Get ready for a mind-bending journey into the world of AI agents and their potential impact on our lives. The future is here, but is it what we expected?
The year 2025 was supposed to be a game-changer, with big AI companies promising a revolution. Instead, we got a year of hype and delayed dreams. But what if the reality is even more surprising? What if the answer to 'When will AI robots take over our tasks and run the world?' is a resounding 'Never'?
A thought-provoking paper, published amidst the AI frenzy, dares to challenge this notion. Titled 'Hallucination Stations', it mathematically argues that language models have their limits. The authors, a father-son duo with impressive AI credentials, assert that even advanced reasoning models fall short. Vishal Sikka, the former SAP CTO, puts it bluntly: 'They can't be reliable.'
But here's where it gets controversial... The AI industry isn't backing down. Coding, a recent success in agent AI, is a prime example. At Davos, Google's AI head, Demis Hassabis, reported progress in minimizing hallucinations. And now, a startup called Harmonic is making waves with its mathematical approach to AI coding, claiming to have cracked the reliability benchmark.
Robinhood's CEO, Vlad Tenev, and mathematician Tudor Achim, co-founders of Harmonic, believe they've found a way to guarantee AI trustworthiness. Achim asks, 'Are we doomed to a world of AI slop?' Their solution? Using formal mathematical reasoning to verify LLM outputs. It's a narrow focus for now, but they're confident it can be expanded.
Achim believes that most models have the intelligence to handle tasks like booking travel. But is that enough for widespread adoption? Himanshu Tyagi, from Sentient, an open-source AI company, thinks not. He highlights the disruption caused by hallucinations, undermining the value of agents.
Yet, the AI powers persist. They argue that guardrails can filter out the imaginative noise, and even Sikka agrees this is a likely path forward. Achim takes it a step further, suggesting that hallucinations are necessary for surpassing human intelligence. 'Systems learn by hallucinating,' he says, 'sometimes leading to unique, human-unthought ideas.'
So, is agentic AI impossible or inevitable? Perhaps both. While hallucinations persist, the industry is determined to narrow the gap. Alan Kay, a computer pioneer, suggests we focus on the bigger picture, quoting Marshall McLuhan: 'Find out what is going on.'
We're on the brink of massive automation, but will it enhance our lives? Only time will tell, and the answer might not be found in mathematics.
This article raises intriguing questions. Do you agree with the industry's optimism? Or do you side with the critics? Share your thoughts in the comments!
