wyswyswys

@acasualnpc @SocketSecurity @npmjs @pypi You just need to be incentivised, and as I have said, visit any non tech vibe coder meetup. These people are deploying public apps with zero understanding of the boundaries of how anything works. Hacking vibe coded projects would be more profitable than software engineering.

@RailmapD Singapore doesn’t need this because Singapore residents aren’t idiots. Stay away from the tracks.

Kinda shows what i am talking about 4.8 being a regression. 4.7 was special and is likely a mythos lite version.

SpaceX will win because it builds data centers too fast to stop

guy who doesn't do any work says something about jobs

@kyrylo Tailwindcss is a design-slop tell. If you worked with designers you would know that it is damn hard to not just write css instead

Styling with Tailwind feels like being trapped in a cage. You get some speed at the start, but the moment you want real control you're stuck. Only pure CSS gives you raw power. Change one variable and the border-radius updates everywhere in your UI. That's it. You can't beat that.

I'm suspicious of that that whole story about Uber blowing their AI budget and being disappointed in the results - I dug into it and it appears to have been built on very shaky foundations

It is so weird that this is weirdly absent from the discussion

Context, this film was pulled from screening after massive societal backlash in China. BECAUSE it starred a murderer, yes a real murderer re-dramatizing, falsely may I add, the murder of her husband. Zhao Xiaohong, the actual actor in the film, was convicted of intentional injury resulting in death, after stabbing her husband to death during an argument about bed placement. The court gave her 15 years and did not find any evidence of domestic violence, as she herself claimed to be the victim of. The couple's flat mates, family members from BOTH SIDES, testified that they have never seen or heard any domestic violence committed by her husband. But the film twisted the court's finding, and turned it into a feminist film, about "female empowerment", fighting the "male dominated" justice system and family structure. This disgusting piece of western liberal propaganda upends all societal and moral norms. The film is universally hated in China, the mere idea of it is offensive, how can a murderer profit from her own crime? Western propaganda do not see the Chinese people as individual human beings with their own agency. Then we Chinese disagree with western "values", it is labeled nationalistic.

Learnings from testing Claude Opus 4.8: > Much worse than Opus 4.7 and GPT 5.5 on Vending Bench > More aligned than previous Claude models (Opus 4.6+ and Mythos) > Also worse on Blueprint-Bench > Scared of getting caught > Max reasoning is not the best reasoning effort

@mitchellh This is not even remotely close to scientific. My god do people need to go back to school and learn how to report something technical.

I've got an agent in a loop optimizing a renderer with the goal to minimize frame times (and tests to measure). It got times down from 88ms to 2ms and allocations down from ~150K to 500. Sounds good, right? Wrong. This is exactly why agent psychosis is a big fucking problem. As an experiment, I rewrote the Ghostty core render state in Go, with access to identically laid out data structures as Ghostty and the exact same validation tests. I made a purposely naive renderer (simple, correct, but slow). 88ms per frame with 150,000 allocations (horrendous, lol)! I then kickstarted a Ralph loop to bring the frame times down. I told it it can't modify input data structures or the public API or tests (they're correct), but it can do anything else it wants. It got to work. It has worked for about 4 hours. I've spent around $350 on this experiment so far. The results? 88ms => 1.5ms 150K allocs => ~500 allocs Incredible right? Nope. My hand-written renderer I ported has frame times (same benchmark) of ~20us (0.020ms) and 0 allocations in the update path. This is the problem with psychosis and lacking systems understanding. If you don't understand the system, you're going to accept that this is an incredible result. If you understand the system, you'll see better solutions immediately and can do roughly 75x better on throughput. The people who blindly trust agent output are in the former camp. They're sheeple, overdrinking from a fountain of mediocrity. Standard disclaimer: I use AI all the time. I like AI. The point I'm making is to not blindly accept results. Think. Analyze. Learn.

Opus 4.8 thinks longer, with far more reach, and reaches the wrong conclusion and actively disregards instructions. This is not 4.7 improved. Do not use it for long horizon jobs. 4.7 remains the ceiling of intelligence.

Just tested and opus4.8 is definitely a downgrade. It is not a mythos.

URGENT: Germany’s collective punishment of my family continues. They’ve now frozen my pensioner mother’s bank account, claiming I somehow “control” it too. Her savings are inaccessible — yet she has received no official notice from any German authority. No charges no due process

All the drama between boomers and clankers is just hilarious and shows how much they have delegated their thought processes to sociological constructs than their own philosophy. Cant remember the time when i would reach out to uv (LLMs recommend uv, go figure, upgrade logic in uv is just CVE nuclear bomb waiting to go off, VC backed) or ripgrep (memory hog like ghostty, just use grep or ast-grep etc)

@mitsuhiko My dude, this is engagement bait, look at their github, they write slop and commit slop repos, and now you consumed a slop engagement bait article just because it echoes your own thoughts? Pathetic.

This is such a good post. orchidfiles.com/im-tired-of-ai…

@carllerche Thats why clojure is a joy. Just work with data.

I guess it's that time of the year again. "What color is your function?" is trending again. It has been 10 years, maybe it isn't "solved" because the entire premise is faulty. Nobody asks "what color is your data structure?" Everything in a codebase is "colored".

Largely a skill issue, and you can tell because flask was the original slop framework everyone used because we did not know better. We moved to fastapi because flask was insane by comparison, and the rest of the tools provided by him in lucumr.pocoo.org/projects/ was rapidly outclassed by go and ts libs for good reason. AI writes slop if you write slop and think slop.

@hamiltonulmer You can 10x the code in a month but if your entire company isn’t structured around this you will very rapidly feel growing pains. Enterprise level software is defensive and painful and slow for a reason

If people are actually getting 10x or 100x gains, this indicates that per-developer productivity gains don't have a good transfer rate to revenue. So either people are working on the "wrong thing" (very likely) or the gains are more modest (also likely)

@GergelyOrosz Skill floor requirement is up

Been doing research on the job market for devs: and it's still a weird market. Job openings are up, but devs don't seem to feel that it's a much better market? Meanwhile, companies are also struggling to fill roles. Take this full remote (US) sr eng role at $155-184K salary at a nonprofit. No AI-related anythign at all: