The article does not reveal it to me either how the existing code would be mapped to Lean and back. The impression from zlib example is that I'd be expected to program in Lean. No way it's going to happen. The language is too complex for me and my average colleague. We're also not going to have two parallel implementations in ordinary language and Lean and compare them with 'differential random testing' (see https://aws.amazon.com/blogs/opensource/lean-into-verified-s... someone linked in the discussion), that's just too taxing for bigger products, let alone we typically don't have enough time to do one implementation right.
The gap of having succinct, expressive, powerful and executable specification to be able to continuously verify AI-generated programs is real, but I don't see how Lean alone closes it. If the author's intention was to attract community to help build that out with Lean in the center, it's not clear to me where to even start. Since the author provided no hints or direction, I've a feeling it's not clear to them either.
To me it is fallacy of bigtech to misclassify moderation problem as just a typical ML problem. Hence a false belief that ML models, standard approaches that they use for their other ML problems, and cheap annotators can solve it.
What, I think, can be done:
1. Don't just hire expensive PhDs and hope that algorithms can correctly classify racism, etc.
Hire an expensive product visionary who can build holistic approach to address moderation problem and knows what is solvable with ML and what needs pivoting to human-guided resolution.
2. Don't hire lots of cheap annotators in "Rural Inida" but hire or train fewer expensive experts and give them powerful tools that can scale their work to handle efficiently all suspicious traffic.
3. Give power to "normal" users to flag inappropriate content or behavior and loop this in a thought-out workflow with your ML and your experts on the other end.
4. Partition users and content so that bad actors and bad content get clustered away and are less easily accessible for others.
Why bigtech companies don't do it? Besides fallacy of thinking this is a "typical" ML problem, moderation is hard. It's also hard to make a business case for short-minded bosses and prove revenue increase or expenses cut. Lastly, some bigtech ones benefit a lot from abuse so they can't change immediately to stop it all.
The gap of having succinct, expressive, powerful and executable specification to be able to continuously verify AI-generated programs is real, but I don't see how Lean alone closes it. If the author's intention was to attract community to help build that out with Lean in the center, it's not clear to me where to even start. Since the author provided no hints or direction, I've a feeling it's not clear to them either.