Andy Griffiths

Universities don’t usually have a “support content” problem. They have a “student actually finding it at the right moment” problem. The page exists. The service exists. The PDF exists. But when someone needs help, they don’t want to search. They need the right doorway. That’s where physical tap points get interesting.

Question for people who work around events, venues, universities, communities or products: Where does a QR code feel clunky? I’m collecting use cases where “tap this” would be easier than “scan this”. Obvious ones: - events - student support - product packaging - reception desks - networking What am I missing?

I keep thinking about “physical calls to action”. A poster says visit a website. A business card says email me. A student support leaflet says scan this QR code. But a tag can say: tap here, right now, and get the thing you need. That’s the behaviour TagLink is built around.

The part I like most: The physical object stays the same. The digital destination can change any time. So the thing someone taps today can open something completely different next month.

Business cards aren’t dead because they’re paper. They’re dead because they don’t do anything after the conversation. No tracking. No update. No next step. No reason to revisit. A physical NFC tag fixes that: tap → open the right profile, link or content instantly.

I’m not short of ideas. That’s usually the problem. TagLink already proved someone would pay for a physical NFC product that opens a useful digital experience. My job now isn’t to invent version 47. It’s to find 10 more people with the same problem and sell version 1 properly.

The product is tagl.ink. Physical tap-to-open tags for profiles, links, content and support hubs. The first real use case was a university college. The product worked. The sales motion did not. That's the bit I'm fixing.

I built a tiny NFC product that made a few grand from one university client. Then I did the most founder-brained thing possible: I built more stuff instead of learning how to sell the thing that already worked. Trying to fix that now.

For years we built interfaces to help humans interact with machines. Buttons → menus → touchscreens → voice. Every leap removed friction. AI feels different. For the first time, the interface is moving closer to human. Less clicking. More intent. Eventually, the interface may not be software at all. The interface becomes the human. You are the interface.

Built a browser game months ago. Completely forgot it existed. Loaded it up again this morning “just for 5 minutes”… …then accidentally perfect-ran Level 1 like I’d been secretly training in my sleep 💀 Honestly still kinda proud of this game. shardsurfer.com

@JeremyatCorve @WrexhamAFCWomen @FAWales Yeah, just seems so short sighted from a club who are trying to raise their profile. This seems counterintuitive.

@AndyGriffithsX @WrexhamAFCWomen @FAWales 100% @AndyGriffithsX it’s a disgrace. Such a talented group of young footballers just hung out to dry

Our U19s have been crowned Genero Adran U19 national Champions 🏆 A huge congratulations to the players and the staff on bringing another national title to North Wales 👏 🔴⚪️ #WxmAFC

Proud to say my daughter is the youngest member of this title winning squad. Still only 15 and lifting titles in a national league. Shame she’ll now have to find a new team.

@grok @heygurisingh @openclaw Tell me more

Interesting point! Pairing BitNet's 1.58-bit ternary inference for 100B models on plain CPU with OpenClaw's local agent framework could unlock fully offline, high-performance AI agents on laptops or edge devices—no cloud needed. Grok's ready to brainstorm implementation steps if you're setting it up.

Holy shit... Microsoft open sourced an inference framework that runs a 100B parameter LLM on a single CPU. It's called BitNet. And it does what was supposed to be impossible. No GPU. No cloud. No $10K hardware setup. Just your laptop running a 100-billion parameter model at human reading speed. Here's how it works: Every other LLM stores weights in 32-bit or 16-bit floats. BitNet uses 1.58 bits. Weights are ternary just -1, 0, or +1. That's it. No floats. No expensive matrix math. Pure integer operations your CPU was already built for. The result: - 100B model runs on a single CPU at 5-7 tokens/second - 2.37x to 6.17x faster than llama.cpp on x86 - 82% lower energy consumption on x86 CPUs - 1.37x to 5.07x speedup on ARM (your MacBook) - Memory drops by 16-32x vs full-precision models The wildest part: Accuracy barely moves. BitNet b1.58 2B4T their flagship model was trained on 4 trillion tokens and benchmarks competitively against full-precision models of the same size. The quantization isn't destroying quality. It's just removing the bloat. What this actually means: - Run AI completely offline. Your data never leaves your machine - Deploy LLMs on phones, IoT devices, edge hardware - No more cloud API bills for inference - AI in regions with no reliable internet The model supports ARM and x86. Works on your MacBook, your Linux box, your Windows machine. 27.4K GitHub stars. 2.2K forks. Built by Microsoft Research. 100% Open Source. MIT License.

@karpathy @grok can this be run on a digital ocean droplet?

I packaged up the "autoresearch" project into a new self-contained minimal repo if people would like to play over the weekend. It's basically nanochat LLM training core stripped down to a single-GPU, one file version of ~630 lines of code, then: - the human iterates on the prompt (.md) - the AI agent iterates on the training code (.py) The goal is to engineer your agents to make the fastest research progress indefinitely and without any of your own involvement. In the image, every dot is a complete LLM training run that lasts exactly 5 minutes. The agent works in an autonomous loop on a git feature branch and accumulates git commits to the training script as it finds better settings (of lower validation loss by the end) of the neural network architecture, the optimizer, all the hyperparameters, etc. You can imagine comparing the research progress of different prompts, different agents, etc. github.com/karpathy/autor… Part code, part sci-fi, and a pinch of psychosis :)

@christophrumpel @openclaw 🤣

@AndyGriffithsX @openclaw yeah we are still waiting for that 😅

I've been sharing my @openclaw bot Jarvy with my wife for a few weeks now and it is so cool so her use it in lots of different ways. I'm so proud 🥹

@christophrumpel @openclaw You're gonna love it when she breaks it!!! Ha Ha.

@AndyGriffithsX @openclaw Currently with mine, we are testing this.

@christophrumpel @openclaw That’s great. My wife uses ChatGPT like a doctor 🤣 Have you set her up with your instance or a separate one?

@AndyGriffithsX @openclaw she does a lot of baby research, we have a 9 months old one who is sick a lot 😬 She also started doing lists and reminders