Exploring Key Takeaways from the Mindstone AI Meetup in NYC

Hello readers, I attended an AI Meetup by Mindstone in New York at Relay GSE on November 1st, 2023, and it was very informative.

The meetup featured three major sessions.

• First, a talk by Alexander Vasserman, Co-Founder at SerenityGPT. He talked about several things.

  • First, he discussed the difference between traditional web search and vertical search. A typical web search is just a search on public data, whereas a vertical search could be done on personalised private/proprietary data or on a specific domain.

  • Nowadays lots of developments in this space are going on, thanks to AI.

  • Alexander further went on to discuss how SerenityGPT is implementing this vertical search, and how they are giving users options to select a specific date range, document category, and much more personalized options to search on their proprietary documents.

  • Next, he discussed about semantic search.

    • Traditional web search is mostly just keyword based search. It mostly finds the exact terms entered in the query.

    • This keyword based search is fast, and cheap, and also domain independent.

    • The difference in semantic search is that semantic search is more about meaning, not just words.

    • A hybrid search is one which submits user’s query to both the systems, and merges results together.

    • What basically happens in semantic search is that an ML model converts the text into it’s vector representation (embeddings), finds embeddings which are nearest to it by vector similarity, and those similar vectors now act as a ‘context’ to the LLM.

    • The prompt to the LLM would be something like -

      You are KB GPT, below, I have given some context. You answer user’s questions from the context only. If the answer is not in the context, you say ‘I don’t know’.  
          Context - <top_k similar vectors>
          User’s query: <user prompt>
      

    • This method of using semantic search is known as Retrieval Augmented Generation (RAG).

• Next, Joshua Wöhle, CEO at Mindstone presented how to leverage OpenAI’s GPT API for day-to-day tasks such as assignment grading for teachers, or for people who are attending a tech conference.

  • First he showcased Octoparse, which is a cool web scraping tool.

  • Joshua took the website of AI Summit, New York’s speakers page, put that into Octoparse, and Octoparse intelligently figured out that we wanted to extract information about different speakers who’ll be speaking at the conference. Octoparse went on to scrape information about all the 300 speakers, their company, their name, their post, everything.

  • He then exported this data to excel, put it on Google Sheets.

  • Joshua then showed how to leverage Add On’s in google sheets. We used the GPT add on.

  • Basically what we can do is say we have 4 columns of speaker’s data, we can create a prompt which uses data from those 4 columns in the prompt to gpt, which then results in whether we would want to meet that specific speaker or not.

  • The result was pretty interesting to watch as GPT starting rating all 300 profiles one by one out of 100.

• Next, there was a fireside chat between Joshua Wöhle, and Jenny Fielding, Co-Founder and Managing Partner at Everywhere Ventures.

  • Jenny gave a lot of insights on what they look for, in a company, in the founders when they look for investing.

  • She explained how the founder’s attitude towards building the product is more important than what actually the product idea is.

  • From a VC perspective, who’s investing in pre-seed startups which are at their founding phase, the pivotal ability of the founders, and how much they are willing to shape their product is the key.

  • She also went on to tell that they are not investing in startups which completely depend on GPT API to implement a single use case by just calling their api or such ideas. In the AI space, they’re more interested in a systems approach, where the founders don’t try to replace humans, but to augment the humans tasks, to make the job easier.

  • I got a fresh perspective towards the AI world from a VC’s point of view. It was a great fireside chat.

Overall, I enjoyed the meetup a lot, learnt a lot of new things, got a new perspective in the field I’m already in. I also had the opportunity to meet top AI Developers from Bloomberg, and Google.

Some photos -

I look forward to the exciting discussions, collaborations, and discoveries that await us. Stay tuned for more insightful content, and let's embark on this journey together!

Connect with me on LinkedIn