Hacker News: tmzt

New comment by tmzt in "MicroVMs: Run isolated sandboxes with full lifecycle control"

tmzt — Fri, 26 Jun 2026 18:50:25 +0000

You can with the newer instances that suport nested VM. There was a recent story about this here https://news.ycombinator.com/item?id=48556561.

New comment by tmzt in "Cyberdecks, going analog, and convivial technology"

tmzt — Wed, 24 Jun 2026 11:37:36 +0000

Personally, I'm repurposing an older HTC phone with slide-out keyboard (the Speedy) with a new Linux port.

New comment by tmzt in "How we run Firecracker VMs inside EC2 and start browsers in less than 1s"

tmzt — Thu, 18 Jun 2026 16:45:32 +0000

Do you see much of a difference between started Chromium instances with the same configuration in terms of the contents of allocated memory? Are they deterministic?

If not, could you template the memory and apply runtime patches (like timers or other initialized values) before releasing the process to run?

Would forcing the isolates to allocate memory better help at all, such as reducing fragmentation making your 2MB page sizes more effective?

New comment by tmzt in "Iroh 1.0"

tmzt — Mon, 15 Jun 2026 16:38:11 +0000

I've been working on a mesh network for private AI models running remotely, controlled by mobile devices (smartphones, tablets, etc.). The mesh is constructed like a piconet, a few devices controlled by a single individual, layered on top of the internet.

How does it support semi-connected devices, intermittent connection failures, etc?

New comment by tmzt in "Stateless Actors"

tmzt — Mon, 01 Jun 2026 00:14:28 +0000

I thought they were itenerant troubadours.

New comment by tmzt in "A Eureka machine that thinks like nature and explores what AI cannot"

tmzt — Fri, 29 May 2026 18:01:55 +0000

Sounds a bit like Transmeta Crusoe.

New comment by tmzt in "Show HN: Forge – Guardrails take an 8B model from 53% to 99% on agentic tasks"

tmzt — Fri, 22 May 2026 19:41:57 +0000

Same. I switched my efforts to a larger Gemma 4 MoE model (26B-A4B) and llama.cpp and started getting meaningful results. I also implemented subagents for querying, determining which object/action to execute, and composing a short title. This is all running on an M4 in approximately 16 gb of ram. Also using Google's native tool calling channels.

New comment by tmzt in "Show HN: Forge – Guardrails take an 8B model from 53% to 99% on agentic tasks"

tmzt — Fri, 22 May 2026 19:35:43 +0000

It's basically restricting what logits are allowed when sampling the model to conform with the JSON (or whatever) shape. It can also cause the model to get "confused" though and doesn't always result in the output you want.

New comment by tmzt in "DeepSeek 4 Flash local inference engine for Metal"

tmzt — Sat, 09 May 2026 00:31:58 +0000

Doing the same for Apple M-series with fused wgsl shaders specifically targeting Qwen3/3.5.

My effort is called shady-thinker and is on github at github.com/tmzt/shady-thinker.

This was inspired in part by Antirez's earlier work with C kernels as well as other efforts to support in-browser LLMs. I've adapted them to Rust and the wgpu library.

Gemma 4 is also the next likely target (with the MTP work) as I'm experimenting with local AI agents.

I'd love to see what you've done to improve prefill and decode even if its not directly applicable.

One difference, I'm using MLX and GPTQ 4bit quants including AutoRound with safetensors as my shader pipeline is pretty much fixed for each model, ggml just adds unnecessary complexity.

New comment by tmzt in "A desktop made for one"

tmzt — Thu, 07 May 2026 15:28:40 +0000

I'm working on this too. I'm building a distributed environment where compute and GPU resources can be on one system, display on another, organized into a type of piconet. So far, I'm working with offline/local AI though running into limitations with it. The goal is to allow for a user to customize the environment with personalized cards powered by open schema databases. The same UI works on mobile devices, tablets, desktops, even TVs and IoT devices (ESP).

New comment by tmzt in "Asahi Linux Progress Linux 7.0"

tmzt — Mon, 27 Apr 2026 23:42:13 +0000

While it's true that early Linux ARM devices where embedded and generally only supported a single configuration, they didn't actually use devicetree.

Originally, embedded Linux ARM devices used a board file with a platform bus and hard-coded device metadata. The bootloader had to pass a machine id which told the kernel which hardware you were running on and which board file to use.

You can see remenants of this in the kernel still, though it's quickly being removed. I'm actually working on a hybrid kernel with the goal of bringing modern Linux support (on an lts branch) to old MSM7x300 devices, like the Evo 4G Shift I intent to use a tmux console/cyberdeck.

On another note, ACPI/UEFI doesn't always give you a clean abstract surface to work with either. ACPI is notorious for building in OS checks into it's compiled bytecode to the point that Linux often lies to it about what OS is running.

New comment by tmzt in "Show HN: OSS Agent I built topped the TerminalBench on Gemini-3-flash-preview"

tmzt — Mon, 27 Apr 2026 19:52:16 +0000

Has anybody thought about encoding AST tokens as LLM tokens, similar to how different words can have different meanings and that's reflected in their embedding?

New comment by tmzt in "Tendril – a self-extending agent that builds and registers its own tools"

tmzt — Mon, 27 Apr 2026 19:25:59 +0000

Also working on something similar and using a dual LLM architecture (small router, larger deep thinker) with offline models, as well as determinisitic skills encoded as TSX.

It's evolved into a mesh-based operating system, gained it's own GPU-based AI library/runtime, and even molted and extended itself to ESP nodes.

Getting closer to a full release sometime in May. For now, pieces are released on my github.

Feels like a good time to seriously rethink how OSs and UIs are designed

tmzt — Mon, 27 Apr 2026 01:15:27 +0000

Article URL: https://twitter.com/sama/status/2048428561481265539

Comments URL: https://news.ycombinator.com/item?id=47916685

Points: 11

# Comments: 2

New comment by tmzt in "OpenAI's response to the Axios developer tool compromise"

tmzt — Fri, 24 Apr 2026 15:48:57 +0000

If that type of scanning works and discovers these attacks, why not make it part of the publishing process?

New comment by tmzt in "Audio Reactive LED Strips Are Diabolically Hard"

tmzt — Thu, 09 Apr 2026 18:06:04 +0000

Using LLM to target an FPGA would be interesting, generate the Verilog/VHDL.

New comment by tmzt in "Audio Reactive LED Strips Are Diabolically Hard"

tmzt — Thu, 09 Apr 2026 17:56:03 +0000

is there a link to the project, the one in the comments won't open on mobile

New comment by tmzt in "Show HN: AdaShape-3D modeler for intuitive 3D printing parts / Windows 11"

tmzt — Tue, 07 Apr 2026 20:40:27 +0000

If it helps I've been doing a lot of projects with TSX (via OCX) as a file format.

It works nicely for these things because it already supports typed parameters and can be treated as a a raw AST without JS engine.

Its also easy to manipulate with the same library.

New comment by tmzt in "Why did Harvey choose a top-down enterprise GTM while Cursor went bottom-up?"

tmzt — Wed, 01 Apr 2026 20:06:16 +0000

Because one is for lawyers? It doesn't seem like a hard choice.

New comment by tmzt in "iPhone 17 Pro Demonstrated Running a 400B LLM"

tmzt — Wed, 25 Mar 2026 18:44:13 +0000

Same here. Then you see SOTA in a browser from Ex0byt, online 10x training (JIT-Lora), TurboQuant (Google), etc. Just saw KV prediction mentioned in this thread, so looking into that too.

I'm adapting all of this to Rust+WGPU with compute shaders if you want to follow along.

See this repo: https://github.com/tmzt/shady-thinker

Goal is Qwen3.5 27b on a Pixel 10 Pro running GrapheneOS.