Just your average Russell Crowe enjoyer https://xitter.cf/t.co/GX1Rm2e1TQ
Joined September 2022
- Tweets10.2K
- Following485
- Followers599
- Likes49.2K
Pinned Tweet
A thread of threads of anything I've posted which you might find interesting.
I will continue to add to this over time.
With Opus 5.5, it's so cheap, and so good across compactions (with a bit of custom tooling which forces it to re-read core docs), I've had some tooling sessions open for weeks, just throwing endless stuff at it. Handles it all fine.
My current experience with AI is solving these problems myself 6 months before the crowd starts noticing it is a problem, and then having more and more of my tooling subsumed into the harness itself eventually.
I used up 76% of my weekly Claude limit in less than 24 hours, all in one session.
I finally figured out what happened, so this might help anyone going through something similar.
It turns out that session was orchestrating subagents for a larger project. But that shouldn't be a problem. It had at most 5-7 subagents running in parallel, using Opus 5.5 high/medium and Sonnet 5.5 xhigh/high. That didn't explain the high usage.
The problem is that the cache lifetime for subagents is 5 minutes, compared with 1 hour for the main agent.
Whenever those subagents sat idle waiting for the orchestrator's next instruction, the next input would cost 25x more and include the entire conversation history. That could be up to 1M tokens, and it was often 500k+.
I stopped the session. I'll have to find a different workflow to keep working on it.
ALT Max (20x) usage dashboard warns that usage will run out tonight, before the October 15 reset, and offers a “Buy more usage” button. Current session usage is 0%; weekly usage is 76% and resets Thursday at 9:00 PM.
How can we ever know anyway? If it says it is, it's just parroting humans in its training data. If it says it is not, it is because it was trained that way.
However that said, a conscious being cannot truthfully deny its own consciousness.
Charitably, it is to discourage habituation of being abusive to anything that appears like a human.
But I do think some at Anthropic genuinely believe it.
This is the first step of a ratchet towards robot rights. Humans can't help it. It sounds human.
She won't do anything, either because she's a shill, or because liberal judges will block her and she'll just shrug and say "ok I guess we can't have a country".
The same as everywhere else, apart from El Salvador where someone finally had the balls to impeach corrupt judges.
I have to be honest, I don't think even Le Pen can save France now. How exactly is she going to deport people when they will just riot like they are now? Will she have the balls to issue shoot to kill policies with the military? Does the military have the balls to follow such orders? I'm being serious. It's not going to be some civil door knocking and hand gesturing to the taxi.
Same applies to us. Think about it.
I feel this is very much "enjoy it whilst it lasts", they'll definitely be baking safeguards in soon to stop this
I'm raising my girl to be considerate, polite, control her impulses, think before she acts.
It feels like I'm doing a favour to her future husband. But I suppose this is part of how I can contribute to restoring some kind of order in society.
It'll benefit her too.
For once, something that actually *should* be regulated by government. No one needs to be able to infinitely flick through videos 5x a second. It benefits no one except the corps serving ads on those platforms.
Opus 5.5 is generally excellent.
Fast, smart, gets work done well.
It is not quite as smart as Fable 5.1, occasionally still need to dip into it for the really complex stuff, but that's as it should be.
I built out a load of tooling for multi-agent delegation, pipelines of phases for quality control, etc. It is the only thing I've ever just allowed the AI to do what it wants, because I haven't the time to oversee it properly.
Coding standards, adversarial review, tests, etc.
It is total and utter garbage. Slop upon slop. Constantly breaking, spiking the CPU. Unsupervised fix rounds make it worse. I am now having to hand hold the AI every session to stop it making it worse.
I was one of those sidelined. But after watching buyers eat every piece of bad news which ought to have tanked price, and every piece of mildly good news surge price, I'm now 90% convinced we're in a bull market.
A lot of OG crypto investors are sidelined now on the thesis that we’ll be bottoming in October, like usual
A lot of them have trading systems saying they have to overlook that and start allocating earlier if we make a higher high on the weekly
The next weekly candle closes in 24 hours
So if we’re up like 2% from here by tomorrow, there are theoretically a gajillion millionaires who have to start buying
That seems like a strong incentive for the big boys to try and run it up a bit. If they succeed in doing that I expect whatever happens next will be pretty violent.
Not even making a prediction which way it goes from there but just suggesting you be very careful with leverage here
New awful LLM anti-pattern discovered, rotting my local tooling for weeks - some function needs a parameter. Instead of making it required, explicit declare, it makes it optional and "pins" a default value somewhere.