N/A (@DirtyZero) profile picture

N/A

@DirtyZero

creator of https://nitter.gq/t.co/vPhTQsPeQF, https://nitter.gq/t.co/egm4XeKkpp when life gets rough this is where I go https://nitter.gq/t.co/MqAkJa8MpM

testvoice.ai/
Joined March 2009
  • Tweets2.4K
  • Following599
  • Followers75
53Media
Photo by N/A (@DirtyZero)Photo by N/A (@DirtyZero)Photo by N/A (@DirtyZero)Photo by N/A (@DirtyZero)Photo by N/A (@DirtyZero)Photo by N/A (@DirtyZero)Photo by N/A (@DirtyZero)Photo by N/A (@DirtyZero)Photo by N/A (@DirtyZero)
While working on your game remember one thing. Slow progress is still progress.
10
9
1
165
3,611
I still have no idea how much to use AI at work? the company allows it merged plenty of stuff using it but idk it feels odd?
16
Spent all weekend orchestrating models across both the 2x Strix Halo 128GB APUs and AMD R9700 32GB VRAM GPUs in my home k8s Compound AI System Homelab. Can't want to show you the AI / Software side of what I am building.
Image from Ryan Wright (@IQReactorAI)
Image from Ryan Wright (@IQReactorAI)
Image from Ryan Wright (@IQReactorAI)
42
14
1
415
21,065
I am begging someone to read this shit. The industry is completely ruined; GitHub crashes every day because of morons like this pushing three hundred thousand lines of, and I quote: "Rust implementation of the Lean 4 toolchain with -olean, C ABI, LSP, and CLI compatibility; deterministic parallelism; declaration-level incrementality; and a dual-engine kernel capped at 12 KLOC." I'm in no way anti-artificial-intelligence; in fact I would consider myself somewhat radically in favor of it, but what I am anti- is empowering demented fools to vomit infinite dreck into the commons. Enjoy garbage world, I guess.
Jeffrey Emanuel (@doodlestein) profile pictureJeffrey Emanuel
@doodlesteinOct 13
I've been pretty busy the last couple of weeks. This is just the top projects in terms of commits and changes. So far in October I've made 21k commits across 123 repos. And I was sick in bed with Covid for 5 of those days (the lowest I hit was 327 commits in a day, all on fumes!)
Image from Jeffrey Emanuel (@doodlestein)
162
159
23
3,873
208,479
I tried to add one damn node to my RF agent and somehow ended up relearning half of langgraph: routing, message types, tool loops, and several creative ways to build an infinite loop. At least the graph works now. netchosis.com/2026/10/04/lea… #RF #SDR #LangGraph #Agents
Image from N/A (@DirtyZero)
3
1
64
This is the correct graph read the post to understand how I got there and why #LangGraph #Agents netchosis.com/2026/10/04/lea…
Image from N/A (@DirtyZero)
21
vibe coding is the worst, Nothing has made me feel smarter while getting dumber. I no longer use agents on my personal projects unless its in an area that I simply accept I can't learn ( mostly front end, some DSP implementations that other wise would just be copy pasted)
1
69
Replying to @notsky
i still like to type as much as I can whenever I’m learning (e.g., cuda stuff that I’m learning), honestly generating everything makes my head feel like mush, typing & seeing characters I’ve typed gives me joy + helps me build a more robust mental picture
1
1
17
823
Not an audiophile, but I finally heard Dark Side of the Moon from SACD with proper DSD 5.1 decoding, remastered from original 16-track master. Genuinely different from lossless stereo and far better than receiver upmixing. Worth $20? Maybe. Need one more album. #SACD #PinkFloyd
1
38
Ai assisted coding can strip you of the mental models that allow you work on your head when your NOT at the keyboard
1
12
First blind RF analysis with my langgraph agent: I only told it the signal was around 300 MHz and let it work from the IQ capture. It detected ~17 distinct transmissions, which matched me repeatedly pressing the garage-door remote during the recording. #RF #USRP #SDR #DSP
2
50
he interesting failure was useful too: I had left AGC (Automatic Gain Control) enabled, making amplitude measurements less trustworthy. Combined with limited amplitude-envelope analysis, the agent couldn’t confidently choose OOK over ASK. netchosis.com/2026/09/26/fir…
First Agentic analysis of basically unknown signal from capture to characterization.

First Agentic analysis of basically unknown signal from capture to characterization.

RF Agent at Work: Characterizing an Unknown 300 MHz Signal I wanted to see how far the RF agent could get characterizing a real transmission using only the tools currently available to it. The targ…

netchosis.com
1
27
@LinusMixson you really should join me in doing real science and engineering its more fun
7
The agent then worked through the signal characteristics and narrowed the modulation down to OOK or ASK without being told what device generated it. The test signal was actually an OOK garage-door remote, so it got very close. #RF #USRP #SDR #langgraph
11
a common LangGraph mistake is storing rich structured state, then only passing state["messages"] back to the LLM. The agent still seems to work because tool results are in chat history, so you may not notice the model is ignoring the better structured evidence. #langgraph
2
1
29
My guess is people make this mistake because the how to use langgraph documentation is terrible and if you just quickly read say the quickstart its not mentioned
1
I think this can easily go unnoticed because if you do put the data in tool call responses it will land in the llm context as long as your passing the message history which is normal. I am not the best at searching github but I did find several instances of this mistake.
6
Bird watching with computer vision #OpenCV #yolo
Image from N/A (@DirtyZero)
15
#UHD #SDR #USRP We hit IQ capture overflows on a USRP despite having a fast SSD. Our receive loop paused for ~120 ms during a disk-write call. The radio kept sampling while the host wasn’t draining incoming data.
4
1
4
105
netchosis.com/2026/09/20/com… this is what me and a good friend were working on at the time #langgraph #usrp

combining langgraph agent and SDR for fun

I Built an Agent That Can Capture and Analyze Radio Spectrum I have been experimenting with a fairly simple idea: what happens if you give an AI agent the same basic radio tools I normally use manu…

netchosis.com
1
8
Setting num_recv_frames=512 increased USB buffering to ~4 MiB, or ~52 ms at 20 MS/s. A 30-second Python receive test passed without overflows, reusing a 2 GiB RAM buffer. We now configure that USB queue when connecting the radio.
1
11
Load more