I don't like LLMs

I have a lot of mixed feelings about AI and LLM technology, a sentiment that feels almost contradictory yet entirely accurate to the current moment. I am deeply fascinated by its effect on our profession, excited by the potential gains in productivity that could allow us to build products at a pace previously thought impossible. The sheer velocity of iteration offered by these models is intoxicating, promising a future where the bottleneck of human cognition is finally broken.

On the other hand, I am fearful of the damage AI might cause, a fear that is not merely theoretical but rooted in the tangible realities of modern infrastructure. The idea of agent swarms taking over our virtual and physical infrastructure, making decisions based on flawed objectives or recursive loops, is a nightmare scenario. We are essentially handing over the keys to our digital civilization to systems that lack common sense, empathy, or an inherent moral compass, designing tools that could theoretically be repurposed to design bio-weapons or destabilize global supply chains with terrifying efficiency.

Yet, despite these genuine terrors, the flip side of the coin is just as heavy and compelling. Fundamentally, LLMs might also design miracle cures for diseases that have plagued humanity for centuries, coming up with clever ways to raise our prosperity through optimization problems that no human mind could solve alone. The potential for good is vast, ranging from personalized education that adapts to every child's learning style to logistical solutions that could eradicate poverty through hyper-efficient resource distribution.

Fundamentally, I don't think we have a choice about riding on the AI technology train. We are not in a position to simply turn off the engine and wait for a safer, slower alternative to emerge; the technology is already embedded in our economy, our research, and our daily workflows. To reject it entirely would be to cede the future to those who do accept it, potentially locking out the very benefits that could solve our most pressing problems.

It's a wild ride, one that requires a level of vigilance and technical literacy that we currently struggle to muster. We are building these systems with incredible speed, often outpacing our understanding of their limitations and risks. The challenge lies not in stopping the train, but in ensuring that we are the ones steering it, equipped with the guardrails and ethical frameworks necessary to navigate the treacherous terrain of artificial intelligence. I just hope we'll figure it out before the consequences become irreversible.