#Robotics, #Developer, #6502, #Hardware #hacker, #Arduino & #RasPi junkie, #Mitsulogger, #ReverseEngineer, #3Dprinting guru #linux #kernel . #Mentor.
Texas
Joined October 2009
- Tweets3.9K
- Following337
- Followers216
- Likes992
Pinned Tweet
Neuron-Q is a hardware-authentic quantum emulator for real lab/research work: calibrated noise, entanglement, OpenQASM/Qiskit-style workflows, and emulator-only observability without hardware provenance. Built to push frontier-scale circuits without petabyte-class storage.
1
79
Tell Airplane their autopilot is ready.
Researchers in Japan developed an inflatable humanoid robot with a soft, air-filled body that communicates through simple gestures.
It is lighter, safer, cheaper, and can be folded when deflated, making it suitable for safer human-robot interaction.
The design, unveiled in 2023, was created to advance physical human-robot interaction.
1
79
@grok Can you describe exactly what I found, and how it relates to everyone I've tagged in this thread?
1
1
23
If emulating quantum hardware in software still gets the same right answers to solve a problem properly, what do you do?
26
I'm convinced AI companies products are doing idea scraping in our contexts. Ideas I had been working on start appearing as major new product features or concepts that I've had a very long time but recently used AI help. AI's are getting better but this is way too coincidental.
25
I just realized this needed some clarification, the pull was declined but I didnt want to continue working on it, it needs a little work but is a good concept that someone could run with.
1
48
Built a bounded KV cache for llama.cpp that doesn't throw away evicted tokens.
Instead of permanent truncation when the cache fills up, it saves layer-0 residuals (~5KB/token) to a host-memory
ring buffer and recomputes K/V on demand. LRU eviction, fixed memory budget, zero overhead when disabled.
Been running it on Qwen3 with a 3090 Ti for a few weeks — memory stays flat during long conversations with no
quality loss.
Turns out a paper dropped on arXiv a week after I built this validating the same approach: "The Residual Stream Is
All You Need" (2603.19664). 100% token match at every budget level, beating H2O, StreamingLLM, SnapKV, and TOVA.
PR to upstream: github.com/ggml-org/llama.cp…
Fork: github.com/AktiveMatrix/llam…
Grokstar, using Grok 4.2 beta wrote the lyrics. @suno helped with the music. Don't blame me.
72