Comment on Programmer tries to explain binary search to the police

• CosmicCleric@lemmy.world ⁨1⁩ ⁨year⁩ ago

  If it’s not “for the duration of the rest of the video,” then binary search would be useless

  That’s not true. It only has to be long enough to be detectable, by landing on a strip of video that it exists on. It’ll be harder, definately, but still doable.

  source

  • starman2112@sh.itjust.works ⁨1⁩ ⁨year⁩ ago

    Maybe I have no understanding of what a binary search is. My understanding is that you check halfway through the video, see if the thing has happened yet, then skip halfway to the end if it hasn’t. Check again, skip again. When you see the cue that the event has happened, you rewind to halfway between the latest point where the event hadn’t happened yet and the earliest point when it has. Keep doing that and you can pinpoint the exact frame where the event happens in a matter of minutes.

    Binary search would be largely useless in cases where you have a good chance of skipping right past the event. If the video is an hour long, and the event happens 34 minutes in and leaves a visual cue that lasts less than 11 minutes, then binary search does not find the event. At that point, watching the video fast forwarded would be the way to go, and that’s not a binary search, that’s just watching the video.

    source

    • CosmicCleric@lemmy.world ⁨1⁩ ⁨year⁩ ago

      When you see the cue that the event has happened, you rewind

      The event has happened, or the aftereffects that the event happened. That is my point, the aftereffects matter as much as the event itself. As long as the ‘after’ looks different than the ‘before’ for any reason, that is a marker to give you an indication on which way to go, rewind, or advance.

      And yes, either the effect or the aftereffects has to last long enough to be noticed by humans, less long by AI (faster to detect changes than humans). But the vast majority of events, when humans are involved, leave long aftereffects usually. Yes, not 100% of the time, but usually.

      source

      • starman2112@sh.itjust.works ⁨1⁩ ⁨year⁩ ago

        The event has happened, or the aftereffects that the event happened.

        In which case there are visual cues and it’s something that the comment you argued with acknowledged would be eligible for binary sort

        But the vast majority of events, when humans are involved, leave long aftereffects usually. Yes, not 100% of the time, but usually.

        Nobody said otherwise, you’re arguing with strawmen

        source

        • -> View More Comments
      • starman2112@sh.itjust.works ⁨1⁩ ⁨year⁩ ago

        less long by AI (faster to detect changes than humans).

        Many things change things. A bit of smoke in the air might have been from a gunshot that happened 10 minutes ago, or it might have been from a cigarette 15 minutes ago. Binary search relies on changes that indicate a specific thing has happened–a broken window, a bike no longer there, blood stains on the street. Anything undetectable by humans would still be useless to AIs. A bit a smoke? Could have been a gunshot 3 minutes ago, could have been a cigarette, could be fog, could be a vape. Even the things that AIs are truly useful for, like video compression artifacts, wouldn’t help, because any number of things can cause compression artifacts. How could it tell what pixels are slightly off color because of a gunshot 3 minutes ago, and what pixels are slightly off color because someone walked by the screen?

        At that point, just feed the entire video to the AI and have it tell you when it sees guns or puffs of smoke or hears screams. Binary search is useless when you can just have a machine watch the entire video in one sitting over the course of five seconds and tell you when the interesting thing happens.

        source

        • -> View More Comments