taf2

CTM | https://ctm.com/ | Looking for great engineers

I remember my parents working for a company in the 90s and the mailroom was one the most common places things went missing… aka theft. I recall my mom had her purse stolen at her cube… theft from within similar to security exploits from within seem to be the most real security vulnerability to be focused on. That sshd that can only be accessed from a few specific ips with the right port knocking is not how you get hacked… at least not normally

Bought the rtx 6000 pros one per month starting in January- since they doubled in price I wish I just used a line of credit back in Jan to buy all of them. For me it’s about keeping internal company content internal. Slack channels etc with tools for teams to use. Triage tools for Zendesk etc.

Spent most of today reworking a rack and rig of gpus for all of our internal ai work… our big server is 8 rtx 6000 pro and 3 psu, I definitely feel I made a mistake not upgrading our wall power to 240v but so far we have multiple 30amp 120v and with 3 PSU uninterrupted power we have been very stable. My big upgrade will be moving to epyc motherboard from threadripper so we get gen5x8 with bifurcation instead of what we are stuck with today gen5x4 due to bios limitations . What has been so encouraging though is first deepseek v4 flash at 200+ t/s for single user and much more in aggregate- now on qwen 3.8 flash next for image support and eyes on glm5.3 flash for some testing … next is realistically considering co location and quiet a sizable loan to scale this to real hardware instead of miner rip vibes

love this and it's timely as I think with AI more of my code will just be native assembly for example: https://github.com/taf2/tic-tac-toe my m series native assembly based tic tac toe windowed game... very portable too you just simply execute codex --yolo port my app to x y z host...

Fun fact you can even ask codex what questions should I ask about this code because I seeing the following x,y,z (image 1), etc…

Reminds me of the children of time books - think it’s the second book with the octopus

i trained another AI on all my codex logs... it's pretty good actually

a quick scan of tech things once or twice daily helps pass the time in boring situations and also catch me up on the latest

vim, ls , cd, grep, cat being close seconds

The one thing that keeps me in codex is that Claude and grok have done all of this work to make the cli tools feel like windows application with mouse etc… I want to scroll back with my terminal history not inside a window within my terminal….

As a business though I can spend 100k and run this model and support about 10 fte- not bad…

Got this up and running at 250 tok/s about 2500 agg tok/s for 16 users and now this morning got https://huggingface.co/webbrain-one/DeepSeek-V4-Flash-0731-V... working - patched it to support multiple images... should post this soon to huggingface - having about gpt-5.6 terra high locally feels pretty good

we got this running on 4 RTX Pro 6000's and for single request we're getting around 250 tok/s we can support about 48 concurrent requests we're seeing around 2400 agg tok/s peaking around 24-31 concurrent users. Model performance feels like gpt 5.4 - mostly using it with pi agent. the only thing i'm missing with this model is vision and i see some folks have done some work like https://huggingface.co/webbrain-one/DeepSeek-V4-Flash-0731-V... but have not yet tried it out.

Llm are great at things that are repeatable. Otherwise they are just like us - great at believable bullshit… a sign of true intelligence

I remember bugs always end up being more expensive then the first time I implemented the feature… so for me always sol always max… I don’t want to pay for a bug later

yeah i can't say if it's 2x or 10x or 100x i don't know. i just know i barely open vim anymore. I still find my rate of development is limited by how much my single mind can hold at any given time. I am however accomplishing with a much higher rate of confidence many many more daily tasks usually by 10am... the limit is usually my willingness to imagine and think critically about an idea with enough clarity to result in something that make sense. It's a different energy then normal development but similar enough.

404 is resolved let the downloads begin

it's a 404 link now

eager to see how it benchmarks on https://deepswe.datacurve.ai/

August 2025 for me but now days I’m much more extreme- it does it all max sol in mornings and then I do other things in the afternoon. I get more done than before with far less stress.

Would love to see the results if they had used /goal or similar

I'm not finding this on huggingface yet is and open model or is kimi now a closed model ?

Not finding anything about this while searching huggingface: https://huggingface.co/search/full-text?q=SWE-1.7 i assume this is another closed source model?

i love the built in AI - i created this tic tac toe game with an AI trash talking opponent -> https://taf2.github.io/ai-tic-tac-toe-trash-talk/

The timescale of this analysis is a big issue IMO- Covid hiring was all kinds of whacked out - with FANG companies competing to hire literally entire graduating cs classes.

this gives me children of time vibes

Best way to make your M series macbook pro feel like a good old fashion intel macbook pro. Run a local model.

when i read "keys" i figured "names" like in my .ssh/config a named host that i access with a key... but listening more it sounds like a new way to do networking over QUIC...

kill -9