An Idea/suggestion for 2023. I remember an interesting discussion on one of your social media about very large models, and about how only institutions with a large amount of money have access to the infrastructure needed to train them. We now start to have open source models, such as Bloom, that can be fine tuned and combined with other innovative algorythms using affordable hardware resources. I'd like to hear your take about these new possibilities.
AI so far is very vulnerable to what I call the bagel problem: the center of a bagel isn't in the bagel. If I start with a set of accurate, truthful texts there is no promise that the latent space generated from them will be accurate and truthful. The space between truths may not be true.
What is the co-working session you mention, please? I couldn't find any link. Thanks for the video, it's interesting to hear your takes on things! I am in a similar major and I relate especially to the issues you mention with adhd. I am also wondering what your take is on the shift that youtube seems to be taking with promoting short form over long form. My subscription box is basically the only way your video gets to me nowadays!
Oh, I must have missed it - here’s the link! in.flow.club/h/jordan?inviteCode=Jordan382 In terms of the shift to short form, I don’t mind it but I do hope that long form remains as the primary push for RUclips in the long term. I liked making short form stuff on Tiktok, but I don’t think I’d want to primarily do that.
I'd say a large part of the improvement of ChatGPT in particular was also using a bunch of underpaid workers in Kenya and other places to keep training them using them with human labour.
Thank you for making these complicated subjects more accessible for those of us who do not have specialized education about them!
An Idea/suggestion for 2023. I remember an interesting discussion on one of your social media about very large models, and about how only institutions with a large amount of money have access to the infrastructure needed to train them. We now start to have open source models, such as Bloom, that can be fine tuned and combined with other innovative algorythms using affordable hardware resources. I'd like to hear your take about these new possibilities.
Will add that to the list!
@@JordanHarrod thanks for considering my suggestion!
Always nice getting one of your videos in the notification feed!
AI so far is very vulnerable to what I call the bagel problem: the center of a bagel isn't in the bagel. If I start with a set of accurate, truthful texts there is no promise that the latent space generated from them will be accurate and truthful. The space between truths may not be true.
Love ur hair girl !
Go Jordan! Outstanding video as always.
You explain these things fantastically. Thank you!
Always informative and insightful!
What is the co-working session you mention, please? I couldn't find any link.
Thanks for the video, it's interesting to hear your takes on things! I am in a similar major and I relate especially to the issues you mention with adhd.
I am also wondering what your take is on the shift that youtube seems to be taking with promoting short form over long form. My subscription box is basically the only way your video gets to me nowadays!
Oh, I must have missed it - here’s the link! in.flow.club/h/jordan?inviteCode=Jordan382
In terms of the shift to short form, I don’t mind it but I do hope that long form remains as the primary push for RUclips in the long term. I liked making short form stuff on Tiktok, but I don’t think I’d want to primarily do that.
i literally wanna be like you, people in AI industry are so cool
Thanks for sharing your thoughts and opinions.
I'd say a large part of the improvement of ChatGPT in particular was also using a bunch of underpaid workers in Kenya and other places to keep training them using them with human labour.