Having grown a large healthcare review platform, I can attest to the success we had mapping specific policy violations to natural language is incredibly useful. At scale, patients having terrible situations and/days can write about in ways that can be deeply unhealthy for the community or the doctors reading/receiving the feedback and sometimes very threatening beyond that purposes for the community. We built a custom ML engine to handle our levels of traffic for reviews, which was among the largest in the US typical ranking top 3 on Google for the domain keywords. Back when BERT was the edge, a policy-adaptive model like this one from Mistral would have been an incredible cold-start solution. Most sites never have the massive volume nor budget needed nor skillset needed before you can train domain-specific models that outperform OOTB solutions. Generally, most people and site mean well and try to do well, so empowering those people with models like this can help the collective in my opinion, so I’m happy to see this released in this manner myself.
A bit of editorial and cultural note from a US native, the subsection of the original article “Teach discrimination, not memorization” is better worded as something like 'Differentiation' or 'Distinction' instead of ‘Discrimination’. In English, the word 'discrimination' can (and in this social context may) imply social prejudice or unfair treatment. I think this may have been a bit of carry over from the rather benign French translation of “Enseigner la discrimination" which I also see awkwardly translated in the paper as well.
Discriminative has a meaning in machine learning that I think is relevant here. There are "generative" models like LLMs that are learning joint probabilities P(X, Y) and "discriminative" models like logistic regression that learn conditional probabilities P(Y | X)
Thanks for the clarification. I’m very happily wrong here, and I appreciate the correction. This does emphasize from an editorial review that the double meaning can be distracting for someone who isn’t very deep in the terminology of this particular area, so it may still be worth rewording the subheading for a broader audience which will be interested in this model.
