I think part of the reason is that interviews have gotten more adversarial as time has gone on. Ie less respect for the interviewee, more of a power dynamic, more careless wasted time in lazy take home assignments, etc.

I wouldn't mind 9 hour interviews to 3 companies if i knew i'd land the job. It's the 1,000 companies that many people experience that gives them pause after so much disrespect through time.

[1]: https://news.ycombinator.com/item?id=49999702

The first real issue where i wanted to leave was the Claudish nonsense. If not for 5.5 i'd be on OpenAI by now.

Is OpenAI worse at airgap/etc than Anthropic/etc or are its models worse at this rogue behavior?

Do we have special benchmarks for agentic systems going rogue in this manner? Sure it's OpenAI atm.. but it could be my grandmas PC next week. Concerning honestly.

The nice thing about math is it can easily plug into a tool, making it even less of a concern.

This happens most though when the speaker doesn't (or care to) understand their audience.

Eg i find effective communication requires expertise in both the subject matter domain but also the reference of the listener. Eg in ELI5 framing, if you don't know what information 5yr olds are expected to know you'll do a poor job at an ELI5.

It often feels like Claude does poorly at both framing the response relative to what it "thinks" the listener knows, but also the prose is... sideways, just weird as you said.

Or at the very least, make more mistakes.

Yea, for clarity this is what i was describing. I use Porkbun's API and the Caddy instance isn't reachable from the internet, only local and over tailscale

I need (and am exploring) custom review tooling to improve this AI->Human code flow. Reviewing PRs were always the hardest part for me in programming. They were often full of the developers decisions and you have to rediscover those as you're reading code for it to make sense[1]. However i find this even more difficult to discover these decisions from AI.

However unlike human PRs we can ask more of AI. Rarely have i had a developer put on a presentation for a PR - but AI could right? AI could produce a guided walkthrough of the code. Not sure if it will help of course, but my thought is we're all stuck in the old "PR review flow" but instead of PRs it's AI - and the volume of them is far greater than anything prior. So i expect we need to tweak how we review, how we get information from LLMs.

[1]: I'm speaking generally, and about larger PRs. Not some small func where you can easily see what it does. Business logic and complex code can be difficult to decipher in PRs, imo.

Semi related, but i've caught myself nitpicking small things from the code of an LLM only to realize i'm putting in more work trying to convince the LLM to make the code the way i want rather than just modifying it myself lol. It's as if i'm trying to educate it in the ideal way, as if it would learn and not repeat the mistake.. but it rarely does of course.

It seems to me that interacting with LLMs has a large pile of cognitive traps for humans.

Intelligence is important but it feels as if the LLM is only capable of conveying information while drenching it in vomit.

Which is exactly what it is like for many trackers these days. They have an "economy" where some type of bonus point system allows you to buy upload credits. You often acquire this by either simply seeding over time, or by uploading special torrents with rewards, eg double upload credits, etc.

Some trackers even double down on this concept and basically only care about long term seeding. Ratio doesn't matter as much if you're seeding 24/7.

As an aside i keep meaning to review Git LFS to see if there's some fundamental reason that it requires a server. Eg could Git LFS write directly to an Object Storage?

I run a lfs proxy at home and it bothers me that it exists heh. Though i don't run ObjectStorage (minio/etc), so i guess swapping out my lfs server for OS wouldn't really net me anything - but still, feels like it was designed first and foremost for the Github API rather than generically for Git users with Git principles.

Really frustrating.

I'd love to know what the hell Antrhopic has done to make Claude's writing so, so bad.

Opus/Fable output these days though is... not enjoyable. It's just really bad. The code quality is fine, but i want information from claude and it's just awful to read.

My biggest problem honestly is that i can't move my day job.. we're using enterprise claude and i'm not sure how much effort it would be to get access to another provider. I should inquire though, claude is really frustrating these days.