4A still doesn't permit you to destroy the evidence. Resist on 4A grounds, destroy the evidence, 4A reasons get overturned - you've got yourself a conviction. US v. Akram Musleh.
7,970 karma · joined October 20, 2009
You can refuse to hand over access. You can't go torch evidence. Caught Ollie North as well.
Given your description of the behavior, my guess is context window compression bit you. Less likely if it's a root level CLAUDE.md and you used a recent version of Claude Code.
AGENTS.md is a poor substitute for agents actually learning, but it somewhat works.
And sure, you can (and should) enforce constraints via githooks whenever you can (determinism is good), but if the agent only finds out at checkin time, it's created a whole bunch of code it needs to redo, and that doesn't run properly if you do the caveman thing and actually run your code before committing. Again, token & context waste.
(Or the equivalent incantations for other languages)
They aren't there for others, they are there to make sure the code produced measures up to my standards. Sure, it helps that it also applies to requests made by others, but that's a bonus.
"Sure, the model makes repeatedly the same mistakes, and I could prevent that, but then it wouldn't be the same model" sounds like an excuse for masochism.
The one small bit of truth is that yes, instructions might become outdated, and they might affect negatively how the system performs. You fix that by... updating your instructions.
And if you deeply care about that potential negative impact (most models are pretty good at just ignoring long outdated issues), you run evals on model upgrades. You certainly don't refuse to use anything but the bare model.
(Of course, there are also folks who refuse to write code in any way except with sed, so there we go)
Which is going not too badly for the shorts, so far. There's a reason almost 50% of the float is loaned out.
We'll see once the earnings call rolls around.
We as society are choosing to be in that place.
Yep. I, too, get more productive by spending a PhD worth of effort on choosing a window manager.
This means the certificate is independent of author and source text.
There's nothing stopping you from sending fake counts/duration to the semoi server. It's a certificate that only says "at this point in time, this is the information I was provided with".
You can then attach it to any piece of text you like.
At the very least, you'd need the ability to prove that there is an underlying event stream with these characteristics, and that this exact event stream creates the document in question. You still can fake that event stream, but it becomes enough work to distract at least casual abusers.
But really, it's the equivalent of saying "I wrote this without AI, honest" in-doc and signing that with your personal key. The value depends entirely on your willingess to be truthful. (IOW: I predict we'll see a resurgence of reputation systems, to some extent)
If you outsource that work to customers/patients, you'll end up with the car dealership model, where the sales rep begs you to give a 10 on every single question including on the interior design so they don't get fired.
That's the part most of this discussion misses. Supervisors exist for a reason. Congrats on your flat org structure, you fucked up an important feedback channel.
"Just" is doing a lot of work there. SpaceX is planning to launch 100 GW of compute annually, that comes with ~ 2.5 square kilometers of radiator (assuming an optimistic 800K radiator temp and emissivity of 0.9, double sided)
Go for advanced carbon composites, you can do that with just 5,000 metrics tons or so of material. That's 34 starship launches just for the radiators. We haven't solved assembly, we haven't brought up power panels or core compute. Planned launch cadence that SpaceX hopes to reach end of this year: 12/year.
Do those exist? I'm usually happy to see a mild candle flicker in the ethics window.
See also self-checkout etc.
I.e. I was responding to the original "in 'most of the world' EMS quality is ass.", trying to say that you only get that outcome if you compare apples and oranges (developed vs developing)
8-10 minutes in urban areas is pretty much universal (for the developed world). Rural areas are where differences come in, and even there standards for the developed world are somewhat on the same level.
And if it's got value for you, hey, it's open source. Spend the 15 minutes and 10 cents, it's a bargain.
He who brings the slop cannon shall be drowned by slop rain.
I've had it reject looking at pages served from my local network because it "can't find it with my search tool" and had "ethical concerns about consent for access".
The People's AI Concern Front has gotten the classifier they want, and it's made Claude hilariously useless. I am waiting with bated breath for their next set of revenue numbers. (And happily hand my money to competitors instead)
But given the sudden wide audience, a quick "here's what this is for" at the top might be helpful.
Society cannot work with too many corrupt civil servants. Yes, "autocrats", "civil liberties", and yet - the guy slurped up $325M to put his finger on the scale, not to change the model of governance.
I wish we in the west took corruption more seriously, but I suppose we're more interested in cage fights on the lawn these days.
Not really. Truly robust engineering includes a cost-benefit analysis of which edge cases you handle. We don't live in a world of unlimited time & money.
Turns out that if you try to be "15 minutes into everyone else's future", you get better and more frequent trips than any psychedelics could provide.
Is it just rehosting the list, plus a reformatted copy of the papers? I was hoping you'd have at least annotated them with what you'd learned?