krig
Ornamental hermit and mushroom fancier.
Profile goblin by @heyheymomo
#Music as Oferlund. #Podcast at Kodsnack (swedish lang). Runs a #SmallTech software business named Ziran. Interests include #Climate, #TechEthics, #Music, #MusicProduction, #FOSS, #Photography, #Gardening and programming languages. #Antifascist #Anarchist #Butlerian #Luddite.
There is a ”Technical Program Manager at Google” posting LLM-generated slop articles about concurrency to lobste.rs and HN getting onto the front page, and it annoys me on several levels. The articles are plausible-sounding (to a casual reader) but wrong, which is bad enough. But I can see how to an unscrupulous person this kind of thing makes perfect sense. They use the slop to boost themselves and their careers, and the risk is minimal since someone has to actually understand things like Erlang message passing and concurrency to notice the factual issues, and be familiar enough with the flavour of LLM-generated bullshit to notice the confident-but-wrong patterns.
The suits at Morgoslop Slophub are starting to feel the consequences of wanting too much. They delved too greedily and too deep. More code, more productivity, more content. More is more, as Yngwie says. Well, you got what you asked for. Your product is shit. You are drowning in it. Your lungs are burning. It's in your eyes. No one can help you now.
Feels like we’re stuck in a tech news Groundhog Day just recycling the same news over and over.
- A new open source project turns out to be slop.
- An old open source project adopts slop-based development practices.
- A middle-aged white guy falls in love with his AI girlfriend which he names ”Lucia”.
- A paper is published showing major frontal-lobe damage after minutes of AI usage.
- A Silicon Valley tech giant claims massive gains using AI while laying off half the staff.
- Some cloud service is having an extended outage.
”Here, through a series of randomized controlled trials on human-AI interactions (N= 1, 222), we provide causal evidence for two key consequences of AI assistance: reduced persistence and impairment of unassisted performance.”
”Our results are notable on two fronts. First, while concern about AI-induced deskilling has grown, prior evidence has been largely correlational (Budzy´ n et al., 2025; Gerlich, 2025) or limited to small samples (Kosmyna et al., 2025; Shen & Tamkin, 2026). Here, through a series of randomized controlled trials on human-AI interactions, we provide the first large-scale causal evidence of this effect. Second, we demonstrate how AI can result in loss of motivation and persistence. A rich body of literature in cognitive science and education has shown that the capacity to regulate effort and persist through difficulty is foundational to effective learning, and is among the strongest predictors of long-term academic achievement, workforce adaptability, and resilience (Metcalfe & Mischel, 1999; Duckworth et al., 2007; Maddux, 2009; Metcalfe, 2009; Andersson & Bergman, 2011; Bjork et al., 2011; Kapur, 2014; Guiso et al., 2016; Mooradian et al., 2016). Our results suggest that AI assistance erodes precisely these capacities. People do not merely become worse at tasks, but they also stop trying.”
One question that has been on my mind since I forked Zed is why it's such a huge project. It's a text editor. What does all that code really do?
I still don't know if I understand why. I have looked at a fraction of the code at this point. What does it all do? Fuck if I know. But here, take a look at this for example...
Python support in Gram (forked from Zed) is mainly provided by python.rs, a single file containing 2865 lines of code (as of today). One part of that code is a function which lists virtual environments available to the editor. The code to actually discover which venvs are available is handled by a dependency (the PET library from Microsoft) but sorting the venvs in order of importance is handled here:
It's a single call to the method sort_by on a variable confusingly named toolchains, containing a list of virtual environments. I have been trying to make sense of it for a while. Like, one sort criteria is filesystem distance, which kind of makes sense. Wouldn't it make sense to primarily sort by filesystem distance, and a second sort by Python version? Instead this is primarily sorting alphabetically by executable name, it's looking at multiple environment variables for priority, it's sorting alphabetically based on conda prefix(?)...
Maybe I'm missing something, maybe I am not galaxy-brained enough to sort virtual environments with the big fellas. But I then compare this to helix which has no special handling for virtual environments, you just configure the LSP to use the venv you want to use, and I feel like maybe it didn't have to be this complicated?
Liked them so much I tried making a couple of variants for Gram.
https://codeberg.org/krig/gram-crt-theme