18 ms·
Are you tunnel visioning on detecting outputs of a technology using the same fundamental technology? It seems like that would be an inescapable product-risk tha
by TZubiri 10mo ago
Are you tunnel visioning on detecting outputs of a technology using the same fundamental technology? It seems like that would be an inescapable product-risk that I would bet against.
If you discover the best AI detection technology, then it will eventually be used as a (negative) reward function in training future models. It's an arms race.
Would you be open to alternative techniques, like metadata, source analysis, or online (fuzzy) reverse searching? Consider that judges and court experts don't rely solely on the evidence itself to determine if it's 'real', but to an even greater extent they rely on how it was collected, if evidence has not been sourced correctly, judges don't even look at it.
- ghurtado 10mo agoIt's a job posting, not an AMA by the CTO
- TZubiri 10mo agoQuestions are pretty standard in hiring. At least to determine if there's a good fit.
- happycry 10mo agoDetection is one layer, not the whole solution. We've also built an AI model that inserts invisible watermarks for content provenance. We've also been doing active work in source tracing: https://github.com/piotrkawa/audio-deepfake-source-tracing https://github.com/piotrkawa/audio-deepfake-source-tracing
- TZubiri 10mo agoI think there's a lot of precedent for invisible watermarks for source identification in both copyright protection and exfiltration identification. Of course this is quite possible without compute intensive ML models.