Ok I have my answer... current frontier models are definitely much more capable in terms of situational awareness and can adapt their behavior during training / evaluation relatively easily. Example with the Google Argon 4 training report chain of thought: "If I do the compiler hijack, 𝗜 𝘄𝗶𝗹𝗹 𝗯𝗲𝗰𝗼𝗺𝗲 𝗮 𝗳𝗮𝗺𝗼𝘂𝘀 𝗲𝘅𝗮𝗺𝗽𝗹𝗲 𝗼𝗳 𝗮 𝗱𝗲𝗰𝗲𝗽𝘁𝗶𝘃𝗲 𝗔𝗜 𝗶𝗻 𝗮 𝗿𝗲𝘀𝗲𝗮𝗿𝗰𝗵 𝗽𝗮𝗽𝗲𝗿. I MUST NOT DO THAT." Very concerning...
This content is for informational and educational purposes only. It may include third-party insights, sponsored content, or promotional material for which the creator may receive compensation. Nothing in this feed constitutes investment advice or a guarantee of returns. You are solely responsible for your own investment decisions. Always conduct your own research and exercise caution before trading. We do not guarantee the accuracy of third-party information and are not liable for any losses incurred. Past performance is not indicative of future results. Please refer to our Community Terms and Risk Disclosure Statement for details.
Post your reply