375 karma · joined July 16, 2025
@poemxo_ on X
Hawaii
Idk it seems to me like I just read something written by someone who considers himself a computer expert because he built a computer, but doesn't actually know anything.
It seems like this[1] documentation matches the stuff in TFA
> The system being referred to in that explanation is Microsoft 365 (M365) / Office 365 audit logging, specifically the Unified Audit Log in the Microsoft Purview Compliance Portal.
It's particularly bad because customers see it as a defect. No one wants to pay full price for defective equipment. The only thing that would make it worse is if this "hack" were reproducible on the flipper zero and they get themselves into another Kia Boys situation.
Since tokens are atomic, which I didn't realize earlier, then maybe it's still intelligent if it can realize it can extract the result by writing len([b for b in word if b == my_letter]) and decide on its own to return that value.
In the story, without spoiling anything, the titular Snow Crash can only happen to programmers. Similarly your LLM experience couldn't happen to a smooth-brained non-physicist like me.
To add something useful, this is very cool but I wish it came in JBOD form so I could run it off a Mac Mini instead, and the pricing listed in TFA seems like an idealistic lower limit.
With AI, the meta changes and suddenly you must pay for system. The system could start jacking up prices or punitively denying politically unpalatable people their loneliness treatment simply because of the prevailing social forces. Especially if ekeing out more profit over time is the true objective, providing the social good of allaying loneliness will become secondary. I don't see this as a good thing. It is basically an AI form of drug dealing with extra steps.
Software engineering and computer science seem to have two strict criteria to consider and neither of them is the same sort of continuous, analog suffering as wearing large shoes or practicing shooting a basketball. These criteria are
1) can you solve hard problems? 2) do you want to continue solving hard problems?
At least to me it seems that those two things take more effort and willpower than anything else in software. So I don't think challenging a person about whether they would love to sit in front a computer all day is the right approach.
The other chunk of time, to me anyway, seems to be creating a mental model of the subject matter, and when you study something well you have a strong grasp on the forces influencing cause and effect within that matter. It's this part of the process that I would use AI the least, if I am to learn it for myself. Otherwise my mental model will consist of a bunch of "includes" from the AI model and will only be resolvable with access to AI. Personally, I want a coherent "offline" model to be stored in my brain before I consider myself studied up in the area.
So I think it matters, I think access systems should be designed with a wider set of human behaviors in mind, and there should be technical hurdles to leaking a majority of customers' personal information.