I've had no idea about how to use embeddings or even the value in storing things as vectors prior to this. This clears a lot of it up while generating ideas on what to do for my own projects; thanks for posting this!
Do you have a video on how to turn the Python chatbot into one that is Web-based so others can enjoy it? I've created some amazing chatbots in the Playground that I would love to make available to others on a public site.
David your work is amazing!!! Thank you. Question, how would you suggest I create a knowledge graph using case law/decisions/legislation and legal commentary on cases etc?
Are there any plans to continue this series ? The value here is great, especially for Introducing people to the possibilities. I've send this series to a dozen people by now with IT skills between "I've used HTML on myspace" up to "I know 5 languages but never touched Python". So far all of them found value in these.
Yeah, I am trying to do this full time so please have your friends support me on Patreon. If I get to the point where I can retire from my day job I will produce much more content. www.patreon.com/daveshap
Excellent video sir Do you know how to create a chatbot that is GPT based for a particular topic, say, a family lawyer website or issues of that sort? I have interacted with openai chat and it is awesome, so I would like to know how to make a similar chatbot but focused on very particular topics and areas.
I'm a patreon subscriber and I was following along with part 1 and 2 it was wonderful, but this 3rd one seems like I missed something. Super confused, from the very beginning of this video. The first 2 videos I felt like I found an amazing tutorial for beginners, but this 3rd video seems like there was no explanations or was not even geared towards a beginner. Almost like I missed some kind of explanation?
@Simba Masters Right, so after much d****ng about, I have an update on this. It turns out that cosine similarity function doesn't really have a big impact on the math, but the math is not the problem here anyway. We need to remain mindful that the dot product of the 2 vectors, is not providing us a result representing the similarity of the two input words in a classical "taxonomy of knowledge" or "ontological" sense. What it is providing us with is the similarity of "where in a giant corpus of training text the two words would be found close together". In this sense "dog" would be rank higher under a "feline" category than under a "mammal" category, because quite simply, the term "dog" appears closer in general training text to the word "feline" than it does "mammal". To resolve the issue, there is no programmatical logic change required to @dshap's code. Instead, all that is needed is to prefix the categories with "type of" so that the list of categories becomes "type of plant", "type of reptile", "type of mammal", etc. Then we see very very accurate results, and "dog" then ranks highest as a "type of mammal" rather than a "type of feline" because the WORD "dog" appears in the training data clustered closer to "type of mammal" much more frequently than "type of feline"
I've had no idea about how to use embeddings or even the value in storing things as vectors prior to this. This clears a lot of it up while generating ideas on what to do for my own projects; thanks for posting this!
Man if I weren't broke right now I'd subscribe to your Patreon. Thank you for this video.
It is super interesting when things don't go perfect. You can learn more that way.
That's the formula that my audience seems to like.
i think i got Covid now so great time to catch up with all your videos David! Keep up the good work!
david, tkx a lot for the videos, have been helping much for my engineering course, keep strong youtube is not easy!!
Love the series, thank you for your efforts
Do you have a video on how to turn the Python chatbot into one that is Web-based so others can enjoy it? I've created some amazing chatbots in the Playground that I would love to make available to others on a public site.
Thank you David.
David your work is amazing!!! Thank you. Question, how would you suggest I create a knowledge graph using case law/decisions/legislation and legal commentary on cases etc?
After my experiment I don't think a kg is the way to go. Semantic search and summaries instead
Are there any plans to continue this series ? The value here is great, especially for Introducing people to the possibilities. I've send this series to a dozen people by now with IT skills between "I've used HTML on myspace" up to "I know 5 languages but never touched Python". So far all of them found value in these.
Yeah, I am trying to do this full time so please have your friends support me on Patreon. If I get to the point where I can retire from my day job I will produce much more content. www.patreon.com/daveshap
David thank you for sharing your experience and workflow. Have you managed to find out why the resulting scores was so strange?
Thanks David :)
Another good video
Excellent video sir
Do you know how to create a chatbot that is GPT based for a particular topic, say, a family lawyer website or issues of that sort?
I have interacted with openai chat and it is awesome, so I would like to know how to make a similar chatbot but focused on very particular topics and areas.
Do you got the solution? I have the same question
I'm a patreon subscriber and I was following along with part 1 and 2 it was wonderful, but this 3rd one seems like I missed something. Super confused, from the very beginning of this video. The first 2 videos I felt like I found an amazing tutorial for beginners, but this 3rd video seems like there was no explanations or was not even geared towards a beginner. Almost like I missed some kind of explanation?
Join my Discord server here: discord.gg/VJpppYNX
Hey this link isn't working. Could I have an updated version? Thank you!
Hi @dshap it's been 7 months - just wondering if you managed to crack what went wrong here?
@Simba Masters It most certainly does! Thank you for spending the time on this much valued and appreciated response.
@Simba Masters Right, so after much d****ng about, I have an update on this. It turns out that cosine similarity function doesn't really have a big impact on the math, but the math is not the problem here anyway. We need to remain mindful that the dot product of the 2 vectors, is not providing us a result representing the similarity of the two input words in a classical "taxonomy of knowledge" or "ontological" sense. What it is providing us with is the similarity of "where in a giant corpus of training text the two words would be found close together". In this sense "dog" would be rank higher under a "feline" category than under a "mammal" category, because quite simply, the term "dog" appears closer in general training text to the word "feline" than it does "mammal".
To resolve the issue, there is no programmatical logic change required to @dshap's code. Instead, all that is needed is to prefix the categories with "type of" so that the list of categories becomes "type of plant", "type of reptile", "type of mammal", etc. Then we see very very accurate results, and "dog" then ranks highest as a "type of mammal" rather than a "type of feline" because the WORD "dog" appears in the training data clustered closer to "type of mammal" much more frequently than "type of feline"
Is your code available publicly to use?
It's all on github
@@DaveShap Thank you very much.
Nice