Elaheh Ahmadi, founder of Themis AI, discusses how she and her team plans to empower the world to create and deploy trustworthy AI.
[01:15] Elaheh provided an overview of Themis AI and the genesis of the research.
[03:34] Discussion on recent developments of OpenAI and how Themis AI’s product may be relevant to the industry. There’s the anxiety in the AI community about the power of these applications. Themis AI may be able to help alleviate some of them.
[06:14] Elaheh provided an overview of the how a machine learning may fail. The vulnerability may be bucketed into 3 main categories: 1) model uncertainty 2) label uncertainty and 3) representation bias
[10:46] Elaheh took a deeper dive in the core of Themis AI’s technology. When we have a data set for AI, what could potentially go wrong? Themis AI aims to solve the problem in a general sense, instead of only solving it for a specific domain/case.
[11:41] A critical component of a good AI model should be able to tell you that if the machine doesn’t know the answer, it should tell you directly rather than making something up (hallucination). As such, there is a need for having a systematic solution to addressing these type of failures/vulnerabilities.
[16:29] What are the reasons or sources of failures of the model? That’s basically Themis AI’s approach of bringing different diagnostic tools to the end users who are machine learning engineers and scientists and helping them diagnose their models and pinpoint what is causing the model failure so that they can then further improve it.
[20:41] The company is industry agnostic because its technology is agnostic to model and data. And the profile of the target customers are basically any companies who are building their own models for different applications. Themis AI is bringing them state of the arts technology and research, the only place they can find an automated solution to make the models uncertainty aware. This saves companies a lot of time and R&D effort.
[22:55] Elaheh discussed how Themis AI could be a useful tool for companies (such as OpenAI) who focus on LLM.
[23:37] Conversation on fund raising and lessons learnt.
[30:26] Elaheh discussed the company’s beta release over the summer. It selected about 5 customers from the wait list for priority access. Now it is moving beyond those beta customers and accepting new clients and talking to many companies in different industries.
Book Mentioned in this episode
Fall in Love with the Problem, Not the Solution by Uri Levine
- Kunal Agarwal made Cyber Security a First Class Experience for his Customers - February 23, 2024
- Scot Wingo shared his Entrepreneur and Investor Journeys - February 15, 2024
- Palash Soni Discussed his Journey of Building Goldcast and Scale it to Profitability - February 7, 2024