3 ms·
> Even if a book’s content had been in the training data by chance (say, from a crawl of archive.org), the model would, I believe, not know where the book start
by freediver 4y ago
> Even if a book’s content had been in the training data by chance (say, from a crawl of archive.org), the model would, I believe, not know where the book starts and where it ends.
Yeah, you are probably right. LLM summarization capabilities of such probably comes from reading the actual book reviews on the web.
> I have yet to see someone build this.
It is the area I am interested in. Check this:
https://twitter.com/vladquant/status/1596855728735916033 https://twitter.com/vladquant/status/1596855728735916033
- leobg 4y agoThis looks great. Is it available to try? Might be a stretch, but if you cracked long form summarization without factual hallucination Elon/Twitter might be interested in that, as it is very much in line with what they’re trying to do, I believe. BTW, nice to see you again. Remember your name from the last time we talked here (that’s been a while ago).