I love this, this is incredible!
75 karma · joined April 2, 2022
From that point I can keep building/writing the song more, but at least now it has some identity.
Setting contamination aside, there's a structural issue behavioral testing can't solve. You can falsify "this model is aligned" with one counterexample. You can't verify it with any number of passing tests. The next scenario you try might be the one that breaks the claim. That's not a methodology problem, it's a logic one, and the same shape affects interpretability and theoretical guarantees in different forms.
The most important sentence in the post is buried in the discussion: "our auditing methodology is not yet sufficient to rule out scenarios in which Claude would choose to take catastrophic autonomous action." That sentence describes the actual epistemic situation. The headline number doesn't.
Expanded on this here: https://aaronjholbrook.com/thoughts/2026-05-08-the-negative-...
For me it is beyond trying to make money or become famous, it is simply to enjoy the journey and the creativity that comes with creating music.
How do you handle logged in sessions/user authentication?
I built something much much more primitive, but I have it actually storing session data in the local project folder and then re-using those cookies so the agent can log in without issue.
Perhaps the comment YOU made could also fall into that category? Pointing out a low-effort comment is ALSO a low-effort comment?
But I am glad I read the book first, I got much more out of it - it goes a lot more in depth into the science and engineering challenges that occur throughout. Which I appreciated. I'm not sure I would have read the book in the same way if I had seen the movie first.
Amazing movie and the end credit visuals WERE incredible!
I mean look back at HN classic posts like the initial Dropbox announcement and the classic: this is nothing more than a wrapper over rsync, etc.
Quick context: I've been the lead developer for this client for over six years across four interconnected systems (ERP, competitive analysis platform, iOS app, infrastructure). All the institutional knowledge lived in my head. I brought up the bus factor problem myself and built a continuity package to eliminate it.
A few technical details that might be interesting to this crowd:
The credentials problem was the hardest part. You need secrets accessible for handoff but you can't commit them. I ended up with a two-layer system: the documentation repo has references to every credential (what it is, where it's used, who owns it) but zero actual values. A separate gitignored secrets directory holds the real .env files, SSH keys, and service credentials organized by project. One command pulls current .env files from production. Another command packages the entire secrets directory into a dated archive for the business owners to store in their vault. The two layers never mix.
On the AI side: I used AI as a research and drafting partner to produce the initial dozens of pages. It read the source code across all four systems, mapped architecture, and generated first drafts. I reviewed and corrected everything, then filled in the business context and institutional knowledge that only exists in my head. The repo also includes a prompt file so any future developer with AI tooling can point it at the file and maintain the documentation going forward. That part has been surprisingly effective for keeping things current.
The automation piece: a server audit script connects to both production servers weekly and collects current state (PHP/MySQL versions, disk usage, running services, cron jobs, SSL expiry). A pipeline sync tool pulls CI/CD configs. Everything assembles into a PDF with a TOC that gets emailed to the business owners automatically.
Happy to dig into any of this.
I no longer am even able to see shorts, they don't get suggested to me.
However, this is an extreme nuclear option and also disables the entire 'Home' and for-you type content. You can search and watch videos, but it flips it so it is 100% intention driven. Which means discoverability and browsing is non-existent. You're still able to see your existing subscriptions but that's it.
It's REALLY helped me disconnect from the addictive dopamine of the infinite, short type formats. YMMV
I've been making music now for 8 months and have absolutely fallen in love with it. I'd love to write music for a game, what kind of game are you creating?
Here's my stuff if you're interested: https://aaronholbrookmusic.com
Feel free to reach out!
I cannot possibly imagine putting my fingers near that thing though!
It completely blew me away and I felt suddenly so betrayed. I was paying $200/mo to fully utilize a service they offered and then without warning I apparently did something wrong and had no recourse. No one to ask, no one to talk to.
My advice is to be extremely wary of Anthropic. They paint themselves as the underdog/good guys, but they are just as faceless as the rest of them.
Oh, and have a backup workflow. Find / test / use other LLMs and providers. Don't become dependent on a single provider.
So I decided to vibecode an app for myself and wouldn't you know? It took me a few hours and it's INCREDIBLE! No more relying on someone else to maintain something, I can simply build my own solutions, whenever I want!
So validating to know someone enjoyed my work enough to buy it.
Thank you to everyone who's taken a listen, and thank you to the person that bought it!
<3
In case you'd like to take a look, shameless self promotion: https://aaronholbrook.bandcamp.com/music
I have to go through my back catalog, and add all my music, but I appreciate your perspective and for wanting to support artists!
Zeek Interactive is a small and highly focused group of talented people. We’ve been building great software and solving problems for clients since 1995, and we’ve been focused on the web from day one. As the web has grown and changed, our skills have expanded to match.
We're looking for an experienced full-time Mid to Senior Laravel/PHP Developer to join our engineering team. Someone who has a solid understanding of web technologies and wants to help architect, build, launch and scale major systems and user facing features.
You’re comfortable working in a fast-paced environment with a small and talented team where you're supported in your efforts to grow professionally. You are able to manage your time well, communicate effectively and collaborate in a fully distributed team.
We'd love to hear from you!
Senior Laravel/PHP Developer: https://www.ziprecruiter.com/job/9118d06c
WordPress/PHP Developer: https://www.ziprecruiter.com/jobs/zeek-991870e9/wordpress-ph...
Zeek Interactive is a small and highly focused group of talented people. We’ve been building great software and solving problems for clients since 1995, and we’ve been focused on the web from day one. As the web has grown and changed, our skills have expanded to match.
We're looking for an experienced full-time Mid to Senior Laravel Developer to join our engineering team. Someone who has a solid understanding of web technologies and wants to help architect, build, launch and scale major systems and user facing features.
You’re comfortable working in a fast-paced environment with a small and talented team where you're supported in your efforts to grow professionally. You are able to manage your time well, communicate effectively and collaborate in a fully distributed team.
We'd love to hear from you! https://www.ziprecruiter.com/job/9118d06c