Perplexity AI submits bid to merge with TikTok

- Name? No

- Technology? No because we rely on other LLMs

Do we have anything? No. We've taken in a lot of money from big-name investors, so let's see if we can turn ourselves into a complementary asset before people realize that LLMs are nowhere near what they are sold as.

$500M is nothing to sneeze at, but that is like 3 orders of magnitude less than TikTok’s value.

It was also much better for posing questions that required the most up-to-date knowledge.

It blows every "model only" service because it's 100x more accurate.

I use it instead of Google for every search now.

If anybody at perplexity wants my more-direct feedback (beyond what I've submitted via your platform's conversations), my postal is listed in /hn/bio (I do not use email, so if your platform eventually `requires` this it's an immediate disusership from me dawg).

On that note, Facebook going all in on AI slop is brilliant. Most users will stay at the trough for ever increasing lengths of time.

And when the ownership structure is the problem, it's obvious that this deal structured as a merger would solve nothing. They'd need to bring a couple of hundred billion to the table; and then we're back to this being just embarrassing, because obviously the money isn't there.

Also, AI video's IQ is generally pretty low now, but it still has the capacity to create very tailored experiences, such as with AI influencers automatically responding to every single comment from any viewer that donates X dollars. The are limitations to the digital avatars and video generation etc., but they are already very engaging. And in the next 1-3 years they will get much better.

Where we are headed in possibly 2-5 years is the expectation for many people that the personalities, activities, videos in general are completely customized for them or for the communities they are in. And change immediately based on their whims (or constrained by more realistic humanlike reactions if desired).

LTX video can be fine-tuned to produce a consistent character and runs at realtime frame rates. It is a tiny 2b model and it's understanding of prompts and the world is garbage because of that, but hardware and models keep getting much much better.

There isn't one. The only type of people excited about AI content are people who are into AI, and see it as the aforementioned way to produce potentially algorithmically driven viral content with no skills or effort. And, with anything as unpredictable as various content platform's algorithms, and especially when it's initially starting out, there will be successes, simply by the law of averages. However, no one, and I do mean absolutely no one, is seeking out AI content beyond for the notion of maybe seeing what it's like for it's own sake. People comment regularly on TikTok and elsewhere about how they find the AI voices grating, they dislike the AI captions that don't match what's actually said, and we have ample history of content farms turning out repetitive, odd nonsense that people have already become incredibly sick of... which is basically what you're guaranteed to get when you have an AI generate a TikTok for you, because of both the weight of those videos in the training data, and because of how these models work.

For those in the know, there's been substantial trends going back a decade now of low effort, slop content being driven further and further into irrelevance because for every hundred lazy grindset people trying to generate income, you have one person who knows things and has studied them, who has insight to offer, and figures out the remaining parts by trial and error; how to shoot video well, how to light a set, all of that. And those people naturally move to the top of engagement. You might, MIGHT, catch early accounts or new users who haven't found interesting people yet... but the algorithms on most of these sites, paradoxically, are going to identify those interests and then drive users towards more interesting people talking about them, because that's what the vast majority of users want.

It doesn't need to be important, or like... pillars of civilization type knowledge, this applies to any topic you can think of, no matter how insipid you think it might be. If I was interested in collector Barbie dolls, would that algorithm draw me towards a video of a Barbie enthusiast who owns hundreds of them and has studied the topic to death, or to a content mill churning out detritus of some AI generated Barbie-girl woman, talking in TikTok narrator voice, about how her dolls are so cool and neat? And like, sure, maybe you can make the AI video thing so good that it can reproduce the same room, the same presenter, with perfect features and clarity hosting a collection of dolls you somehow manage to keep in the AI's memory, all of that is difficult, but solvable, in theory anyway. How do you give your AI presenter with her AI dolls anything even remotely interesting to say about the dolls she doesn't collect, in the room that doesn't exist?