‘It is incredibly helpful to give LLMs its response back and ask it to find 5 ways to fix it.’

What's your job and how does it involve AI?

My job has two facets to it. One is researching companies that will either drive AI or benefit from it, i.e. to find "AI winners". Second is to work with GenAI models to build internal capabilities like RAG and analytics. The idea is that both facets complement each other: building with AI models and tools gives me a sense of which companies are building the best products, and researching the companies and products helps me identify the best tools for building our internal AI infrastructure.

What's the craziest thing about AI you've heard so far?

How Deepmind was using LLMs to solve unsolved problems in Mathematics. This for me is the most exciting benefit from AI: the generation of new knowledge. Last time we made significant progress in man's understanding of the world was when we discovered Quantum Mechanics. It led us to discover semiconductors, lasers, LEDs, you name it. As an ex-physicist, I am very invested in this topic.

Which AI tool do you personally use the most?

ChatGPT. Nothing else comes close. Second would be text-to-speech, but it is embedded inside the read-it-later app I use.

How much time do you approximately save daily by using AI tools?

For me it is not about saving time but augmenting my capabilities. GitHub Copilot is a good example. Hard to put a number but I save ~10% of my time using GenAI, some days more.

Do you prefer: ChatGPT or Gemini?

ChatGPT. Two other interesting ones for me are Phind for more coding-related tasks and Perplexity for search.

What area should AI enthusiasts keep an eye on in the near future?

I think the most interesting area to look out for is AI agents. Suppose you ask a LLM to execute a complex task, like analyse a company and tell you if its a good investment opportunity. It would need to spawn multiple-sub tasks: one that researches the financial statements, one that analyses the earnings call, another to analyse the alternative data available on the company, one to combine findings mid-stage to make a plan to fill gaps in research, and so on. Executing such multi-step, complex tasks look theoretically possible to execute with LLMs, but have been hard to build. Starting to do increasingly complex tasks autonomously would be the ultimate productivity boost.