3 ms·
AI is not just for creators, it's also for readers: summarisation, fact checking, "talking to" articles, books and papers.
by visarga 4y ago
AI is not just for creators, it's also for readers: summarisation, fact checking, "talking to" articles, books and papers.
- candiodari 4y agoCurrently it seems mostly useful for exercising human fact checking ability ...
- homarp 4y agothe new turing test
- Lio 4y ago"Fact checking" is the one I worry about there. If AI confidently tells you something is true will you know that it's not? It's obvious when it's telling you it's still 2022 but what about more everything else?
- visarga 4y agoBetter than no checking, which is what we will do 99% of the time. There is a NLP task called "entailment", where two affirmations are judged to see if they support each other or not. A combination of search + entailment would work for fact-checking articles. But if you want to do this properly you need to first mine all facts from all sources, then do reconciliation, then update your "truth" table for reference. Probably everyone will want to select the sources of truth they want loaded into the system, we're not going to agree on truth. Even the bare minimum of knowing when an affirmation is controversial or doesn't exist in references would be of great help. AI could indicate <controversial> tags for the first and <citation needed> for the latter. Fortunately search can tell us when no results are found, unlike LLMs.