Paper explanation videos have got to be the most precious resource available right now. I have been asking chat GPT endless questions about the proper configuration of Weaviate over the last 72 hours. I’m now the proud owner of my very own vector database. 😀 Please keep making these videos !
Thank you so much for writing this Dan, it really means a lot! Haha that's amazing to hear, using ChatGPT to help setup Weaviate!! Will do, thanks again!
Great video Connor. It took me back to the days of your iconic wooden powerpoint background! Really enjoyed the video and learning about the paper. The method kind of reminds me of the card game "BS" where players sequentially place cards (sentences, in this analogy) on the table, and the other players based on how confident the person places the card down can call them on their bluff and set the record straight. But having access to the token probs almost feels like cheating - like attaching everyone to a lie detector or something...
Haha man, thank you so much for remembering that -- insane how much AI / DL has evolved and ofc a bit sentimental for me thinking of my personal story with it and the wooden powerpoints lmao. Hmm very cool, love that -- would be even better if you could somehow see into the confidence patterns of their brains haha! I am publishing a new podcast tomorrow with the inventor of ChatArena, reminder me a lot of the "BS" game but with large language models competing over conversation, I think you will enjoy it! Thanks again Oisin, really really appreciated this comment!
Hi, For fine-tuning the generator model like BART for question answering tasks generally the dataset set used contains question-and-answer pairs If I only have data with questions and their corresponding passages, how can I fine-tune the generator model to improve answers for my custom data?
Paper explanation videos have got to be the most precious resource available right now. I have been asking chat GPT endless questions about the proper configuration of Weaviate over the last 72 hours. I’m now the proud owner of my very own vector database. 😀
Please keep making these videos !
Thank you so much for writing this Dan, it really means a lot! Haha that's amazing to hear, using ChatGPT to help setup Weaviate!! Will do, thanks again!
Absolutely love this Connor! You should make more of these paper deep dives. Super useful
Thank you so much David, really appreciated!!
Great video Connor. It took me back to the days of your iconic wooden powerpoint background! Really enjoyed the video and learning about the paper.
The method kind of reminds me of the card game "BS" where players sequentially place cards (sentences, in this analogy) on the table, and the other players based on how confident the person places the card down can call them on their bluff and set the record straight. But having access to the token probs almost feels like cheating - like attaching everyone to a lie detector or something...
Haha man, thank you so much for remembering that -- insane how much AI / DL has evolved and ofc a bit sentimental for me thinking of my personal story with it and the wooden powerpoints lmao. Hmm very cool, love that -- would be even better if you could somehow see into the confidence patterns of their brains haha! I am publishing a new podcast tomorrow with the inventor of ChatArena, reminder me a lot of the "BS" game but with large language models competing over conversation, I think you will enjoy it! Thanks again Oisin, really really appreciated this comment!
@@connor-shorten Oh, wow. Nice! Will keep an eye out for it 🙏
Looking forward to move videos like this, your explaination was spot on! These videos will be prove to be invaluable
much appreciate this explaination
Thank you so much Soumya!
awesome content. looking forward to more.
This was really informative. Looking forward to more such content. ☄️
Thanks for the explanation :)
Bravo
Thanks Derrick!
Hi,
For fine-tuning the generator model like BART for question answering tasks generally the dataset set used contains question-and-answer pairs
If I only have data with questions and their corresponding passages, how can I fine-tune the generator model to improve answers for my custom data?
What other methods are there? Is there maybe a good overview somewhere?