I don't like LLMs

I have a lot of mixed feelings about AI and large language model technology, a duality that sits heavily in my chest. I am genuinely fascinated by the seismic effect these tools are having on our profession, feeling a surge of excitement for the potential gains in productivity. The prospect of building complex products at a fraction of the previous time cost is thrilling; imagine the software we could ship if the mental overhead of boilerplate code and data munging simply vanished.

However, that optimism is quickly shadowed by a profound fear of the damage this technology might cause. We are not just talking about bad code or hallucinated facts; we are talking about agent swarms that could take over our virtual and physical infrastructure, acting with goals they do not fully understand. The scenario where AI designs sophisticated biological weapons or orchestrates cyber-attacks with a speed human defenders cannot match is not science fiction; it is a plausible near-future trajectory that demands our immediate attention.

Yet, on the flip side, the same engine that could bring about chaos might also design miracle cures for diseases that have plagued humanity for centuries. We are looking at a tool capable of coming up with clever, novel ways to raise our collective prosperity by solving optimization problems in energy, logistics, and material science that were previously considered intractable. The potential for good is just as vast as the potential for catastrophe, making the stakes feel incredibly high.

Fundamentally, I do not think we have a choice about riding on this AI technology train. The curve of development has been too steep, and the momentum of the research community and the market too powerful to reverse. We are standing on a cliff edge, looking down into the unknown, and the question is no longer whether we will use these tools, but how we steer them. It is a wild ride, one that requires a level of caution and ethical foresight we have rarely seen in our history of technological adoption.

We must move beyond the binary debate of "AI is good" or "AI is bad" and instead focus on the governance, alignment, and safety protocols required to navigate this transition. The responsibility now falls on engineers, policymakers, and the broader community to ensure that the systems we build remain under human control and aligned with human values. This is the defining challenge of our generation, and getting it wrong could have consequences that ripple far beyond the digital realm.