This is the second episode in our data-driven VC’s series. Mike is co-founder and Applied AI at Moonfire - one of the pioneering VC firms in the field of data-driven investing. Mike is a hacker at his heart and we go into some interesting technical details in our conversation.
You can listen to Starwatcher podcast over here or pick you favourite platform:
Show notes and topics we are discussing:
- Mike's background as a hacker, engineer and becoming a VC
- Establishing Moonfire with data as a core fabric of the firm
- Sourcing, screening and evaluation as a machine learning driven recommendation problem
- Adopting transformer architecture from very beginning.
- BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding
- Universal approximation theorem
- About the stage of Data-driven investing industry
- Data-driven VC landscape report
- Analysing founders vs. analysing companies. Units of calculation at different stages of company life cycle.
- How LLM's have affected people analytics
- Importance of data across sourcing, screening and evaluation cycle
- Thesis-driven approach
- Having prepared mind in conversations with entrepreneurs
- Importance of identifying which companies don't fit thesis.
- Trusting data vs. making an own opinion
- Assisted expert decision making process
- Building tools to assist human expert
- Agent based approach
- Impact of public Large language models
- Story about improving venture scale classifier model by 20%
- Performance of GPT4 vs. in-house models
- API (outsource) models vs in-house models
- Giving LangChain access to internal tools
- Diminishing AI advantage in VC industry
- AutoGPT and "Thought, Action, Observation" loop
- Alignment problem
- Why it is actually valuable to be kind with LLM's
- Dramatic change of intellectual labor market
- Importance of humans to humans experience
- AI from startup perspective
- Rapid change of information exchange and expressing human emotions
Share this post