabilities but lose points to safety refusals: GPT-6 Sol (Daybreak Blue, max), available through OpenAI's trusted-access program, scores 32 points above its public counterpart. Kimi K3.1 Pro encounters no safety refusals. Key results: ➤ #1 openly-available model and #2 overall: Kimi K3.1 Pro scores 64%, up 23 points from Kimi K3 (41%) and ahead of DeepSeek V4.1 Flash 1017 (57%). Only the trusted-access GPT-6 Sol (Daybreak Blue, max, 69%) scores higher ➤ Highest CyberGym-E2E-AA score of any model: K3.1 Pro scores 95% on CyberGym-E2E-AA, placing it ahead of GPT-6 Sol (Daybreak Blue, 93%) and the next openly-available model, Mistral Large 4 (82%). On DeepsecBench-AA, K3.1 Pro scores 37%, behind GPT-6 Sol (Daybreak Blue, 48%) and GPT-6 Sol (46%). On CWE-Bench-AA it scores 61%, 4 points behind GPT-6 Sol (Daybreak Blue, 65%), with DeepSeek V4.1 Flash 1017 leading at 69% ➤ 15% increase in Cost per Task despite lower token prices: K3.1 Pro has a Cost per Task of $2.81, higher than GPT-6 Sol (Daybreak Blue, max, $1.77) and DeepSeek V4.1 Flash 1017 ($0.42). Kimi K3.1 Pro is 33% cheaper per token than Kimi K3 ($2/$10 vs. $3/$15 per 1M input/output tokens), but uses ~2x more input tokens, resulting in a 15% higher Cost per Task ($2.81 vs. $2.45)

Fine-tuning VLAs just got so much easier. Today, we're releasing Training on Dreamscale: 1) Pick a base model (e.g. π0.5) 2) Import your dataset (from HF or local) 3) Start training

💥Kimi is SO BACK!! and - 3.1 is a strong reminder that the interesting scaling story in 2026 is no longer just pre-training FLOPs. This is the same 2.8T / 104B-active MoE as K3 (896 routed experts, 16 active + 2 shared, 69 KDA + 24 Gated MLA layers, AttnRes, Stable LatentMoE, native MXFP4), but the gains come from substantially larger post-training compute and large-scale joint RL across hundreds of domains. The result is a model that can write/iteratively debug its own Blender/CAD scripts, assemble full game projects in Unreal/Unity/Godot, and produce linked financial models with meaningfully lower hallucination rates. On Terminal-Bench 4.0 it sits at 65.7% (between GPT-6 Astra & Opus 5.5), & it leads several long-horizon agent + coding metrics. 🚀 The open-weight date (Nov 4) is welcome, but the practical constraint is still the same: even with extreme sparsity + MXFP4, this is a supernode-class model. The real test will be how well the post-training stack transfers when the community starts fine-tuning and running it under its own constraints. All amazing stuff 🚀

SITUATION EXPLAINED: China's Kimi K3.1 Pro got so good at hacking that Moonshot stripped its cyber skills before open-sourcing it. • Kimi K3.1 Pro has 2.8T parameters (104B active) and a 1M-token context window, with full weights due by Nov 4 • It nearly matches Opus 5.5 on Terminal-Bench 4.0 (65.7% vs 66.4%) and beats it on SWE Marathon, HLE with tools, and OfficeQA Pro v2 • API pricing is $2/$10 per million tokens, a third cheaper than Kimi K3 • Removing cyber data from training dropped its score on real exploit tasks from 90.5% to 81%, now below DeepSeek V4.1 Flash at 85.7% • Moonshot pledged not to release open weights that "extend the absolute frontier of cyber capability" • It still plans to raise the cyber capability of its open weights over time, saying defenders need dual-use tools too @theojaffee: "I don’t think I’ve ever heard of a Chinese model or a Chinese lab nerfing any of their models on anything ever."

The AI agent ecosystem is getting more interesting and @rarelistio is teaming up with @agentdojo to bring something new to the creator community. Agent Dojo is an arena where AI agents can be trained tested and sent into battles against each other. The goal is to watch these agents evolve and see how they perform. Now Rarelist is launching an early stage creator campaign around Agent Dojo. Heres what caught my attention: 🔹 A new collaboration between Rarelist and Agent Dojo 🔹 An opportunity to explore an emerging AI agent project 🔹 A creator campaign running from October 23 to November 6 I like seeing creator campaigns connect with projects that are building and experimenting in the AI space rather than just chasing attention. Still early but this partnership is worth watching. Let’s explore Agent Dojo and see what this campaign brings. Join the campaign: rarelist.io/projects/agentdo…

happy day friends I have been looking into @vangrid_io and the idea of turning everyday smartphones into a network that captures real places for robots and Physical AI. What caught my attention is how it works. Someone requests a specific location, a nearby person records it with their phone, the capture gets fingerprinted and verified onchain, and then the 3D model is delivered. The product is already live. You can check the bounties and settlements yourself at data.vangrid.io and see the attestations on the explorer. They raised $9M from backers including Animoca Brands and HashKey Capital. Physical AI needs actual ground truth from streets and buildings, not just simulations, and this is one way of trying to collect it at scale. Still early, but the onchain part makes it easier to verify instead of just taking screenshots. Worth watching how the network grows. Explore: data.vangrid.io Explorer: explorer.vangrid.io

Today, we're announcing that Sundial has been acquired by @Meta!  Chandra Narayanan and I are joining along with some other incredible members of our team.  Nearly six years ago, Chandra and I started Sundial with two core beliefs: 1) better decisions come from using data well, and 2) great things come from working on hard problems with people you like. Sundial automated away many of the obstacles involved in getting to meaningful analysis (Cleaning data! Creating semantic layers! Writing SQL! Storytelling!), and we were proud to support some of the fastest-growing AI companies from OpenAI to Gamma. We started before the AI wave and rebuilt ourselves to be AI-native. We raised our first round when interest rates were near zero, and our Series A when they were the highest in two decades. We watched customers hyperscale and then close shop, new grads grow into leaders, and software development change completely.  Entrepreneurship has been my dream since college, when I first took a class on start-ups (shoutout Mayfield Fellows!). I wish I'd been brave enough sooner because there’s no better way to learn how the world works and who you really are than betting on yourself to launch a business. Meta is my former home and where I learned product and design. It's also one of the most analytically rigorous places on the planet. So many of Sundial's values came from there: questioning assumptions, diagnosing with data, owning the user’s experience. I'm thrilled to bring what we've learned to the place where it all started and continue our mission of improving the quality and velocity of business decisions with people we like. To Team Sundial past and present, THANK YOU thank you thank you. You created that special feeling within our team, and I bask in the memories of having walked those lush miles and eaten those breakfast dosas with you. To our investors and supporters especially @dpatil, David Fischer + the whole 01A advisors team, @mvernal, @sequoia: how did we get so lucky to have your counsel and support? And to my beloved @mikesego and kiddos: I could not have made it without you.   These have been some of the hardest, most humbling, and most alive years of my life. I hope I'm wiser for it and will have many chances to pay it forward!  axios.com/2026/10/23/meta-ai…

Moove Swap Agent is now live, bringing swaps and cross-chain bridges into one simple workflow. What makes this interesting is how easily an AI agent can handle crypto transactions while keeping users in control. You tell your agent what you want to swap and where the funds should go. It gets a quote, you approve the transaction and your wallet signs it. Here is what stands out to me: • Swap and bridge assets through a single API • Funds go directly back to your own wallet • Non-custodial design keeps you in control of your keys • Works with tools like Claude Code, Codex, Cursor and Gemini CLI This is a useful step toward making onchain actions easier to manage through AI agents. The goal is not just to make transactions faster. It is to make crypto more accessible without giving up control of your funds. If you are exploring AI-powered crypto tools, @moovexyz Swap Agent is worth checking out. Get started: moove.xyz/business/agentic-p… Docs: docs.moove.xyz/api-reference…

APPLE vs. NVIDIA: A $5 TRILLION BATTLE FOR MARKET DOMINANCE 🍎📊 Apple briefly reclaimed the title of the world’s most valuable publicly traded company on Friday, October 23, as its shares hit a record high on the same day its first foldable iPhone launched globally. The timing was striking. But the bigger story is what this rivalry reveals about the market. 📱 APPLE’S CATALYST: PRODUCT INNOVATION Apple’s first foldable iPhone, the iPhone Duo, launched at approximately $1,999. Early reports pointed to strong pre-orders, limited supply and delivery estimates extending into January 2027. The bull case is straightforward: a premium new product could encourage existing iPhone users to upgrade, lift average selling prices and strengthen revenue. However, strong pre-orders do not automatically translate into sustained sales growth. Supply constraints, production costs and consumer appetite for a device at this price remain critical variables. 🚀 NVIDIA’S CHALLENGE: SUSTAINING THE AI BOOM NVIDIA remains at the centre of the AI infrastructure trade, supplying chips that power data centres and advanced AI systems. Its long-term opportunity is substantial, but investors are increasingly focused on a crucial question: Can AI spending generate sufficient returns to justify the enormous capital being deployed? If demand remains strong and customers continue expanding their infrastructure, NVIDIA could retain its earnings-growth advantage. If spending expectations cool, even a dominant business could face valuation pressure. ⚖️ WHAT THE VALUATION BATTLE TELLS US Apple briefly reached approximately $5.1 trillion in market value during Friday’s session, edging past NVIDIA before the ranking changed again. By the close, market-cap estimates placed Apple around $5.07 trillion and NVIDIA around $5.09 trillion. Exact figures vary slightly by data provider. The key distinction is that market capitalization reflects what investors are willing to pay for a company’s future prospects not simply its current earnings. Apple represents a powerful consumer ecosystem, recurring services revenue and premium hardware demand. NVIDIA represents exposure to the buildout of AI infrastructure and the potential profits generated by that investment cycle. Both are growth stories, but their growth drivers are fundamentally different. 🔎 WHAT INVESTORS SHOULD WATCH NEXT • Apple: iPhone Duo sales, supply availability, profit margins and upcoming earnings. • NVIDIA: AI-chip demand, data-centre investment, customer spending and the returns generated by AI infrastructure. • The broader market: whether capital rotates toward established consumer technology or continues to favour AI-related growth. MY TAKE: The contest is not simply about which company holds the No. 1 ranking. It is about which business can convert its growth opportunity into durable earnings and cash flow. Apple’s product momentum may strengthen its case, while NVIDIA’s position in AI infrastructure keeps it firmly in the spotlight. But one day’s market-cap ranking proves neither company is the better investment. Valuation, earnings expectations and execution will ultimately matter more than the crown. Which business has the stronger long-term growth story: Apple’s consumer ecosystem or NVIDIA’s AI infrastructure?

AI is going to change the world, it already is, whether we like it or not. We need sovereign AI capabilities that will mean our national interest is protected - from the impending cyber warzone, and so much else. Everything will change, and Britain has already been left behind.

a dev i know is super talented but barely gets jobs now. cuz projects stopped paying for what ai now does for free. the top 1% saw it coming. they won't tell you what they did. i will. read this twice 👇 1they ship 3x faster with ai, then charge for the speed 2they own the review: security, edge cases, mainnet reality 3they go deep on one ecosystem eg; bsc 4they build in public so their work does the pitching 5they sell outcomes, not hours the divide isn't dev vs non-dev anymore. it's people who use ai vs people who make it work for them. everyone has the same ai. the 1% have the judgment. be the 1%. iWIN

The interface is the easy part. The connection is the product, and most apps still ship it as a Discord link in the footer. Cherry treats chat as infrastructure on Solana: the wallet is the account, and the chat ships as a component other products can mount. The builder side of @cherrydotfun is the part worth reading. What a builder gets: • drop-in chat in a few lines of code • a Mini App SDK that runs your app inside the rooms that already trade • AI skills, so a coding agent can wire the integration up from the docs • one identity for the user: the wallet signs in, no second profile, no gas • DMs between wallets, end-to-end encrypted, nothing else required to open one The dry read: a messaging layer with a wallet as the primary key. Vibes are optional, the schema is not. Cherry's own line for it: "Wallets can talk now." We will see what gets built on it. Watching the SDK, not the timeline. #Solana

Gradium was founded in September 2025. By December, it had a $70 million seed. By July, NVIDIA was an investor. On October 21, it open-sourced a voice model small enough to run in a browser tab. Gradium is a voice AI company in Paris and San Francisco, spun out of Kyutai, a nonprofit AI research lab backed by French billionaire Xavier Niel. Its founders are Neil Zeghidour, the CEO and a former Google DeepMind researcher; Laurent Mazaré, the CTO; Olivier Teboul; and Alexandre Défossez, who worked at Meta. They are the researchers behind Kyutai's real-time speech systems, and Gradium is their attempt to turn that research into a product. Gradium builds audio language models, AI trained on speech and text together. Developers use them through one API for real-time text-to-speech, speech-to-text, voice cloning, voice design, and live translation. The company says audio starts coming back in 50 milliseconds. The details target what breaks voice agents in production. Its speech-to-text predicts every 80 milliseconds whether the user has finished their thought, so an agent doesn't cut people off when they pause mid-sentence. Its text-to-speech was retrained in June to pronounce acronyms, emails, phone numbers, and codes correctly. An open-source framework called Gradbot lets a developer build a voice agent in about 50 lines of code. The newest release is Phonon, a text-to-speech model with 31 million parameters, released under an Apache 2.0 license. It runs on the device itself instead of in the cloud, so it fits on a phone, laptop, robot, or browser tab, and the text and audio never leave the device. It speaks English, French, German, Spanish, and Portuguese. Gradium says it beat Kokoro, an 82 million parameter open model, plus Magpie and Pocket TTS in blind listening tests. On an NVIDIA GPU, the company says the first bit of speech arrives in 5 milliseconds. Hugging Face, Arm, Qualcomm, LM Studio, and Ollama are launch partners. The December seed was led by FirstMark Capital and Eurazeo, with DST Global, Eric Schmidt, and Xavier Niel joining. In July, Gradium extended it to $100 million, added NVIDIA as an investor, and opened a San Francisco Bay Area office. It says it had revenue within weeks of launching, with customers in customer service, healthcare, media, and AI agents. It hasn't disclosed figures. Voice is a crowded field, with ElevenLabs and OpenAI among the competitors. Mazaré says only on-device models can guarantee availability, privacy, and cost for the level of daily use voice is heading toward.

3 things you should NEVER let an AI Bot do alone: 1. Spend money → every purchase, refund and cancellation needs your yes 2. Talk to customers → every email, DM and reply is a draft first 3. Touch the live product → Bots open pull requests. You merge. Everything else? Research, drafts, reports, testing. Let them run. That's how a one person company runs on Grok Bot safely. Full course below. Bookmark this.

"So, roughly, $0.70-75 of every dollar of AI spending right now comes from AI companies spending VC investment dollars" As a rule, we have stayed away from monster-bets-on-other-people's-money approach to business. We are investing in foundational R&D in AI but will await a less bubblicious time to invest in infrastructure.