More on Language Models

LLMs represent many dramatic technical achievements in algorithms, machine learning, statistics, human-computer interaction (HCI), chip-design, and many other fields. They are impressive artifacts! Anyone who tells you LLMs are not impressive technical achievements is not looking carefully.

However…

You are being routinely misled about LLMs by secretive actors who want you to believe these models are more capable than they are.
The labor required for reinforcement learning with human feedback is low-paying, deeply traumatizing, and disproportionately falls on the Global South.
LLMs are already leading to degradation in online information ecosystems, human creative production, and environmental resources.
While boosters promise more leisure and creative opportunity, the long-term trend of automation and AI is to reduce wages. Generative language models are already contributing to layoffs.
Companies training LLMs are stealing from authors and publishers to assemble their training data.
LLMs systematically reproduce harmful stereotypes along axes of gender and race.
Increases in the amount of resources required to train LLMs are concentrating power in the hands of a small number of tech corporations.
LLMs regularly generate responses that are humorously or dangerously wrong.

LLMs Are Like Cars

I encourage you to think of LLMs like you think about gas cars.

Like cars, LLMs can be extremely helpful. One day, it might even be necessary to use LLMs in your daily life.
Like cars, every time you use an LLM you make a small contribution toward environmental degradation.
As with fossil-fuel cars, large companies with profit motives are regularly misleading you about the impacts and harms of this technology.
Like cars, LLMs can get you where you’re going faster. But by going fast, you might miss out on important joy, learning, or experience.