3 ms·
What is your plan for AI safety, and how does OpenAI fit into it? In http://blog.samaltman.com/machine-intelligence-part-2 http://blog.samaltman.com/machine-in
by jimrandomh 11y ago
What is your plan for AI safety, and how does OpenAI fit into it?
In http://blog.samaltman.com/machine-intelligence-part-2 http://blog.samaltman.com/machine-intelligence-part-2 you laid out sensible requirements and safeguards we should demand of late-stage AI development when it gets further along. But OpenAI's founding statement seems to have baked in a mission and vision that's incompatible with those safeguards, requiring immediate sharing of breakthroughs even if the technologies required to use them safely aren't ready yet. That founding statement used to have the word "safely" in it, but this was edited out without any sort of announcement or comment, and, to the best of my thirdhand-rumor knowledge, OpenAI is not currently doing any safety-related research.
(I wrote a blog post on this back in December, and no one has given any plausible answer: http://conceptspacecartography.com/openai-should-hold-off-on-choosing-tactics/ http://conceptspacecartography.com/openai-should-hold-off-on... )
- sama 11y agoWe're still working through this--it turns out to be a research problem in itself. We talk about it a lot, and I'm working on a post about it (for example, we may not share intermediate breakthrough that we think pose major safety risks), but I've been much busier than expected the last few months and haven't gotten to it yet. Definitely expect us to do safety research.
- darnton 11y agoWhat can a run-of-the-mill technical person do to help? We know that super-AI is more likely than not to be existentially dangerous. It would be remiss to foresee the apocalypse and assume someone else will fix it.