Casa de Brain

My Thoughts


(Why should a software engineer spend $100 more, on tasks that can be done for free (or < $10)?

Today's technical offloading, will lead to the evaporation of that skill from society.

AI's original goal retention capacity is not there yet. Things like changing something at 100 step without effecting other parts built in step 2 or 3 OR complex understanding of what to change in a huge pile of data OR having a resistance to change because of possible negative effects on other parts of the generated data, with a knowledge on immediate and long-term effects, is what something AI can't do.

It is good in repetitive bounded tasks, where it can learn well from documented examples but having an understanding of how something works without getting trained, is still far away.

The fact that even after all the huge amount of data, LLMs have a long way to go to understand language meaning, specially in a large context, the fact that companies STILL need more data, seals the fact that, current neural network models will not stay 100% accurate. They will fluctuate, deviate and will provide their own version of the truth, without properly understanding the implication and causal effect.

In some case someone can argue this exactly many of us DO.)