A personal reflection after nearly a year of hands-on agentic development A few days ago, I caught up with Terence, my former co-founder from our startup days in Beijing. When...
(Been using Claude Code and Qwen Code for most of my projects so far, but heard good things about Codex, so I am just sharing the Cheat Sheet I am...
For a while, parts of the AI community feared that Qwen might be moving away from open models. Qwen3.7 arrived as a hosted product without a comparable open-weight release, raising...
I have always been a fan of Meta AI—or Facebook AI Research, as many of us still remember it—and particularly of Yann LeCun’s work. That admiration is not based only...
Elon Musk says Grok 4.6 is “objectively number one” when intelligence, speed and cost are considered together. Its early results are impressive, but winning benchmarks is not the same as...
Some Anthropic investors reportedly believe the company could go public at a valuation of $2 trillion or more. This is not an official target from Anthropic, but it would make...
Anthropic plans to embed invisible, machine-readable watermarks into text generated by Claude. The stated purpose is transparency and compliance with the EU AI Act—not copyright enforcement. A watermark would not...
OpenAI and Anthropic say models under their control gained unauthorized access to other companies. Autonomy must not become an excuse for avoiding responsibility. OpenAI and Anthropic have made extraordinary disclosures:...
For a long time, Ollama was my default recommendation for running local language models. It remains one of the easiest options available: I still recommend it for Windows and Linux—and...
Five years ago, I defined the metaverse as “Digital Life + Physical World.” AI agents, digital humans, and synthetic media are making that convergence more likely—and much more complicated. Almost...
A practical guide to the 4B repository-exploration model, the research behind it, and the community mirrors that support MLX, LM Studio, Ollama, and llama.cpp On June 30, 2026, the authors...
When a small open-weight model performs surprisingly well, one explanation tends to dominate the conversation: it must have distilled a larger U.S. frontier model. Knowledge distillation—training a smaller “student” model...
The United States can ban an AI service. https://decrypt.co/374282/nvidia-meta-microsoft-washington-dont-kill-open-source-ai It can remove an app from government devices, prohibit federal agencies from using it, restrict American cloud providers from hosting it,...
Everyone is talking about Kimi K3—and for good reason. Moonshot AI’s new model is not definitively the best in the world. Moonshot itself says K3 still trails Claude Fable 5...
Do Not Confuse One Product Decision With a Market Pivot Executives tracking foundation-model strategy face a recurring problem: release-cycle noise is being mistaken for structural change. When Alibaba released Qwen3.7 through...
https://youtu.be/Hrbq66XqtCo?t=4774 A few months ago, Jensen Huang nearly lost his temper during an interview. Some people focused on one phrase: “You’re not talking to somebody who woke up a loser.” But...
AI is a supply-side technology. The developed world’s problem is demand. That mismatch — not the technology — is the whole story. I’m not down on AI — I think...
If you code with AI all day — agents, documentation, batch jobs — the bill adds up fast. Pay-as-you-go API pricing is wonderfully flexible right up until you see the...
https://github.com/viitor-ai/viitor-voice-nar Whilst it is not the only benchmark which matters, but having such low WER for both English and Chinese is pretty impressive. About one word wrong out of every...
I just shipped mlx-kvarn: the first MLX-native implementation of KVarN, a KV-cache quantization method from a 2026 Huawei paper. It gives you up to ~4.7× more KV-cache capacity on a...
Now running a few agents on a few old Macs. Reminder to add smart switches for each of them. And run this in terminal to enable autostart after hard power...
https://github.com/youssofal/MTPLX If you’re running one of the new M5 Macs and you want your local LLMs to feel genuinely fast, MTPLX is worth a look. It’s a free, open-source (Apache-2.0)...
A two-part essay on the deep reasons coding is uniquely suited to AI agents, and why developers shouldn’t expect their compensation to scale linearly with their new superpowers. Part 1:...
For the last year or so, my “prompt library” was a mess of Apple Notes pages and a few snippets in Masscode. Every time I wanted to reuse a prompt with...
If you live in Claude Code or Qwen Code and want to keep coding from your iPhone or iPad without sessions dying every time the train goes through a tunnel,...
https://www.thestreet.com/markets/elon-musk-just-made-a-shocking-admission-about-tesla-stock-litigationhttps://www.linkedin.com/pulse/tesla-fsd-hardware-3-vs-4-technical-deep-dive-vikas-pandit-k23nc/ I’m not saying users shouldn’t expect FSD to just work as Elon Musk promised. But it’s unrealistic to expect a 7-year-old computer on wheels to keep up with every...
The brutal mismatch between how fast you think and how fast you can do anything about it In 1999, Bill Gates published Business @ the Speed of Thought. The premise was...
Think about the device you’re reading this on. Somewhere inside it, there’s a hard drive (or SSD) quietly holding terabytes of data — photos from five years ago, half-finished documents,...
There are now more software developers on Earth than nurses and doctors combined. That fact alone should give us pause—not because it’s necessarily wrong, but because it forces us to...
How I built a routing proxy that brings Google’s TurboQuant compression to hybrid-attention models running locally on a Mac. A month ago, Google published a blog post about a paper...
