Nils Durner

Software Engineer | 🏗️ LLMs & Agents | 🤝 Digital Trust

[Nils Durner @ndurner]

@thsottiaux Move chats between projects/folders. I often start experiments in /tmp, then want to continue them in a proper project folder with the context preserved.

[Nils Durner @ndurner]

@BoyuanChen0 This is a pretty reliable case with hand/arm breakdown.

https://t.co/wmLcFzojUv

https://t.co/LQZVRCW0ja

[Nils Durner @ndurner]

@theSamPadilla @emollick Google AI Edge Gallery app is perhaps most straightforward. I found it overly cautious about what model can be run on a particular device, though.

[Nils Durner @ndurner]

@karpathy The latter used provider-supplied Agentic Memory indeed - and I wasn't happy with it. Your new Wiki approach appears like something to try instead. Thanks!

[Nils Durner @ndurner]

@karpathy #1: https://t.co/240Y8LcCVz
#2: https://t.co/nhrFBEgwWZ

[Nils Durner @ndurner]

@PITTI_DATA EuroEval attempts to measure European language proficiency, and I have concluded for myself that there *is* a language gap, even for supposedly high-resource languages including German. (Dan Saattrup posts updates regularly on the blue jobsite).

https://t.co/6HmRf3kC2Z

[Nils Durner @ndurner]

@dkundel The auto-translated German version is a tough read: wrong terminology, convoluted sentence structure. Also, couldn’t figure out how to switch - tried URL guessing to no avail

[Nils Durner @ndurner]

@Miles_Brundage "New project" still shows up in the sidebar (never used it, though). Version 1.2026.027 (1769832365)

[Nils Durner @ndurner]

@simonw Could the research attention that @margaretstorey calls for be in the management theory that @emollick refers to? What would be the conclusions from that? HUMANS․md, including what bits?

https://t.co/n6YT75UpPt

[Ethan Mollick @emollick]

As a business school professor, its striking that a lot of the AI folks on this site, as they increasingly delegate authority to coding agents, are re-encountering the basic problems that underlie management theory and practice. Many delegation problems are old & well-understood!

[Nils Durner @ndurner]

@embirico It suffers from the same connection issue as Codex CLI: "stream disconnected before completion". Has been reported by others through Github issues, but hasn't been fixed despite issues closed.

[Nils Durner @ndurner]

@AliSananS @ajambrosino It's Electron.

[Nils Durner @ndurner]

@OpenAIDevs Is there a way to add minimal reasoning to the model selector, like in VS Code?

[Nils Durner @ndurner]

@OpenAIDevs Do automations run only run when the Codex app is open or are they also active on a system background level?

[Nils Durner @ndurner]

@petergostev @dkundel Yes. Primarily so it can look up niche APIs, but I really had it always-on before. Didn‘t hurt, perhaps less rifling through local libs that way - but haven‘t compared.

[Nils Durner @ndurner]

@thorstenball Invitations from people I don‘t know got less once I switched the Connect button that is default to Follow via the settings. I accept connection request from people met IRL or had meaningful conversation with online.

[Nils Durner @ndurner]

@xeophon Blog: https://t.co/x3HDiyYlUe. Supports just English 😕

[Nils Durner @ndurner]

@petergostev There is a „correlation is not causation“ point, though, that goes back to the original gpt-image-1 release. I unfortunately forgot who made this. I think they said that this could simply be for „entertaining“ users - by running a faster model? - and isn‘t really how it works

[Nils Durner @ndurner]

@benvargas @diegocabezas01 @OpenAIDevs I got this error with the Images API (OpenAI.images.generate), though. Python package is openai == 2.13.0