Matt Asay

The inherent weaknesses of large language models are reason enough to explore other technologies, such as reinforcement learning or recurrent neural networks.

Apache Airflow is a great data pipeline as code, but having most of its contributors work for Astronomer is another example of a problem with open source.

Yes, a tiny number of companies have relicensed their open source code. Let’s worry about actual problems, like security and megacompanies that contribute almost nothing.

Without skilled developers supervising AI coding assistants, they are likely to break your code rather than write it. Right now, only people can fine-tune and evaluate AI.

Take a look at this unscientific survey to discover some ways to keep developers happy. A few are even simple.

The schoolyard chaos at OpenAI could detract from corporate adoption of AI. Microsoft may be best suited to restore order and trust.

Overspending on AI infrastructure by cloud providers has some forecasting an AI bust, but there are signs that enterprises are starting to put AI to work.

Considering how much we depend on open source, we should worry about the financial health of our most critical projects.

Fueled by vibes and with stars in their eyes, enterprises are not taking the time to understand genAI’s limitations and to create their own rules-based approach.

Once we get past the disappointment that AI can’t solve everything, we can use it for things it’s actually good at.