plaguuuuuu

It's the wrong approach, because the public demand to reverse engineer games (a pretty benign activity, for humans at least not corporations) is massive and a new subset of hackers who will release public jailbreaks that could be used for hacking which is more of a grey area

The way I use Opus is basically adhering to its training/emergent behavior. Anything else is always worse in some way. Using default harness, mostly letting it do what it wants the way it wants to (as much as possible anyway) and making it work for my own flow.

For sure it's annoying building up so much knowledge about a single product that changes every 6m anyway, but what can you do.

compromised extension then backdoors your code anyway

It's a lot more like a bunch of people bought Cisco networking equipment and then the NSA secretly intercepted shipments, implanted surveillance backdoors, then resealed the boxes and sent them on to unsuspecting overseas targets.

it depends on the task and it's hard to say

Despite Opus 5.5 being newer and "better" if you read benchmarks, I've found Fable to be better for some things (subjective stuff)

what setup did you use to train it?

nobody is going to read that README

The answer is that AI tools are not easy to use. It took me quite a long time to learn actually.

Eventual productization will probably make that effort meaningless but hey. whatever. I'm too old to sell my ass for cash too.

there's a problem with that. Probabilistic slot machine works for anything with a deterministic eval (coding, troubleshooting, etc). It doesn't work for anything subjective, like the thing this model is targeting, professional work.

Wait, you're saying that mathematics isn't useful to software engineers?

You should give it a go.

eh?

I have ADHD and I tend to be the total opposite

I skip over details because my short term memory is bad and I work in extremely abstract abstractions all the time

I have a hard time fleshing out work presentations past like 10 minutes. I figure things get boring. to be honest... maybe that's accurate and nobody's listening past 10 minutes.

The highly heritable aspect of ADHD makes that proposal a bit weird

The extremely basic use case is that I'm doing something at work and it's not finished yet when I leave the office.

Or my laptop crashes, ugh.

Yes, if I could ssh into a random server it'd be fine. But I can't.

I had more success starting from scratch. Especially claude's skill-creator skill, it micromanages, which is in fact worse for 5+ models than just leaving the instructions out and crossing your fingers.

Start from scratch, do some test runs, find the bugs, add the minimal possible text to avoid the bug, iterate

You can get 95% of my impl workflow skill by just telling Claude "split the work into slices and use ephemeral subagents" and the other 5% takes like 10x as much text to achieve