More than *anything* the people who actually know how technology works, who actually build things, wish that people would treat LLMs like every other technology, and be normal about them. Don't build a religion about them, don't force them on people, don't ignore the problems. Just be normal.

At least they have good taste.

They are tools for thinking if after the prototype you sit and write down what you learned in the form of specifications. But most of the time people use them as a first iteration of the product, and having something tangible gives them the false impression they have the product figured out.

In a recent experiment with AI coding, the agent generated a lot of tests and achieved an impressive 80% coverage. The catch? Most tests "validated" the implementation, not the specification of the behavior. For instance, is this variable set with the input value? Pretty useless in many cases.

That's not true. Hitting a pedestrian with a bike moving at full speed can cause severe injuries to the pedestrian, more than to the cyclist, who may be wearing a helmet.

This goal is easy to achieve because AI tends to produce verbose code. So instant "productivity" just by turning it on!

Just jump out of bed when I wake up and activate myself: take shower, make breakfast (used to do 10 minutes of HIIT exercise but no more). The more I stay in bed awake, the harder it is to get out and activate.

During the pandemic this was my favorite genre of you tube videos.

Only if you bake it.

Barcelona is equally scary and dangerous. It's too late for me, but they can still avoid my mistake of moving here.