Post #3882612

2026-07-17 10:12 UTC

I don’t like playing the futurologist, but after seeing AI companies warn EU institutions about the supposed risks of open-weight models, I suspect some are lobbying to regain control over genuine open source and open-weight AI. Don’t fall into the trap: the greater danger lies in opaque, proprietary models, not open-source ones. #opensource #ai #cybersecurity

Replies (2)

@flo@mi.cmbg.ws 2026-07-17 10:15

@adulau@infosec.exchange They are quite obviously worried about their business model, cf. also the open letter co-signed by AI company economists.

Open ##3882705
@tyzbit@toot.now 2026-07-17 15:43

@adulau@infosec.exchange I say this with respect and after searching earnestly myself - what large language models are open source? DeepSeek? Qwen? Those aren't open source, although the term has been used for them. Open source means releasing all of the artifacts you need to build your own, and I don't know of a single mainstream LLM that also released the training data used. Again, respectfully and not trying to sealion, do you know of any mainstream models like this? If there isn't any, both proprietary and open-weight models are outside the reach of individuals without a large corpus of training data and are effectively the same: something you rely on others to make for you. The only difference is open models can be shared.

Open ##3891088