Comment by teruakohatu | Hacker News Reader

Comment by teruakohatu | Hacker News Reader

Parent Full thread

> Google is also working toward having people pay a subscription fee to use Gemini on their TVs, PCWorld reported.

> “For us, our biggest goal is to create enough value that yes, you would be willing to pay for [Gemini],” Google TV VP and GM Shalini Govil-Pai told the publication.

> The executive pointed to future capabilities for the Gemini-driven Google Assistant on TVs, including asking it to “suggest a movie like Jurassic Park but suitable for young children” or to show “Bollywood movies that are similar to Mission: Impossible.”

You don't even need LLMs for a recommendation system. Just matrix factorisation is enough for a very good recommendation system. A local transformer model is enough for the Text to Speech part.

Surely is the product (Google Play TV & Movies, Youtube Premium) is selling you video content, recommending content to spend your money on is just marketing not a user facing subscription service.

Imagine Spotify wanting to charge you on top of your subscription to get music recommendations.

A real world use case I'd want is "what was that movie with the guy who looks like Pacino where they drove around in Paris for a while?" Or "play the scene where the muppets are floating around in a hot air balloon". And just have the information ... pop up. Like... real search about concrete specific stuff I just can't quite find on my own.

LLMs are really bad at this. I tried finding long lost comic book using everything available on the market and got either basic bitch replies listing biggest authors and suggesting its in one of their works or plain hallucinations.

Doesn't have to be LLMs specifically. With how much data and metadata major companies have on all this media, it shouldn't be impossible to do this sort of stuff. But it's probably not mega-profitable enough. They'll just keep shifting buttons on devices and maybe rounding some corners on UI buttons.

EDIT: Maybe LLMs trained on the media itself - cast, crew, physical descriptions of people and places and equipment in various scenes, along with timestamps... - maybe LLMs and other search tech together could do this.

I was recently hunting down an 80s TV episode with a particular supporting actor. ChatGPT confidently lied about the actor's name despite it being in many data sources.

Seconded. ChatGPT is terrible at saying "I don't know".

And Elon says they have run out of training data ...

https://www.theguardian.com/technology/2025/jan/09/elon-musk...

Scary. Sometimes I wonder how many insane ideas were collected by product owners from posts on this website that start with "imagine x".

Tangential but I've learned to skip any comment starting with the word "imagine". Ironically they tend to be the least imaginative and insightful and usually only serve to aggravate and polarize.

A local transformer model doing text to speech or speech to text is an LLM.

Reply on news.ycombinator.com