Expert opinions on Transparency

Future risk reports, system cards and external evaluations should be assessed for reproducible methods, clear separation between simulated and real actions, disclosure of missed detections, and evidence that safety interventions do not simply make pr...

For the most advanced frontier AI models, an independent, federally overseen, industry-backed organization could set safety standards and verify independent, voluntary audits, similar to how other critical sectors are managed.

It is really important, given the stakes, that people who develop AI and those who test it don’t have relationships with each other.

You’ll see industry oppose outside auditors and the main reason is they would mandate something that can’t be done [today] in a robust and uniform manner.

The following practices are mapped to Australia’s 8 AI Ethics Principles, demonstrating how governments can practically apply them to their assurance of AI. [...] Governments should identify and consult with stakeholders, including subject matter and...

For funders and international community: support and/or collaborate with local actors best placed to detect AI harm, and back efforts to establish cross-border AI incident information sharing.

Rather than creating entirely new channels, policymakers should focus on ensuring that existing channels can effectively incorporate information related to frontier models and agents. Building on trusted relationships will be faster and more effectiv...

It’s very dangerous to make this a dichotomy between complete openness all the way to complete closedness. In complex software systems as well as scientific systems it’s much more nuanced.

If I were to try to give one prescription, it would be to promote openness, because AI is amazing technology and I want it to be in everyone’s hands.

Every action and every outcome needs to be clarified and justified to a board, an auditor, or a customer.

We have a very natural instinct to say: we want as much interpretability as possible. One of the sci-fi worry cases is that agents start speaking in languages to each other that we don't understand. [...] I bet there are ways to define boundaries tha...

Explainable AI (XAI) is often offered as the answer to the black box problem and is broadly defined as “machine learning techniques that make it possible for human users to understand, appropriately trust, and effectively manage AI.” Around the world...

If you are able to prove that I did an AI video of you robbing a bank when you were running for office, and I didn’t have a disclosure, I could be fined and face up to two years in jail.

You can say dishonest things whether they're AI or not.

The way to get there is a shared machine-readable format. I call it a Safety Framework Card.

So I don't see any risk associated with open sourcing models. I only see benefits.

The physical foundations of frontier artificial intelligence, the fabrication of advanced semiconductors and the operation of computing infrastructure at the largest scales, shall be subject to international transparency, so that the concentration of...

The largest training runs have now likely exceeded 10^26 FLOP.

AI training run reporting rules should be developed through a distinct study process. We also recommend clarification of the rules to distinguish between pre- and post-training activities like fine-tuning. Fine-tuning a qualified model should not tri...