Skip Navigation

Posts
23
Comments
704
Joined
3 yr. ago

  • assuming API pricing

    API pricing is not what it costs openai to offer the inference.

    API pricing is what they're currently selling their inference, with profit included in price.

    "But if the user didn't get a subscription, he would have spent $14k in token" is a wrong assumption.

    The user is wasting the tokens on useless tasks because they're free to him, but once he has to pay them, he wouldn't ask to generate 100 images of a toilet eating carrots anymore

  • It turns out that in modern versions (for Mac) they inserted a timebomb that disables the license check.

    "Sorry the software now is EOL we can't send an update to disable the timebomb that we set five years ago"

    Users of office 2021 are luckier (for now) because it goes EOL in October, so they can receive the June patch that moves the July timebomb forward another five years

  • They mean that when it was originally posted on the db0 Lemmy piracy community, there was a disclaimer to say it was sarcasm

  • It appears almost every single time I login to gmail

  • When you use Google services there's a very annoying popup "why u no use Chrome" that appears at least once every three week and you can only choose between "ok I give up, install now" and "no, ask me again for the 134th time"

  • I'm shocked to learn that a dedicated browser always running in the background for a chat app uses 100x the memory and resources

  • What do you mean, the app is the web version now

  • Am I wrong or this article looks completely written by a LLM? The way it always repeats the redditor username is so weird.

    Articles like this are a valid reason to choose an username like 'igargleexpiredcum'

  • Me too, I assumed all their engineers were acqui-hired by Google for the Pixel line

  • I saw in the teardown that the battery was made in the Philippines

  • I'm surprised it's a gold painted HTC phone from two years ago (good quality even if probably HTC had a warehouse full of unsold inventory that they were happy to repaint at huge discount) and not the cheapest shit that you can find on Alibaba with a soc from a decade ago and the ROM already preloaded with malware.

  • Me too, I just saw the empty spaces reserved for the ads

  • Meta: what do the website mean with "I'm out of free articles" when there's an ad break every single sentence?

  • He updated the video a few hours ago saying that some genius accomplished that

    Disclaimer: I didn't actually watch the video, only saw the thumbnail on my feed

  • The GPUs they're using don't have video output, take the energy of a furnace and are fanless (rely on the high air flow of a very noisy server rack). And they're not using a standard pcie connector.

    The CPUs are the same, big like a smartphone, require a massive cooler and a massive motherboard

    Motherboards are difficult to repurpose as they require two massive processors

    RAM is registered ecc and thanks to Intel almost no consumer hardware supports that.

    SSDs maybe we can buy a cheap pcie adapter on AliExpress, but the data lines on a consumer processor are not too many so you can realistically put 1-2 in your PC, not dozens like the old SAS server drives. And also they're toasty as they assume the case is giving high air flow

  • so, train the computer vision model for a gun and train again for a shotgun. Run the two detection models at the same time.

    Your approach is the typical "but if you really want you can use an atomic bomb to kill mosquitoes" - yes, you could do that, but nobody is paying $1 mil/year in inference costs (+some expensively licensed software to wrap around that) when it can be done locally with a $300 GPU (+ some expensively licensed software to wrap around that)

  • Using a LLM for detecting a specific object on an image is possible but stupid: if your object is always the same (like in this case) it's several orders of magnitude cheaper to train once on that specific object then use the computer vision model running directly on the local server that's recording the video.

    Otherwise:

    1. the api costs would be colossal, 0.001$ per each image, at 30 fps it's $100 per hour, nobody would pay that
    2. The detection latency would be several seconds vs almost instant
    3. Without internet connection the system wouldn't work

    Use cases for LLM-based image recognition is if the object changes at every request or it's ultra specific with brands and colors

  • I tried to scroll his "truth" profile but after ten minutes I still was at "1 day ago" and I couldn't bear the amount of shit to continue

    A true dedicated shitposter

  • If I understood right, Clemens tönnies is the uncle, deeply involved in the slaughtering and with the racism controversies. A real asshole. Clemens tönnies Junior is the one funding the software and he sold all the shares of the slaughtering company to his brother a decade ago. Maybe a better person?