from Ethan Mollick | by Ethan Mollick

As someone who has spent a long time studying open source (recently open weights) as an academic, I am a bit surprised by how much it is being embraced by large firms releasing AI models. I absolutely get why scientists want it but company strategy doesn’t always match OS goals.

OS is great for platform building, getting community work you can exploit, and for judo moves against competition. A lot of OS AI releases seem likely to be useful to competition and don’t lead to robust and durable platforms.

The argument I hear most is that it attracts developers, but that is fungible- pay enough and/or let people play with interesting ideas and smart people and you can still get talent. See OpenAI & Anthropic.

It seems you can learn from having one other firm do open models. Mistral alone would be enough as a basis for community research that Microsoft, Google & Meta can learn from. But if you were going to build a competitor against Meta’s AI efforts, you would use Llama to do it.