YT’s blocked on it.

Just tried it. It works.

And yet whenever some achievement is made, the headlines are “Musk achieves great feat”

Not at all. It’s not “how likely is the next word to be X”. That wouldn’t be context.

I’m guessing you didn’t watch the video.

I’m not wrong. There’s mountains of research demonstrating that LLMs encode contextual relationships between words during training.

There’s so much more happening beyond “predicting the next word”. This is one of those unfortunate “dumbing down the science communication” things. It was said once and now it’s just repeated non-stop.

If you really want a better understanding, watch this video:

https://youtu.be/UKcWu1l_UNw

And before your next response starts with “but Apple…”

Their paper has had many holes poked into it already. Also, it’s not a coincidence their paper released just before their WWDC event which had almost zero AI stuff in it. They flopped so hard on AI that they even have class action lawsuits against them for their false advertising. In fact, it turns out that a lot of their AI demos from last year were completely fabricated and didn’t exist as a product when they announced them. Even some top Apple people only learned of those features during the announcements.

Apple’s paper on LLMs is completely biased in their favour.

it just repeats things which approximate those that have been said before.

That’s not correct and over simplifies how LLMs work. I agree with the spirit of what you’re saying though.

15 years ago, maybe. But I haven’t had any compatibility issues in many years.

The funny thing about that story, and the outset that no one covered after the fact, is that Munich reversed direction again and ultimately did go with Linux and open source stacks.

This is corporate AI against open source AI.

Show me where I can download Midjourneys full model to run it locally and then we can agree to call it “open weights”. Unless their base model and training data is also available, it’s not open source.

It’s a tech illiterate YouTuber for tech illiterate people that think they are tech literate

That’s a great way to say it. I usually just call him a “tech entertainer” that real tech people look down on. But I like your version.

The video that really polarized my opinion on them was their “storage server upgrade” video where they worked on replacing their horribly and amateurishly configured storage server.

Wendell from Lvl1 and Allan Jude (maintainer for OpenZFS) commented on LTT’s setup and, while they didn’t outright say anything negative, they didn’t have anything good to say and their tone heavily implied they thought LTT are posers.

Linus Sebastian

I cannot believe he still has a channel worth anything. I think LTT favs might actually be worse than Swift8es

So the guys on the street that scam tourists are high profile scammers?

I did acknowledge that it’s not exclusive to the US. And I didn’t say “it is”, I said “it feels like”.

FTX, Theranos, Fyre Festival, Enron, Bernie Madoff, Logan Paul’s CrytoZoo, Charles Ponzi (the OG Ponzi scammer), etc.

While scams exist everywhere, the US seems specially suited to embolden people to run scams. At least high profile ones.

It’s a whole hell of a lot harder to rig when your name is everywhere when you win.

This also sounds like a uniquely US problem. Not that there aren’t scammers everywhere, but it feels like it would be more prevalent in the US.

You are the one basing your argument on an article from 2008 , not me.

… what? You literally linked the article from Android Authority, not me.

You are completely deranged.

Says the person claiming a model’s computational power usage scales with the number of classes trained.

Now come back with some hard evidence

Hard evidence for what? I’ve never once claimed phones are listening to people’s conversations. This whole thread has been about the technical viability of such a system. Not evidence of it’s literal existence.

You, on the other hand, have spewed nonsense this whole time.

So like I’ve said more than once, come back with something real or stay in your lane.

I already did multiple times

No you didn’t, because you keep saying wrong things.

you just refuse to read it

I don’t need to read it, because I read it when it came out… back in 2008. I read their stuff regularly. I also read all the other stuff about this topic (AI tech). An article from 2008 is irrelevant at this point. Technology has advanced leaps and bounds in 17 years. AI wasn’t even a thing back then. Things like Picovoice didn’t even exist until recently.

It also says a lot that your source of truth is a near 20-year old article from Android Authority.

How often do you say Nike ?

Personally? Never.

More interesting would be “I will buy a pair of new shoes” now shoes can be mentioned in tons of context so you better have a way of separate it.

I don’t know about “interesting”, but I do agree that it would be much greater context to better target ads. But that’s not what the discussion was about. I said way back that I’m not positioning this idea of phone’s listening as an absolute certainty. My whole point was that at a technological level it’s well within technical means to accomplish the whole “our phones listen to what we say” all while not draining the battery enough to be outright noticeable.

Another thing to note, is that most (if not all) of the anecdotal stories about people talking about a topic and then seeing ads about that thing are often generic conversations. Even in my own tests, which are anecdotal, confirm that. I never talk about boating. I never search anything about boats. I also never saw any ads about boats. Etc. So I did a little test on my own recently and openly talked about “getting the boat ready”, “can’t wait to go boating next week”, “need to get the boat in the water and ready for the season”, and so on. I did this for about an hour solid. Then waited and hour and visited some generic websites that show ads, and lo and behold there were lots of ads for buying a new propeller, ads for nearby marinas, ads for marina supply shops, ads for boating accessories, and so on.

Like I said, it’s entirely anecdotal and in no way conclusive, but it does lead me to believe that there might be truth to the rumours. And it’s the kind of thing I’ve heard from many other technical people who deliberately tried to trigger ads on topics they never deal with otherwise.

And also like I said before either come back with something real, or go away and concede you’re out of your depth.

No you are wrong

Lol. “Nuh-uh” doesn’t work with me.

https://stackoverflow.com/questions/64008486/effects-of-number-of-classes-on-inference-time-in-object-detection-api

Seems your making things up on the go

I speak from knowledge and experience. What do you bring to the table?

More wake words to listen to more battery drain. Fact.

1 trained class = 1 model

100 trained classes = 1 model

Tell me how running 1 model would drain more battery than running 1 model? I’ll wait…

You have ZERO context then. Completely useless .

The person said “NIKE” a few times, show them ads for shoes. The person said “mechanic” “car” “fixed” around the same time, show them ads for local car repair shops.

You don’t need the full context of what was said to get some context from just the words. The spacing in time and the revelations relationship between words can give you a whole lot of context. Plenty to target ads.

Now, either come back with something real, or go away and conceed you’re out of your depth.

Give CodeBerg a look. It’s starting to pick up some steam.

https://codeberg.org/

keyword detection like “Hey Google” is only used to wake up a device from a low power state to perform more powerful listening

That’s more applicable for something like a Google Mini. A phone is powerful enough, especially with the NPU most phones have now, to perform those detecting efficiently without stepping up the CPU state.

Is there some kink of roleplaying AI dev?

Is there some kink on your side in pretending you’re smart? You have no idea who I am or what I know.

Increasing the number of keywords to thousands or more (which you would need to cover the range of possible ad topics) requires more processing power

Again, you’re showing your lack of knowledge here. A model doesn’t use more power if trained on one class or a hundred. The amount of cycles is the same in both instances.

It’s usually smart speakers that have a low powered chip that processes the wake word and fires up a more powerful chip. That doesn’t exist in phones.

Edit: just to hammer home a point. Your example of “hey Google” simply waking up the device for more complex processing just proves my point. The scenario we’re talking about is the same as the wake word. We’re not looking to do any kind of complex processing. We’re just counting the number of times a word is triggered. That’s it. No reasoning out the meaning, no performing actions, no understanding of a question and then performing a search to provide a response. It’s literally a “wake-word” counter.

I don’t have any questions. This is something I know a lot about at a very technical level.

The difference between one wake word and one thousand is marginal at most. At the hardware level the mic is still listening non-stop, and the audio is still being processed. It *has" to do that otherwise it wouldn’t be able to look for even one word. And then from there it doesn’t matter if it’s one word or 10k. It’s still processing the audio data through a model.

And that’s the key part, it doesn’t matter if the model has one output or thousands, the data still bounces through each layer of the network. The processing requirements are exactly the same (assuming the exact same model).

This is the part you simply do not understand.