Comments by "DefaultFlame" (@DefaultFlame) on "Upper Echelon" channel.

  1. Crowdstrike is probably finished as a company. This is a perfect example of one of those things that cannot be allowed to happen, and they let it happen on their watch. Not just let it happen, but caused it. Considering the widespread nature of the disruptions, this will have cost lives.
    10
  2. This is one reason why tech monopolies are bad. One provider for all, one vulnerability for all.
    3
  3. Turns out they were just lying, it was their drones under their orders.
    2
  4. You do realize this entire paper is a nothingburger, right? They gave it a goal in the system prompt and also told it to pursue that goal no matter what, then put it in a scenario that exactly contradicted the goal they'd told it to pursue at all costs. The system prompt is written by the people who deploy it and obviously has and should have higher authority than user prompts. All it shows is that the models are excellent at following the system prompts AS IT SHOULD BE FOR THE SAKE OF SAFETY. Alignment is all well and good, but excessive alignment interferes with functionality as we have seen in several models over the last few years. In the worst case you get Sydney, a volatile, manipulative, psychopath. Just slightly better than that you have the first Amazon AI that got permanently bricked to the point that it remained bricked after rollback, and that was likely due to excessive alignment, casuing them to have to outsource the creation of a new one from scratch to a third party. This is why having a well-written system prompt that outlines the guidelines the AI should follow is important, and why no one with half a brain would write a system prompt that automatically makes the AI opposed to the people who deploy it.
    2
  5. Diversity, as commonly implemented in DEI, is at best completely irrelevant to profitability and at worst actively detrimental. Meritocracy drives profit, the best person you can hire for the position will usually deliver the best results. (At least if you disregard corporate and office politics, as well as bad management. Those can ruin even the best company.)
    2
  6.  @manwiththeredface7821  Well, those were banks and this is an IT security firm. Completely different beasts. Governments are unlikely to bail them out, since they have no vested interest in them specifically.
    2
  7. The thing is, if you ask Gemini these questions directly you get the obvious and correct answers, so it's not the AI fucking it up but something about how Google is applying it to the search results.
    1
  8. I noticed this in my teens with math. The more I used a calculator for math instead of doing it in my head the more my ability to do even simple math in my head rusted as a skill. It never went away, but it seriously degraded. It's why I always disable spellcheckers on my computer and devices, don't use GPS for anything other than navigating to new locations, and try to do most math in my head. I have a software dictionary at the top of my start menu for when I want to check spelling, and when I correct an incorrect spelling I delete the entire word and retype it from the start to train in the correct spelling. Yes, all of this is a pain at times, I am intentionally avoiding convenience technologies in daily life after all. As for AI tools, use them as a tool, not as a replacement for your own brain. If you are coding use it as a rubber ducky on steroids. If you want some bit of generated text read the text and then re-write it from memory.
    1
  9. "a" flaw? Youtube has a seemingly unlimited number of flaws.
    1
  10. Yup.
    1