PaLM 2 Meets LangChain

[@dylanhashem1894]

Thanks for another great tutorial Sam!

[@Spock-AI]

Nice job Sam! I use both Vertex AI and Langchain so I found this video very useful. Thanks. Your right insofar the difference between the two models are the search results as I have read the article from the first search results and it's report was accurate insofar as the article. Since the difference between ability to answer from the content was which article it chose to answer the question, I wonder if in time it will be able to retrieve more than one article for it's answer source. Insofar as the math part, if I was not going completely private, I would include Wolfram's API to give me accurate results with any math problems. Anyway, I put together a copy of DemoGPT yesterday. It automates the making of demos of different models based on the contents of it's templates. Would love to see you do a video of it and making new templates for different models. Thanks again and best wishes.

[@samwitteveenai]

I will check it out. Just did a search and it certainly looks interesting. Thanks for the tip.

[@Spock-AI]

@@samwitteveenai No probs. With your skills it should be pretty easy for you and interesting for us if you decide to do it.

[@Pingupinga123]

You can try adding system prompting to make your agent get multiple sources of news before coming into conclusion.

[@Spock-AI]

@@Pingupinga123 Good point

[@EstebanAstudillo]

Great video as usual! I noticed that for the alphabet question (odd letters), the program is looking at the string's index starting at 0, hence the odd indices are indeed B, D, F, etc. How would you change the prompt to hint it to use the other index convention? Interesting.. great stuff!

[@samwitteveenai]

Good point!

[@officialseethesky]

Can we use it for commercial purpose, like creating our own project using PaLM's API to earn money from it?

[@lskua6808]

Hi Sam! For the chat retrievalqa with chat-bison model, could you share a tutorial or your codes on how we could integrate this with a front-end user interface like streamlit? Thanks!

[@SalehShamoon-mk5kr]

I get list index out of range when im trying to run the qa chain. Please help

[@georgekokkinakis7288]

Great tutorial. Thanks. By the way do you know any open source LLM which supports the Greek Language? It would also be very helpful if you do a tutorial about how we can use langchain with Petals in order to run big LLMs which doesn't fit in ram of free colab.

[@samwitteveenai]

Really good question about Greek, a lot of it has to do with the tokenizer supporting the language well. You can look at Bloom or MT5 if you need a generation model or something like GreekBERT if you are classifying language etc. github.com/nlpaueb/greek-bert There are probably other but I don't know them. I will checkout Petals I have heard about it but not tried it.

[@hacktry4157]

hey can you share video with makersuite ?

[@samwitteveenai]

I have already uploaded one with MakerSuite, take a look at my recent vids

[@fabianaltendorfer11]

Can I also use this if the documents contain many screenshots, is there a way to feed screenshots in? Thx for the video btw, very interesting!

[@samwitteveenai]

Not this model, but there are others and a some more coming that allow for that.

[@TzaraDuchamp]

About the Alphabet, maybe it starts counting from zero. That would make B odd.

[@azmo_]

What's the price for palm 2 api calls compared to ChatGPT?

[@samwitteveenai]

cloud.google.com/vertex-ai/pricing 0.1 cents per 1k characters (not tokens)

[@torstenschindler1965]

Scientists are used to provide literature references. How do you include these in the search results? How do you keep them in memory?

[@harigovind511]

You can consider each query to the LLM as a small instruction, it could be something like: “Answer the question at the end based on the below context. Context: {REFERENCE} Question: {QUESTION} You can use this prompt with a Retrival Question Answering chain, or define the retriever as a Tool and use it with an Agent Chain. Hope this helps.

[@user-wr4yl7tx3w]

I guess, when you restart, it doesn't mean you have to install again?

[@Krishna-ue3bo]

yes, you don't have to install it again