Lewis Ho

As we build increasingly powerful AI models, we’re committed to responsible development. We’re implementing our latest Frontier Safety Framework – our most comprehensive approach yet for identifying and staying ahead of emerging risks. Find out more → goo.gle/3W1ueFb

Per our Frontier Safety Framework, we continue to test our models for critical capabilities. Here’s the updated model card for Gemini 2.5Pro with frontier safety evaluations + explanation of how our safety buffer / alert thresholds approach applies to 2.0, 2.5, and what’s coming. storage.googleapis.com/model-cards/do…

AGI could revolutionize many fields - from healthcare to education - but it's crucial that it’s developed responsibly. Today, we’re sharing how we’re thinking about safety and security on the path to AGI. → goo.gle/3R08XcD

My ep w @AllanDafoe (director of frontier safety & governance at DeepMind): "Tech doesn't force us to do anything, it merely opens the door – and it's military-economic competition that forces us through." (32:25) "We're not at peak returns to generality." (1:38:06) "The counterposition I would offer is: you don't want to equip groups trying to shape history with a naive model of what's possible..." (20:31) "The agricultural revolution, evidence suggests, was not great for a lot of people... probably health and welfare went down." (13:27) "He gave a demonstration of cannons bombarding the shore... The Japanese asked him, 'Will you bring your ships again?' And he said, 'I'll bring more.'... That was the opening of Japan. It led to a 15-year period of revolution." (40:17) "I'm glad you think it's obvious. Let's not underestimate the bias that comes from a scientist using the tools that they prefer to use..." (33:30) "In my view, Demis and Shane are extremely impressive, from their safety orientation to… their wisdom and broad character. AGI is probably the most important historical development… Demis and Google DeepMind are very likely to be influential" (7:40) "...imagine we had no investment in AI alignment. Maybe it [only] delays it 2 years until the market demands we solve this problem..." (51:50) "We can talk about what I've called the 'super-cooperative AGI hypothesis': that as AI scales to AGI, so will cooperative competence scale to sort of infinity" (1:08:50) "Increasing the cooperative skill of AI will make those AI systems better off, but it may harm any agent who's excluded from that cooperative dynamic" (1:25:09) "...the Frontier Safety Framework I don't think is cheap talk. It really does represent an expression of Google DeepMind's sense of the nature of the risks..." (2:29:05) "...this is a new phenomenon, something new under the sun. It's quite incredible that finally the safety has reached the number of nines required..." (2:40:48) Links below, enjoy! 4:31 Why join Google DeepMind over everyone else? 9:30 Do humans control technological change? 38:11 Competition took away Japan's choice 1:03:34 How AI could boost cooperation between people and states 1:08:44 The super-cooperative AGI hypothesis 1:43:39 It matters what AGI learns first vs last 1:48:24 How Google tests for dangerous capabilities 2:00:57 Evals 'in the wild' 2:15:05 DeepMind's strategy for ensuring its frontier models don't cause harm 2:18:52 How 'structural risks' can force everyone into a worse world 2:29:01 How much do AI companies really want external regulation? 2:40:48 How AI could make life way better

As we make progress towards AGI, developing AI needs to be both innovative and safe. ⚖️ To help ensure this, we’ve made updates to our Frontier Safety Framework - our set of protocols to help us stay ahead of possible severe risks. Find out more → goo.gle/42IuIVf