stefanve

so far so good:)

I spent 1500 hours building an app with AI coding assistants. The code compiled and tests passed. But it was full of silent issues: patterns violated, conventions ignored, technical debt accumulating invisibly. This is the hidden cost of AI-assisted development. The code works today. The codebase becomes unmaintainable tomorrow. I built a tool to prevent this, then benchmarked it across 5 different AI models. Key finding: top-tier models don't need help writing working code. They need help writing code that belongs. Full analysis and open source tool linked below.

I wonder how much of the people that are being upset about an option that recognizes landmarks and that is build in a privacy minded way are using chatGTP, windows, google or socials

The writer of the post admits that he doesn’t understand the very tech lite explanation of Apple nor read about Apple AI in general and the way it was setup. A lot of people are upset that (maybe, but unlikely) Apple knows that you made the 10 billionth photo of the Golden Gate Bridge. But continu using all sorts of services, living in sec camera heavy cities etc.

Not an apple fan but from all big tech corporations they have the least amount of interest in mining user data from a business perspective

created a car selection script, that collected data from several platforms, reliability data, occasion adverts, safety data, reviews and specs.

I used it to buy a secondhand car based on several parameters (price, doors, features, reliability etc) and by assigning weight to the parameters

Python in total a day or so of work (including tweaking and adding features) only used it a couple of times to buy a car

there are many types of a wrong answer, and the difference is based on how the answer came to be. In case of BS/Hallucination there is no reason or logic behind the answer it is basically, in the case of LLM, just random text. There was no reasoning behind the output or it wasn't based on facts.

You can argue if it matters how a wrong answer came about ofc but there is a difference

I get the idea but the "science" is based on reports it doesn't look like this has been tested with actual malware. Would be interesting to know how well it works

Also make it OSS and ask for donations. Not sure what your feature earning model is but is seems easy to replicate and as point out several times right now it asked to blindly thrust you

While it sounds impressive from a technical point of view it doesn’t say anything about how much better it is or if it is now truly full self driving.

I think the 17m is only for the 2% of the voting shares the rest is political donation and thus taxfree

you properly mean the average user. The average user would not download this browser. so the user of ladybird might want to live with the shortcomings. That being said , Firefox still exists and is a very good cross platform OSS browser

e91 was a very nice car to drive the f36 is the first car that I modified because it didn't driver that well. changed the suspensions and now drivers in some ways nearly as good as the e91 and in other ways better