Keynote: The Cost of Democratization? Open Source and the Future of Ethical Use

on YouTube ↗19m

" summary

OpenAI’s GPT-2 withheld release is cited as a concrete case where a text generator can produce realistic fake news, while an open-source deepfake tool sold as deepnudes $50 app shows how easy it is to produce revenge porn. The talk advocates concrete mitigations: adopt model cards and data sheets for provenance, run benchmark attacks/defenses, architect systems forensics, and evaluate civil liberties risks after examples like the Gaydar paper and government use of facial recognition.

AI-written summary · about

" chapters (10)

1. 0:00Introduction
2. 0:46What is GPT
3. 6:15Open Source
4. 7:28Responsibility
5. 8:06Worlds
6. 9:02Nuclear
7. 10:58Law and adversarial machine learning
8. 13:13Benchmarking
9. 15:47Civil liberties
10. 17:57Conclusion

" related talks

• 13mKeynote: State of Open GenAIJim Zemlin · Cassandra Summit 2023
• 21mKeynote: "Open as in:"Zeynep Tufekci · Open Source Summit North America 2017
• 13mKeynote: State of Open GenAIJim Zemlin · AI.dev: Open Source GenAI & ML Summit North America 2023
• 28mEthical AI?!Laura Summers, Vilas Dhar, Mehran Sahami, Navrina Singh · DLD Munich 23
• 27mChallenges of Generative AIJ. Clippinger, JP Rangaswami, M. Ruiz, R. Socher, J. Thornhill · DLD Munich 23
• 01h 16mBook Launch: Regulating the Synthetic SocietyBart van der Sloot · CPDP.ai 2024
• 33mKeynote: Fine-tuning our way towards openness in AICarol Chen, Cedric Clyburn · DevConf.CZ 2025
• 59mKeynote: The past, present, and future of AI for application developersSteve Sanderson · NDC Melbourne 2025