Lifting the Hood on Big AI - The Future of Transparency and Accountability in AI
summary
Panelists show concrete findings and tools: adversarial audits using scraping and bots reveal reliability gaps (one project finds one-third errors in a deployed copilot over two months), an open-LLM leaderboard evaluates models across 14 dimensions of openness, and Hugging Face hosts about 600,000 public models with model/data cards and leaderboards. Policy levers matter: DSA Article 40 creates structured researcher access for very large platforms (45M+ users), while safe‑harbor and pre‑release red‑teaming programs remain inconsistent.
AI-written summary · about
chapters (13)
- 0:00Transparency in AI
- 4:17AI forensics and auditing
- 7:53Evaluating model openness
- 13:11Hugging Face’s open AI
- 16:58Researcher access programs
- 22:21Challenges in AI auditing
- 28:22Approaches to AI openness
- 33:20Best practices in AI eval
- 39:19Barriers for external researchers
- 46:44Policy gaps in AI access
- 52:00AI forensic access regime
- 55:57Risks of open washing in AI
- 1:02:03Q&A on explainable AI
related talks
29mThe Open Source Imperative in AI: Transparency, Governance, and Global Collaboration
01h 13mTowards a Safe Harbour for Public Interest AI Research
44mAI, Corporate Responsibility & Democratic Legitimacy
18mDeveloping AI in a privacy conscious world
11mKeynote: Trusted AI: Bringing Trust Back into AI through Open Source
39mA Hopeful View of Truly Open Source AI
28mEthical AI?!
48mPanel: Why Open Source AI Matters for Europe