Highly recommended, I thought it was just the usual recap of the same few allegations but it's so much more. Sad this slid off the front page because I assume most people that.
Same. I'm doing something security related right now and I swear it was smarter than this before. Right now it keeps pushing me to just allow all connections, or leave a password in plain text, or other insane suggestions. Maybe placebo, maybe bad timing though, but it feels ridiculous.
Edit: Just to add some more ridiculousness:
> The last squash stalled because it opened an editor to merge descriptions. I've stopped it. Checking what landed:
I was using jj with Claude for a while now, suddenly it forgot how to squash? It also started doing that stupid thing where it edits files through python scripts so I don't see the diff during manual mode. This is like 3.5 levels of stupid.
If this is in Claude Code, check whether your session got switched to Opus 4.8. After a cyber safeguards refusal it continues on claude-opus-4-8 with a one-line notice. I saw this on ordinary auth-token fixes, in 5 of 60 runs last week.
That is a setting. The first time you get refused it offers to stop or switch to 4.8 and if you choose 4.8 it automatically downgrades silently whenever it happens again. I just shut it down and switch to Codex when this happens.
I would also agree that 5.5 feels dumber the past week. More weird logical problems. it's not as dumb as 5.1 or 4.8 but it feels like it regressed.
I was always uncomfortable that it's on by default and surprised anyone uses it it. It's also more cumbersome compared to just using a key. I feel like using keys is the password manager analogue for SSH.
This always seems bonkers to me. I think SSHD should give you ONE password free connection of setup to copy your key over and then it just shuts the door behind it.
Not sure why you're getting downvoted as from security standpoint that would be the right thing to do; but actual good server providers take it up a notch and allow you to upload the public key through their admin panel during server set up and the password you get only works locally for sudo etc. There's no reason why a fresh Linux setup should ever offer a publicly available SSH server with password based auth out of the box.
Yeah, the price is so high and things are moving so fast I don’t wanna risk committing to something that becomes obsolete in a year. I wanna see the rumoured M7 first. Hopefully hardware demand doesn’t make waiting a stupid decision…
Waiting is never a stupid decision because you can always save a good deal on apple refurbished. I got a mac on bestbuy refurbished a couple years ago shortly after it was released. It had like 4 cycles on the battery and I saved hundreds.
I was in the same boat, but then I tried out Pi which works with the OpenAI subscription and now I prefer it more. I wish Anthropic let you use a custom harness with a subscription but I heard people get banned and don't wanna risk it.
even if claude is nice these hostile tactics from anthropic make me refuse to use their tooling. codex works fine and openAI i havent seen banning ppl for using their API.
there is something in there, and im quite sure someday it will bubble up to the surface and people will be sour. people who use claude that is.
yes, its just a hunch. but i can see these tactics, and u cant tell me its just corporate policies that were whoopsiedaisy. They are getting more from people using claude code than other harnesses, enough to ban and fearmonger people into not using those other harnesses.
the reason will be $ related. Likely their tooling wastes tons of tokens.
now, codex is also a bit wasteful ofc. (very much). but they dont mind u optimizing and rolling ur own.
kinda fun reading last slop report posted here after this which shows opus5.5 higher setting than medium is 'harmful' which means that if u run it higher, it will cost more, and also need more cycles. cant say if its exponential scammery but pretty sure is close :')
I am surprised how many people like the actual writing part, it always felt to me like doing micro in an RTS game. I know a lot of people love that too but I always played RTS games to feel like a general. But I'm still in the camp of reading every single change the LLM does, so I feel like I get the best of both worlds, but maybe that will end up being a waste of time soon. Like if the job ends up being just writing and reviewing MD files for the agent to digest I'll hate it too.
Edit: I really don't understand the down votes, what did I say that's do bad?
It's funny how differently people see things. I find reading/reviewing LLM generated code to be much more difficult and tedious that writing it myself.
Sure, I guess, but I've yet to see an LLM generate code that I'd consider either easy to read or particularly well-written, regardless of who was driving the agent.
And even if it does end up being good code, if it's something non-trivial, having a reasonably deep understanding of what it's actually doing is quite hard simply from reviewing the code.
I just don't get how you can properly reason about the code if you're not involved in writing it, though.
I keep giving AI-assisted programming a serious try, but time and time I land in the exact same place: reading the code is more exhausting than writing the code, _and_ I end up with a far worse understanding of what is being built than if I had just written it myself. The fact that LLMs tend to write _way_ too much code certainly doesn't help.
And what makes me sad is that, if you follow this line of inquiry, you eventually realize that capital isn't challenging the value of humans _writing_ the code, but rather the value of humans _understanding_ the code. Because, as I've experienced many times when talking to those how have completely bought in on the hype, the answer is always to just "let the AI handle that." AI writing too much code? Have it summarize it for you. Your understanding of the code is worse than if you wrote it yourself? If it works, who cares!
And yes, I think we are well on our way to becoming Markdown jockeys.
You probably micromanage the LLMs architectural choices. Thats actually a bad idea, you will probably get better results if you let the LLM decide the architecture and just tweak it where its assumptions are wrong. It has much more context than you in its codebase anyway. Then you won’t feel like a general more, but you’ll be much more productive :)
I have a mix of heavily reviewed and more vibe-coded projects, and the ones where I manage the architectural choices end up working a lot better, even just for the LLM itself. I can get features built much quicker with less tokens and bugs. Meanwhile the more vibe-coded stuff always takes longer and sometimes gets bugs
Edit: To add to that, even in my "micro-mangaged" projects, the parts I didn't review end and treat like a block box temporarily up being the parts that introduce bugs.
I’ve seen advanced models implement function calls in an interpreter by instantiating an entirely new interpreter to handle the call, so no I’m not going to delegate architecture to the agent. And no, it didn’t do that because it had more context.
reply