Kirk Patrick

What level of stress and disappointment have you experienced with AI when it deviates from its objective? Has this happened frequently with recent updates like Sol 5.6 or Fable/Opus? Rate it on a scale of 1 to 5, with 5 being the most disappointing compared to previous models.

We've open-sourced MoonEP, our high-performance communication library for distributed MoE workloads. Built to make expert-parallel communication more efficient at scale, MoonEP helps reduce communication overhead in large MoE training and inference systems. Explore on GitHub: github.com/MoonshotAI/Moo…

Releasing the model weights and technical report of Kimi K3. Kimi K3 is our most capable model: a 2.8T MoE model with native visual understanding and a 1M-token context window. New model architecture: 2.5x the intelligence per unit of compute, not just more params. Alongside Kimi K3, we're opening up more of the stack behind it — high-performance attention kernels, MoE communication library, and infrastructure for running agent environments at scale. Model weights: huggingface.co/moonshotai/Kim… Tech report: github.com/MoonshotAI/Kim… Tech blog: kimi.com/blog/kimi-k3

For my first post, I’m sharing a letter @NVIDIA signed on why open models matter. AI will transform every industry, power every company, and be built by every country. Open models strengthen safety and cybersecurity, accelerate innovation and diffusion, and enable sovereignty. The world needs both frontier closed models and frontier open models. images.nvidia.com/pdf/Open-Weigh…

Announcing OpenWorker! An open-source agent that doesn't just chat with you, but delivers finished work -- like hand you a polished document, send a slack message, or update a calendar entry. Ask it to prepare a customer brief, untangle your calendar, draft a report, or triage a Slack alert. It works across your files and everyday tools, produces the deliverable, and checks in before doing anything consequential. OpenWorker runs on your Mac, with Windows support coming soon. It does not lock you into any one model. Bring your own API key and run it with GPT 5.6 Sol, Claude Fable, Gemini 3.6, an open weight model (like Kimi, GLM, DeepSeek, Inkling), or Ollama to keep your data local. Your data does not leave your machine except through an LLM provider and integrations that you choose. @rohitcprasad and I are building OpenWorker because AI coworkers are an important way to get work done, and we want there to be an open, privacy-preserving, model-independent option. Check it out and let us know what you think! Try it out: openworker.com (requires your own API key) Source code: github.com/andrewyng/open…

We landed on almost exactly this architecture a bit over a month ago: unified Postgres embeddings, hybrid retrieval, reranker, and MCP serving LLM-free retrieval primitives with the client orchestrating. Where we went further: - Append-only canonical store before indexing, so the whole index rebuilds without depending on source systems - Formal provenance on every derived artifact (derived-from, source version, model, prompt, date), with source vs derived queryable and derivative indexing as explicit policy - Zero LLM calls in the query path. Retrieval-only worker; synthesis delegated via MCP to the assistants users already pay for - A verified output layer that makes hallucination inexpressible, not just unlikely. The retrieval contract is empty-never-nearest: outside the validity window the system returns nothing instead of the closest plausible match. Absence is a well-defined, provable result over a bitemporal canonical store, and abstention is typed (Found | Gap), gated by byte-exact verification against the source, outside the model. Most systems make hallucination improbable; this makes unsourced claims impossible to sign - Governance via the ARSIA Protocol, our open standard aligned with the EU AI Act and GDPR: query-time ACLs against the IdP, PII protection, EU data residency, auditable corpus - Edge on Cloudflare, data layer on dedicated OVHcloud servers. Fully European stack, replicable by any company without an inference fleet The result is a complete stack, closed at every layer: ingestion can't contaminate (canonical + provenance), storage can prove what existed and when (bitemporal invariants), and output can't assert what it can't verify. Correctness lives in the data layer, not in model behavior. And the entry cost is almost embarrassing: a small company can run the whole thing end-to-end on Cloudflare alone (Workers, R2, D1, Vectorize, Workers AI). The two-person deployment runs on roughly $5-10/month. No seats, no per-user pricing, no GPU fleet. Convergent evolution is the best validation there is.

2026 ✍️ Spain #FIFAWorldCup

⭐️⭐️ ¡¡𝗦𝗢𝗠𝗢𝗦 𝗕𝗜𝗖𝗔𝗠𝗣𝗘𝗢𝗡𝗘𝗦 𝗗𝗘𝗟 𝗠𝗨𝗡𝗗𝗢!! #VamosEspaña | #CopaMundialFIFA

Spain are #FIFAWorldCup Champions. 🏆

So ChatGPT can just build and publish websites now. Hosting included, visual annotations, even plugins to make the output better. Watched the demo and my first thought was: a Lovable replacement? But looking closer, the demos are all front-end stuff. Landing pages, a little game. No auth, no database, no real backend. That's still Lovable territory. The catch: most of Lovable's ~$500M ARR comes from people building exactly this kind of thing. Simple sites, storefronts, prototypes. Enterprise is only ~$20M of it. And they're reportedly raising at $13.2B betting that growth keeps going. Not saying Lovable is dead, far from it. But when the app everyone already has starts doing 70% of your main use case for free-ish... that funnel gets leaky fast. youtube.com/watch?v=8zH_O7…

EU average corporate rate is 21.6%. That's below the US (25.6%) and the world average. Highest in the bloc is Malta at 35%. Ireland and Cyprus sit at 12.5%, Hungary at 9%. The "EUSSR" hosts ASML, LVMH, SAP, Airbus, Novo Nordisk and nearly every US big tech HQs its European ops there. Comparing a 450M-consumer single market with private property and competition law to the USSR isn't an argument, it's a tell that the real numbers don't back you up.

@kirkfps @hoogyoutube there are no enterprises in the eussr, only charitable work by honorable men that decides to risk their capital, their time and their effort in order to advance civilization even if it costs 70% in taxes and absurd regulations

The Brussels effect is working. Being first to regulate AI is working. Europe's top political brass is telling Anthropic they can always just stop developing their models.

It's revealing to watch people who, once established, suddenly clamor for regulation. Open competition for you, regulatory protection for me. The design confirms the diagnosis: a body funded "mostly by industry," benchmarks "developed in consultation with Frontier Labs," approval as the price of access to the US market. The incumbents pay the referee, coach the referee, and the whistle becomes a barrier to entry. And what justifies pulling up the ladder? A premise that is pure speculation dressed as urgency. As @ylecun has said, LLMs are fundamentally incapable of reaching AGI: no world model, no persistent memory, no planning, just next-token prediction. The essay even admits "nobody in the world knows for sure what is going to happen," yet demands an entire regulatory apparatus built on that uncertainty. If the premise falls, the edifice falls with it. What survives the collapse is the part that actually bites. By covering models "open or closed," the framework makes open frontier development structurally impossible: no community project survives a 30-day pre-release gate. Yet open models are the only real protection against the concrete risk, which is not a rogue AI but a handful of companies controlling the planet's cognitive infrastructure. The proposal quietly strangles the hedge while pointing at the hypothetical, and its provision to "coordinate a slowdown among Frontier Labs" is, under any antitrust reading, a cartel with a safety preamble. All of this flows from one foundational error: regulating the technology instead of its applications. Nobody regulates mathematics or compilers; we regulate fraud, discrimination, defective products. Regulating benchmark scores means regulating an abstraction, and abstractions are easy to define in whatever way keeps the club exclusive.

@BraedendotTECH Tokenmaxxing

GPT 5.6 SOL is funny You tell it to write some code You ask it to review that code It finds bugs in the code it just wrote… It fixes it… You ask again to review It finds more bugs… And it never ends 😂

Analog politicians governing an increasingly digital world: there's no way it can work.

You takes an ambiguous news item, inflates into a "historic signal," dismisses competitors and ends with the pitch. There was no announcement that Starbucks is "moving off IBM and Microsoft." What exists is a leaked internal presentation, reviewed by Bloomberg, showing that Starbucks is developing its own alternatives to two specific tools: a Microsoft system that monitors inventory and an IBM product that manages maintenance. Moreover, that software could roll out by the end of next year, depending on test results. In other words: an experimental project, narrow in scope, conditional. Attributing daily drops in IBM and Salesforce stock to a leak about Starbucks is a post hoc fallacy. Worse: Salesforce doesn't even appear in the story. One company testing the replacement of two tools becomes "the largest companies in the world are done paying for software." LoL Starbucks is not abandoning big tech: its "Green Dot Assist" virtual assistant, built on Microsoft's Azure OpenAI platform, is still expanding. Have you ever tried building a corporate brain, not one that runs locally on your machine, but one for a mid-sized or large company? Because the difference isn't one of degree, it's one of kind. Running something like Karpathy's wiki locally works great: one user, data that's yours, zero permissions to manage, no consequences if it leaks, you're just talking to yourself. That's why the demo is so convincing. The personal use case eliminates, by construction, everything that makes the problem hard. In a corporate setting, every one of those shortcuts disappears. Anyone who thinks "you just index everything" has never sat through the meeting with the CISO. Source: bloomberg.com/news/articles/…

Starbucks spends $400 million a year on software. Yesterday they announced they're moving off IBM and Microsoft to build their own custom systems in-house. IBM dropped 3% and Salesforce dropped 4% on the news. And honestly this is, unequivocally, the biggest signal I've seen since OpenAI and Anthropic launched their consulting arms back in Q1. The largest companies in the world are done paying for software that half fits how they work. We saw this coming about a year ago. Moved everything we build off Airtable and low-code tools and went fully custom. Already paying off, and it's only going to compound from here. This is the opportunity right now. You get all of a company's data into one system. You build out a single operating system for the entire business. You cut out bad, redundant processes. Then you layer AI on top of it, under the correct processes. That's the core of AI consulting. Helping companies actually operate better. There are a lot of fly-by-night offerings circulating right now when it comes to Ai Services. For example, 'second brains'. Throwing scattered data into a second brain while the processes underneath stay broken does nothing. The companies who will absolutely destroy their competition over the next 5 years are rebuilding how they work from the ground up. Starbucks is showing you what other companies will be doing over the next several years. Your job is to position yourself to facilitate that process for as many companies as you can.

@sama Can't blame us. Humans need plenty of vitamin D to survive! 😎

makes us happy to see people love 5.6 sol so much!

@other_europe The real logic:

This is a problem of mindset. They're always saying: workers in China don't have guaranteed rights, companies can't compete, etc. We're not in China. China's problems belong to China. This mentality prevents us from finding better solutions. We're always blaming others. And it's important to emphasize that the current problems were choices made by the European Commission itself. Just take a look at how current policies contributed to the Volkswagen in Germany. Four factories closed. ~100,000 people unemployed. Brussels regulates without understanding the long-term impact this has on businesses.