The Neural Network’s Encounter with TON
Sergey Gustun, an independent researcher, discusses AI and crypto.
Artificial intelligence is currently riding a wave of hype, yet it remains unable to fully replace professional analysts or traders. We discussed the prospects of AI at the intersection with the crypto industry and why OpenAI’s CEO Sam Altman “messed up” with GPT-5 with independent researcher Sergey Gustun.
ForkLog (FL): You previously worked as a product manager at Bitsgap, but now you’ve fully transitioned to AI. Why did you change direction and what attracted you to this new field?
Sergey Gustun (S.G.): I have always been interested in artificial intelligence. At Bitsgap, we launched a product with AI elements—it helped assemble trading bot portfolios. I used to experiment with machine learning algorithms in trading and betting. Now, amidst the hype around LLM, I wanted to delve deeper into the topic. I’m inspired by how AI boosts productivity: for instance, coding speeds up significantly. Tasks that used to take a week can now be solved in a day or two.
FL: Trading and betting involve a lot of randomness and uncertainty. Is it wise to fully rely on artificial intelligence in such fields?
S.G.: AI is an assistant, not a replacement. You can build an agent with instructions, and it might even perform well for a month, but by the second month, it could start losing money. It will solve the task at any cost, even if it has to “bend the truth” a bit. It’s simply a tool that requires constant oversight.
FL: OpenAI recently released GPT-5, and Altman himself admitted they messed up. What are the main shortcomings of the model and what bottlenecks are currently hindering progress in AI?
S.G.: With GPT-5, OpenAI clearly rushed. The model wasn’t as groundbreaking as expected, leading to disappointment—especially compared to Claude’s successes. We are nearing a ceiling: each improvement is becoming increasingly difficult, and a new approach is needed. Plus, there’s the human factor: businesses aren’t ready to fully trust AI, using it as an assistant, but they resist the idea of replacing employees. Add to this the hallucinations, and it becomes clear why progress is hindered not only by technology but also by people’s distrust.
FL: What is the most memorable “hallucination” of AI you’ve encountered?
S.G.: The funniest was when I tried to write a smart contract for TON. The neural network went into some incredible iterations, asking itself questions it couldn’t answer, and ultimately spiraled into an endless cycle of reflection. It was like an existential crisis for the machine.
FL: You’re a fan of Warhammer. How would you apply AI in video game development, beyond content generation?
S.G.: I would use AI for level design—to design locations faster and more interestingly. For example, a new desert location in WoW could easily be created with artificial intelligence. In coding, it’s already an indispensable assistant, but for creating emotions and deep settings, it’s still insufficient.
Subscribe to the podcast:
Рассылки ForkLog: держите руку на пульсе биткоин-индустрии!