I don't like LLMs

I have always been a man of cautious optimism, yet when it comes to the latest wave of Large Language Models, I find myself shaking my head with a mixture of fascination and genuine dread. The technology promises to be the biggest productivity leap since the invention of the compiler, offering the tantalizing prospect that we could build software products in days rather than months. It is an intoxicating idea, one that suggests our collective capacity to solve complex problems might finally outstrip our historical limitations, turning years of toil into mere afternoons of iteration.

However, this euphoria is easily tempered by the darker shadows that these systems cast. We are not merely talking about better autocomplete functions; we are discussing autonomous agent swarms that could theoretically take control of our virtual infrastructure, or worse, design novel biological weapons before human oversight can intervene. The speed at which an AI can iterate through millions of possibilities means that the gap between a helpful tool and a catastrophic actor is dangerously thin, and I fear we have not yet built the guardrails to keep them in check.

On the other hand, the potential for good is equally staggering and perhaps even more immediate. Imagine an AI that can design a miracle cure for a neglected tropical disease by simulating molecular interactions faster than any lab could physically test them. Picture systems that find clever, non-obvious ways to raise global prosperity by optimizing supply chains or energy grids in ways no human planner ever could. The duality of this technology is its defining feature; it holds the power to elevate humanity or to bring it to its knees, depending entirely on how we choose to wield it.

Fundamentally, I believe we do not have a choice but to ride this technological train. The momentum is too great, the investment too deep, and the underlying mathematical principles too powerful to simply ignore or suppress. To try to stop the train would be to invite chaos, but to ride blindly without a plan is to invite disaster. We must accept that AI will become the dominant force in our professional lives, reshaping how we think, build, and solve problems in fundamental ways that we are only beginning to comprehend.

The challenge for our profession, then, is not to reject the technology, but to mature quickly enough to harness it responsibly. We need to develop new methodologies, new ethics, and new safeguards that can evolve alongside the models themselves. It is a wild ride, full of turbulence and uncertainty, but perhaps the only way forward is to steer the ship with both hands, keeping one eye on the horizon of infinite potential and the other on the precipice of unintended consequences.