I was fascinated by them since reading a guide for Cities: Skylines that said that roads were like trees. There's a trunk that moves large amounts of nutrients and little branches that distribute the nutrients to the leaves. Such simple rules, but such complex and deterministic results.
GPT-5 used to do this on release, but seems to have reverted back, especially on the codex versions. This may well be a feature.
I joked that this is the side effect of asking it to act like a senior software engineer. It tends to talk back and do its own thing. There was that one time when the thought processes went "I'm a full stack engineer" > "I'm expanding my connections on LinkedIn" > "I'm establishing myself as a tech writer" > "I'm evaluating classes in professional writing". It does have introspection capabilities, but one could argue it's just a bug or emergent.
Anyway, option 1: why not just use Sonnet? Heck you can use Haiku if you're giving it clear instructions. The thinking ones do perform worse on clear tasks. You also get your rolled back version.
Option 2: Use role prompting [0]
Give it some junior engineer role where it's expected to follow instructions exactly as given.
> thought processes went "I'm a full stack engineer" > ... > "I'm evaluating classes in professional writing"
Let that one run, let's see where it ends.
maybe "I no longer build software; I now make furniture out of wood"
or "Have taken up farming"
or perhaps "I've sold all my RSUs, am taking a Sabattical" > "I'm exploring activities I find more fulfilling: yoga & angel investing" > "I'm the author of a substack blog post where I share life advice with others" [1]
I'm not a fan of the things proposed by 80000 Hours or other smartest people. There's a lot of people who said that AI would be the most important thing to work on, and those people ended up using AI for evil in the end.
Plus it assumes that survival or suffering are the highest priority things, but there's a lot more other things like community and civilization.
The last slides make some good points about going up Maslow's hierarchy. Some companies pick new technology because they have the engineering power to build on top of the new stuff. But not everyone has this, and they're probably better off picking a boring database.
Personally, I'm considering migrating from a stable MongoDB way of doing things to postgres, but this is more because I understand postgres much better recently. It could be that Mongo does the same things I want to do. Keeping on MongoDB is the boring option for me even if it's not the classic example.
However, there's a little misconception that boring means safe. MySQL is popular for "just working" until emojis started breaking it. You'll have new stuff, like RFC 9078 where you can respond to emails with just an emoji. Everything breaks eventually in this world, it's more about what's less of a pain to patch together once it breaks.
Social media has made it normal to have closed social circles. You like a couple posts by Karpathy and LeCun and your feed is now full of AI. Or you like a post by the anti-AI folks and your feed is now on how useless or harmful it is.
With HN, we all see the same news. But now it's far to the extreme of either party.
HN is often around the early majority of the Innovation Adoption curve. I was actually going to ask why Clawdbot wasn't even mentioned on HN when it had been everywhere else... but the news did hit the front page a few days later. There's a few killer updates from the AI coding groups which only get 0-2 upvotes and disappear.
HN is not the place to go to be up to date on AI, it's where you go to see the news that has hit the mainstream. Opus 4.6 was "leaked" on media and hyped as if it were Claude 5 or something. By the time the news was actually out, people had formed their opinions.
I've had a zombie project running on this for many years now. I used to charge people about $25 for 'lifetime', but there's only about ~10 regular users on it, so I try to keep it alive.
Decided this is the time to make the switch over to AWS. They've been rather painful with cancellation. They required all dynos be downgraded to "Eco". Fine.
But this downgrade also incurred another $5 charge which they now required me to pay to remove the credit card. It's not much, but this is shady af.
The billing language has become increasingly shady over the years. Basic is "~$0.010/hour, Max of $7 per month". Eco is "~$0.005/hour, Flat fee of $5.00/month". But in reality, you're just being charged a flat $7 or $5 either way. Eco is visually shown as the "free" option, except it's not free at all.
I'd love to just keep using Heroku and paying some flat rate; we were talking about putting some more work & funding into the project and maybe scaling it up to thousands of users. But I have no idea what Heroku can actually scale to and how much it costs. AWS etc are also not that clear on costs, but at least their specifications is a little more detailed than "Superior performance for your very large-scale, high traffic app"
They meant what they wrote. Merriam-Webster's definition: "to support the weight of"
It means they're transitioning to the absolute minimum to keep it alive and nothing more. That could, in worst cases, mean firing everyone except one guy, or using AI to keep it alive.
There was a time when I put the EA-Nasir text into base64 and asked AI to convert it. Remarkably it identified the correct text but pulled the most popular translation of the text than the one I gave it.
Sucks that you got a really shitty response to your prompt. If I were you, the model provider would be receiving my complaint via clay tablet right away.
reply