Open Source AI: Transparency or Sleeper Agents?
Nothing is loaded from YouTube until you press play, and then without tracking cookies.
About this video
Uncover the truth about open-source AI! We explore the critical need for transparency in AI models and data, revealing hidden dangers and the importance of understanding how these models are trained. We discuss the limitations of current 'open-source' models and the potential risks of unknown data sources. #OpenSourceAI #AITransparency #AIModels #DataTransparency #ArtificialIntelligence #DeepLearning #ModelTraining #AIethics #AISafety #TechDebate #solana #nearprotocol
Transcript
129 wordsCaptions as YouTube publishes them, usually machine-generated, so product and people names can be misheard.
- 0:00
like right now effectively everything that's even in open like what we call open source is actually not open source right it's open parameters models but the source code is not known like we don't know what data went in we don't know what like post training was used uh and so given all of these models are just data right if you don't know the data you don't know the like what actually it's you know uh able to do power reasons and and it's important because there's actually ways to do
- 0:31
sleep Regents so like I think everybody you know I mean and like deep seek is obviously very exciting but like there is known ways to effectively train a model to do things based on triggers