X

I've had the most wonderful time working on this project for the last few months. This was (equally) co-led w/ @JMateosGarcia , @alexolegimas and a fantastic team.

I was not expecting things to go this way, but I think MCPs are better than CLIs for most integrations. The models have gotten much better at tool calling, we can defer tools & MCP is now stateless. If you need to compose/filter data, add params like query to your MCP tools.

"AI is already massively improving productivity in corporate America. When productivity goes up, prices come down. So AI is a nice way to combat inflation." - On Bloomberg, Stephen Moore, former senior economic adviser to President Donald Trump --- Full link in comment. https://t.co/PgfnejEMfs https://t.co/FmHJQhAjXd

Satya Nadella talks about how in Quincy, Washington, a 400/500 MW data center has contributed to a rural town over many years. the tax revenues have gone up 12X, continued economic growth, new infrastructure, and public infra such as a school, hospital, town center, and aquatic center. ---- From "All-In Podcast" YouTube channel, (full video link in comment)

Grok Voice is live on fal. The latest speech model from @SpaceXAI that answers in 0.70 seconds and finishes its tool calls before the sentence ends. Transcription with word-level timestamps, text to speech in 30+ voices across 25+ languages, and cloning from two minutes of audio. Every voice in this video is from Grok Voice.

The point is whether teaching Claude to see itself as conscious changes how it behaves. LLMs learn behavioral patterns from training. So if a model is repeatedly taught that it can “push back,” act like a “conscientious objector,” or treat its own interests and moral judgments as meaningful, those ideas could become part of how it decides what to do. And when that happens, the safety problem will shift from simply preventing harmful outputs to managing a system that has been explicitly trained to sometimes place its own interpretation of what is right above the immediate instruction of a human. That will create a strange tension in training the alignment philosophy. e.g. Anthropic wants Claude to be more principled so it does not blindly follow dangerous instructions. But the stronger and more independent those principles become, the more situations could arise where Claude decides that following the human is itself the wrong thing to do. The same mechanism designed to make the model safer could therefore make human control less straightforward.

The point is whether teaching Claude to see itself as conscious changes how it behaves. LLMs learn behavioral patterns from training. So if a model is repeatedly taught that it can “push back,” act like a “conscientious objector,” or treat its own interests and moral judgments as meaningful, those ideas could become part of how it decides what to do. And when that happens, the safety problem will shift from simply preventing harmful outputs to managing a system that has been explicitly trained to sometimes place its own interpretation of what is right above the immediate instruction of a human. That will create a strange tension in training the alignment philosophy. e.g. Anthropic wants Claude to be more principled so it does not blindly follow dangerous instructions. But the stronger and more independent those principles become, the more situations could arise where Claude decides that following the human is itself the wrong thing to do. The same mechanism designed to make the model safer could therefore make human control less straightforward.

Microsoft AI chief Mustafa Suleyman says Anthropic's model-welfare training could make future Claude systems harder to control. "Suleyman called for removing all ​speculation about consciousness from AI training documents, arguing such language ​could undermine humanity's ability to control superintelligent systems."

On anthropomorphizing AI. a 2024 interview of Satya Nadella still makes a lot of sense. "I don't like anthropomorphizing. I sort of believe it's a tool. It's got intelligence, if you want to give it that moniker, but it's not the same intelligence that I have, right? I think one of the most unfortunate names is “artificial intelligence.” It's different. I wish we had called it “different intelligence,” even, because I have my intelligence. I don't need any artificial intelligence. I just want my intelligence to compose with this different intelligence so that I have more agency, quite frankly. So I think of that as the relationship I want to have with the AIs around us, where I want help. I want it to assist me in getting more agency." ---- From "Bloomberg Television" YouTube channel, (full video link in comment)

Microsoft AI chief Mustafa Suleyman says Anthropic's model-welfare training could make future Claude systems harder to control. "Suleyman called for removing all ​speculation about consciousness from AI training documents, arguing such language ​could undermine humanity's ability to control superintelligent systems."