Tim Janik

Working on jj-fzf (Jujutsu TUI), Anklang (DAW), Imagewmark, Audiowmark;

[Tim Janik @TimJanik]

CMSCOUT 0.2.0 is out! 🚀

🔍 Code-motion aware diffs for TS/JS/TSX/JSX, Go, Bash, C & C++
📦 Matches functions, classes, methods, imports, macros, JSX across renames/moves
📊 Per-block similarity + change summaries
⚙️ Git external diff support + --summary

#100DaysOfCode

[Tim Janik @TimJanik]

Agentic tip:

Next time your LLM messes up, tell it:

"Turn my rejection into a lint rule."

It might just manage to do it *and* clean up every related bad pattern too. 🤯

[Tim Janik @TimJanik]

CMSCOUT 0.2.0 Release Notes & Downloads 👇
https://t.co/JtUd9lDi0Y

#DevTools

[Tim Janik @TimJanik]

CMSCOUT 0.2.0 is out! 🚀

🔍 Code-motion aware diffs for TS/JS/TSX/JSX, Go, Bash, C & C++
📦 Matches functions, classes, methods, imports, macros, JSX across renames/moves
📊 Per-block similarity + change summaries
⚙️ Git external diff support + --summary

#100DaysOfCode

[Tim Janik @TimJanik]

@bnjmn_marie Are there any properly quantized GGUFs with 16bit routers that you would recommend?

[Tim Janik @TimJanik]

@esrtweet Loud != Majority.

[Dr. Daniel Bender @drdanielbender]

Two weeks did pass super fast again and it is time to discuss AI news and lessons learned / best practices in a space. 👇 https://t.co/hrZNJtt4ml

[Tim Janik @TimJanik]

@Elev_30 @btibor91 https://t.co/eDk5Xuxr0C

[Tim Janik @TimJanik]

@Elev_30 @btibor91 Here is a good video explaining the likely mechanisms and to what extend text watermarks may survive attacks.

[Tim Janik @TimJanik]

@sudoingX GLM-5.3-Flash on AMD 495 at 160GB, well priced boxes are starting to appear...

[Tim Janik @TimJanik]

@MiaAI_lab @XiaomiMiMo There surely are some cute or interesting ones.

E.g. Candlelit Choir, Ripple Tank or Weather Vane are really nice to play with.

You have just been SOTA spoiled by Astra and Opus-5.5. lately, think back just 3-6 months ;-)

[Tim Janik @TimJanik]

@neural_avb I have noticed the ordering problem as well, doesn't raise confidence in the scores.

However, wouldn't Qwen3.5-0.8B-Base be a more promising base (newer arch, better benchmarks)?

[Tim Janik @TimJanik]

@Stamatiou Configure your LLMs as multiple OpenAI-compatible providers in opencode.
Then use t3code with the opencode backend, which lets you pick repo, provider, model, branch, worktree (and machine for multiple instances) for a task.

[Tim Janik @TimJanik]

@SapoLollo @sudoingX NP.
Then you probably haven't seen this project:

https://t.co/RH6evNmP25

Claims 26.5s per token on 8.24GB *RAM* (not VRAM)!

And no, I didn't bother to verify ;-)

[Tim Janik @TimJanik]

@drdanielbender To play with Jev like inference at home:

https://t.co/ZhRJb1dLjh

[Tim Janik @TimJanik]

@MiaAI_lab Hm, no synthesizer / epiano page this time ?

[Tim Janik @TimJanik]

@xuandongzhao That is exactly the mission of https://t.co/LWzqIpNtvX

Human provenance attestation instead of marking AI generations.

[Tim Janik @TimJanik]

@iampriyanshu23 @opencode In opencode, you can select "Muse Spark 1.3 Contributor OpenCode-Go" with thinking levels
minimal low medium high xhigh.

AFAIK, "Max" is a non-contributor tier, not offered by @opencode atm.

[Tim Janik @TimJanik]

@keithofaptos Explain….

[Tim Janik @TimJanik]

@WolframRvnwlf I tried the OpenAI 23€ tier. The 5h limit feels hostile:

I am barely touching the weekly cap, yet forced to babysit and ping the agent every 5 hours. 😠

With Grok-4.7, is SuperGrok comparable in quality, but without the 5h nonsense?