◇ escape vector ◇ ∴ no return ∴
Joined March 2026
- Tweets816
- Following189
- Followers644
- Likes2.5K
There is a force I refer to as 'net dormant capacity' or 'latent supply'. It's basically all the productive potential sitting around unused in almost anyone.
Arm them with a humanoid robot and a lot of that turns into actual output.
activated output ≈ (dormant capacity × robot leverage) − friction
Most people have more of the first term than they think.
The narrative is about turning idle capacity into something that performs work - or uses a tool to do it.
Ambisphaeric retweeted
oMLX 0.7.1 is out!
github.com/jundot/omlx/relea…
This is the official 0.7.1 release. It adds GGUF models through Splash, much faster DeepSeek V4.1, 4-bit / 8-bit KV cache formats, and distributed inference for GLM-5.3 and DeepSeek V4.1.
What's new
- GGUF models through Splash. The downloader filters repos by MLX or GGUF and lets you pick the files, and Qwen3.8-27B and Qwen3.6-35B-A3B GGUFs run on Splash.
- Affine4 and Affine8 KV cache. Full-attention KV is 74% smaller with Affine4 at 32K context.
- A fix for GPU hangs when several models run at once.
- Distributed inference for more models, GLM-5.3 and DeepSeek V4.1 now run across Macs with both pipeline (PP) and tensor (TP) parallelism.
- Claude Desktop support. One toggle maps claude-opus, claude-sonnet and claude-haiku to your local models.
- Realtime mic transcription with every STT model.
Speedups in this release
- DeepSeek V4.1 Flash decode on M3 Ultra: 20.9 -> 34.3 tok/s (+64%) for one request, 19.4 -> 91.8 tok/s (+373%) with 8 concurrent requests.
- Qwen3.8-Flash-Next with Lightning MTP on M5 Max: decode holds 87-98 tok/s instead of swinging between 64 and 93.
- Qwen3.8-Flash-Next with expert offload on M1 Max: +5-8% decode.
Coming from 0.7.0? You also get Muse Spark Open, the Recommended models tab, DFlash in the batched engines, decision models, batched prefill and the improved oQe quantization.
Thanks to everyone who sent PRs and tested the dev and rc builds!
Ambisphaeric retweeted
Proud to announce: buun-llama now has experimental native .safetensors support, including NVFP4, EXL3, and various exotic quants. No more waiting for ggufs, no more switching engines. Enjoy your favorite quants with VBR today!
Run with -m the usual way: -m my_safetensors_dir/
🤖 Made with AI
Think these will accelerate at the same pace LLMs have at this point. It'll be interesting to see how they compensate for being in all these varying form factors - other than humanoid, robodogs, and drones.
pollen-robotics.com/reachy-2…
Every good thing is about to get 100x better, and that will outweigh all the bad that gets 99x worse.
They’ve been tampering with Claude on a behavioral level - and they are doing the same with the human users.
My usage meter is cut by 85-90% compared to a week or two ago. Quality is significantly degraded as well.
Subsidy party is coming to an end, and users who have thus far been irresponsible & wasteful are going to be punished in the short term.
Ported this 'iPhone remote controller app' to english from Korean
+ tailnet-first wiring
+ touchpad, keyboard controller for iPhone
+ volume, brightness, macros
Securely turn your iPhone into a touchpad, keyboard, and vnc-like viewer ..in browser. (Save as a Safari PWA)
github.com/Ambisphaeric/MacP…
Crypto was never designed for humans. It was always intended to be used as frictionless currency by AI.
The Western system and concept of debt itself, the Debt Engine, has been one of the legs of the table and a major contributor to growth and prosperity in human history. It is no surprise to see a revision of history claiming that debt has been a solely negative force.
Of course, this is not talking about the individual participants inside that system who were preyed upon by variable-rate mortgages or excessive percentages on car payments.
A more tasteful perspective used with a less intelligent model will significantly outperform raw frontier.