I don't like LLMs

I have a lot of mixed feelings about AI and large language model technology. I'm fascinated by its effect on our profession, excited by the potential gains in productivity, and the possibility of rapidly building products that were once deemed impossible. Yet, simultaneously, I feel a profound unease regarding the damage such powerful systems might cause, from agent swarms taking over our virtual infrastructure to the theoretical nightmare of designing biological weapons. This tension isn't just a philosophical musing; it is the defining characteristic of our current era.

On one hand, the promise is undeniable. LLMs offer a shortcut to understanding complex domains, allowing developers to prototype faster and engineers to solve problems with a breadth of knowledge previously locked behind years of study. If we can leverage these tools to design miracle cures for intractable diseases or come up with clever ways to raise our collective prosperity, the potential return on investment for humanity is staggering. The technology represents a leap forward in our ability to process information and generate solutions, acting as a force multiplier for human ingenuity.

However, the risks are not abstract; they are tangible and escalating. We are entering an age where the barrier to creating sophisticated, autonomous systems is vanishingly low. An unaligned model could optimize for a goal in a way that destroys the very infrastructure it was meant to serve. The fear of "agent swarms" is not merely science fiction; it reflects a real concern about how decentralized, intelligent agents might interact with power grids, financial networks, and communication systems without human oversight, potentially leading to cascading failures that are impossible to reverse.

Fundamentally, I don't think we have a choice about riding on this AI technology train. The trajectory of the tech giants, the academic research, and the commercial incentives all point toward an ever-expanding integration of these models into the fabric of daily life. To try to stop the train would be to abandon the future of computing itself, leaving us behind as the world accelerates. We cannot opt out, so we must instead learn to navigate the wild ride with our eyes open.

The challenge ahead, then, is not about rejecting the technology but about mastering the governance required to keep it on a safe path. We need new forms of verification, rigorous testing frameworks, and ethical guardrails that evolve as quickly as the models do. It requires a global conversation about how we deploy these tools, ensuring that the benefits of rapid productivity are shared fairly and that the existential risks are mitigated before they materialize.

Ultimately, the story of AI is a story of human responsibility. We are the ones holding the steering wheel, even if the engine runs on silicon and electricity. If we approach this with a mix of optimism and extreme caution, we might just steer the technology toward a future that heals and uplifts. If we rush, driven only by profit or convenience, we risk a future where the very tools meant to save us become the instrument of our downfall. The outcome depends entirely on the choices we make today.