Conversational Search Engine for your Dataset using PaLM (By: Emad Ehsan) - DevFest Lahore 2023

Talk by Emad Ehsan (https://www.linkedin.com/in/emadehsan/) at DevFest Lahore 2023 by GDG Lahore.

Transcript

1. Let’s build a Conversational Search Engine Google PaLM LLM Lahore
2. Emad Ehsan Google Developer Expert Sr. Software Engineer @ Bayer @e_mad_ehsan Lahore
4. Let’s build a Conversational Search Engine Google PaLM LLM Lahore
5. [...] [...] [...] [...] What is an LLM? 0.02 0.03 0.9 0.01 0.0 … LLMs Explained Dogs Rain Drops Fish Wind … and cats raining It’s
6. Roses are red, Violets are blue, Sugar is sweet, LLMs Explained
7. Roses are red, Violets are blue, Sugar is sweet, LLMs Explained
8. for(var i = 0; i <10; i++) {
9. for(var i = 0; i <10; i++) {
10. Modern LLMs are large.
11. Classic Natural Language Problems LLMs Explained Entity extraction Classification Summarization Sentiment Analysis Translation …
12. Explain this joke: “I used to be addicted to soap, but I'm clean now.” User: LLMs Explained
13. Explain this joke: “I used to be addicted to soap, but I'm clean now.” Model: The joke is a pun on the word "clean". It has two meanings: Model: User: The joke is funny because it is unexpected. The listener expects the speaker to be talking about the first meaning of "clean", but they are actually talking about the second meaning. This creates a surprise and a laugh. To be free of dirt or other unwanted substances. To be free of addiction. LLMs Explained
14. LLM-Powered Chat Applications Customer service Personal assistants Interactive storytelling Conversational product recommender Customer feedback collection …. A lot more … LLMs for Chat
15. LLM is unaware of our private dataset Challenges
16. A natural language search engine (chatbot) that does not upload our data Challenges
17. Retrieval Augmented Generation (RAG)
18. Credits: James Briggs
20. PaLM for Embeddings Lahore
21. Embeddings Endpoint Convert text into a list of numbers computers can understand Similar text will have similar embeddings You can use these to develop applications, like search! What is an embedding? A vector representation of a concept
24. LLM: “I cannot answer your question due to insufficient information” Limitations
25. Cannot answer meta questions Limitations
26. ai.google.dev Learning more
27. Emad Ehsan @e_mad_ehsan Lahore Thank You
29. bit.ly/pak-devs A Community of Programmers & Roboticists