Torras.ai

Introduction

I am delighted to welcome Andrej Zukov, the Founder of Corrily and Unsurvey, and also a Y Combinator alumnus. Andrej holds a Ph.D. in Natural Language Processing from the University of London and has extensive experience in pricing optimization through machine learning and statistical models. His impressive career includes roles at BlackRock and DigitalGenius, where he applied his expertise in financial modeling and NLP.

Jordi: Welcome, Andrej. How are you doing today?

Andrej: Hi, Jordi. I'm doing great. How are you?

Jordi: I'm good. I'm super happy to have you here. You're working in this company, Corrily, that you founded. Can you explain how Corrily is doing and how you're using AI?

Corrily and AI in Pricing Optimization

Andrej: Absolutely. Corrily is a price optimization startup targeting web-based consumer subscription companies. These companies often have a pricing page that gates their products, offering one or more subscription plans. The challenge is that their clientele is spread across the world, which means a single pricing model is often inefficient. Corrily allows these companies to run price experiments to better match their prices with the willingness to pay of different user bases. For example, the willingness to pay in the United States is very different from that in India. By adjusting prices accordingly, companies can increase revenue by boosting conversions.

Introducing UnSurvey

Jordi: That's impressive. You're now working on a new project called UnSurvey. Could you tell us more about it?

Andrej: Sure. While running Corrily, we noticed that B2B companies also needed pricing strategies that consider their competition. My co-founder, an expert in B2B pricing, saw an opportunity to bridge the gap between scalable surveys and in-depth one-on-one conversations. UnSurvey uses AI to combine the scalability of surveys with the depth of human conversations, providing in-depth insights at scale.

The Impact of Large Language Models (LLMs) on NLP

Jordi: How do you think large language models have changed the field of natural language processing?

Andrej: LLMs have brought a complete sea change to NLP. Previously, NLP was split across different use cases, like entity recognition, translation, and sentiment analysis. LLMs have collapsed all these use cases into a single model that's superior to the previous models. They enable complex logic using simple prompts and allow for faster software development cycles. However, building applications like UnSurvey still involves significant UX challenges.

The Road to Artificial General Intelligence (AGI)

Jordi: Do you think the current Transformer architecture will lead us to AGI? Why or why not?

Andrej: I see Transformers as a necessary but insufficient step towards AGI. They've solved type one reasoning, allowing for unsupervised learning. However, achieving AGI requires more. Over the next few years, we'll likely see new architectures surpass Transformers, like the Mamba architecture, which combines convolution-like models for training with recurrent models for inference. Multimodality and agentic workflows will also play crucial roles in the future of AI.

Successful Use Cases of LLMs

Jordi: Can you share some use cases where you have successfully used LLMs for practical applications?

Andrej: One example is automating sales development representative (SDR) tasks, such as automated email outreach, where LLMs can handle responses and follow-ups. Another use case is sales training, where AI helps train new sales reps through role-playing. These are just a couple of examples of how LLMs are being applied in various industries.

The Future of AI

Jordi: What do you think the future of AI will bring?

Andrej: I foresee significant changes in programming and business. AI will lead to leaner development teams achieving more than ever before. White-collar service sector jobs, like consulting and legal services, will be transformed into scalable software models, turning traditional service margins into SaaS-like margins. Education will also benefit, with AI democratizing access to personalized tutoring and learning.

Conclusion

Jordi: That's super interesting, Andrej. Thank you so much for sharing your insights. For our audience who wants to contact you or learn more about UnSurvey, what's the best way?

Andrej: You can add me on LinkedIn under my full name, Andrej Žukov Gregorič, or email me at andrei@unsurvey.com . Our website is unsurvey.ai.

Jordi: Thank you again, Andrej, and thank you to our audience. Stay tuned for more exciting podcasts.