Things we learned about LLMs in 2024

This overview is useful for anyone interested in starting 2025 with a good basic understanding of where things stand with artificial intelligence (and large language models specifically). It is especially useful for people who think they already have a good understanding. Because here’s the thing with a technology advancing as rapidly as this one: Your knowledge can become outdated quite quickly. And few things are more annoying than being confidently wrong about something that was true when you last checked (and you stopped checking). So, the state of LLMs as we enter 2025 – what they can and cannot do, the collateral damage they might cause, and what to expect in the near future. One thing that remains true: «The key skill in getting the most out of LLMs is learning to work with tech that is both inherently unreliable and incredibly powerful at the same time.»