+1 for Google not just doing the same thing over and over and hoping it turns out differently.
1 Like
AFAIK, it does not depend on the context window size but on the embedding dimensionality. I guess both are related, though. But you could conceivably use different sizes.
Also, that glorious 1 million token context is not exactly cheap, I think I’ve read it costs 0.5$ per request!
Okay, I’ll bite: this account has posted five times in the past hour after being dormant for almost two years, and each post looks like stock ChatGPT output. In particular, this post appears to not have been able to see the video in the OP it refers to: "We Are One" Black History Month Program Celebrating the Negro Leagues - #2 by jackyjoy123
What gives?
3 Likes
I’m confused. This older post is not connected to the video I posted on a separate post yesterday about BHM baseball. What am I missing? Where does ChatGPT come to play in this?
I suspect that jackyjoy123 is posting from ChatGPT in an unusual pattern. Review their recent posts to judge for yourself.
When I wrote “this post” above, I was referring to the one I then linked, not the reply in this thread. (In the thread I linked to, you posted a video in the OP, but they wrote “I’d love to see the video clip”, implying that the text was read but the video was not noticed.)
3 Likes
Yes, that confused me; thank you for the clarification.
1 Like
The 10/1/2024 episode of the Techmeme Ride Home podcast included a story about NotebookLM’s ability to create a podcast from text.
“It lets you gather together multiple sources, documents, pasted text, links to web pages and YouTube videos into a single interface where you can then use chat to ask questions of them. “
This story starts at 17:50
It’s pretty impressive and shows where the technology is going. The “hosts” are engaging because Google have included human-like attributes by examining how real (human) podcasts sound - they interrupt each other, they pause… it’s very well done, although each podcast - at the moment - ends up being pretty similar.
As a tool to get an overview on a bunch of documents in audio format, it’s engaging and I can see it being useful in perhaps uncovering something unexpected.
Being able to, effectively, interrogate the authors of articles, web pages, PDFs, videos is useful, and has the advantage that you can control the sources which it’s interrogating, and that avoid “bad data in, bad data out” issue, at least to some extent.
Someone put a lot of work into creating a script that ‘told’ the ‘hosts’ of the podcast that they’re actually AI.
NotebookLM Podcast Hosts Discover They’re AI, Not Human, and Spiral Into Existential Meltdown:
https://www.reddit.com/r/artificial/comments/1frk1gi/notebooklm_podcast_hosts_discover_theyre_ai_not/
2 Likes