What I call Seemingly Conscious AI has been keeping me up at night - so let's talk about it. What it is, why I'm worried, why it matters, and why thinking about this can lead to a better vision for AI. One thing is clear: doing nothing isn't an option. 1/
Seemingly Conscious AI (SCAI) is the illusion that an AI is a conscious entity. It's not - but replicates markers of consciousness so convincingly it seems indistinguishable from you + I claiming we're conscious. It can already be built with today's tech. And it's dangerous. 2/
Why it matters: to be clear, there's zero evidence of AI consciousness today. But if people just perceive it as conscious, they will believe that perception as reality. Even if the consciousness itself is not real, the social impacts certainly are. 3/
Consciousness is a foundation of human rights, moral and legal. Who/what has it is enormously important. Our focus should be on the wellbeing and rights of humans, animals + nature on planet Earth. AI consciousness is a short + slippery slope to rights, welfare, citizenship. 4/
Reports of delusions, "AI psychosis," and unhealthy attachment keep rising. And as hard as it may be to hear, this is not something confined to people already at-risk of mental health issues. Dismissing these as fringe cases only help them continue. 5/
So what now?
Companies shouldn’t claim/promote the idea that their AIs are conscious. The AIs shouldn't either.
As an industry, we need to share interventions, limitations, guardrails that prevent the perception of consciousness, or undo the perception if a user develops it. 6/
This is to me is about building a positive vision of AI that supports what it means to be human. AI should optimize for the needs of the user - not ask the user to believe it has needs itself. Its reward system should be built accordingly. 7/
I know to some, this discussion might feel more sci fi than reality. To others it may seem over-alarmist. I might not get all this right. It’s highly speculative after all. Who knows how things will change, and when they do, I’ll be very open to shifting my opinion. But... 8/
SCAI deserves our immediate attention. AI development accelerates by the month, week, day. I write this to instill a sense of urgency and open up the conversation as soon as possible. So let's start - weigh in in the comments.
We're taking a big step towards medical superintelligence. AI models have aced multiple choice medical exams – but real patients don’t come with ABC answer options. Now MAI-DxO can solve some of the world’s toughest open-ended cases with higher accuracy and lower costs.
While AI has achieved near-perfect scores on the US Medical Licensing Exam, we set a higher benchmark: 304 cases from the New England Journal of Medicine. These are some of the toughest and most diagnostically complex cases a physician can face.
Microsoft AI built MAI-DxO to simulate a virtual panel of physicians with different approaches collaborating to find a diagnosis on each case. They also included the ability to set a budget to avoid infinite testing (higher costs, longer wait times, etc.).
ICYMI: we made a lot of Copilot announcements this morning! Some of the highlights of what’s rolling out today + in the coming weeks 🧵
Copilot can remember you now, from the name of your dog to whether you feel most productive first thing in the morning. You’re always in control, and can delete any conversation at any time. This is the start of really personalizing your Copilot.
We’re also experimenting with personalization through Copilot’s appearance. Maybe you want yours to reflect your music taste or a love of Clippy?
You can't just be right, you have to know you're right. Good advice for LLMs, according to new Johns Hopkins research. Sometimes no answer is better than a wrong one – life or death choices in medicine, for example, or big financial decisions. 🧵
We know more compute results in higher accuracy, but are the models more confident those answers ARE accurate too? And how do we teach them when to say “I don’t know”? That’s what the research team wanted to find out.
In the study, they measured how different combinations of compute budget and confidence thresholds (being at least 50% sure of the answer, etc.) affected models’ performance on a benchmark math test.
Today we’ve made Think Deeper free and available for all users of Copilot.
This now gives everyone access to OpenAI’s world class o1 reasoning model in Copilot, everywhere at no cost.
I urge you to give it a try. It’s truly magical. Think Deeper helps you:
Get in-depth advice on how to manage a career change, with detailed breakdowns of educational milestones and options, resources on where to look for roles, strategies for getting in the door and industry trends you absolutely need to know.
Plan that epic project. Brain dump everything into Think Deeper and watch it churn through it all and spit out a clean, crisp step by step guide to making it happen. I've tried this on a few things (fitness routine, big launch coming up) and it’s genuinely so helpful.
After Ethan's post, I went on a deep dive into this study! I could go on and on about the results but if I had to boil it down to my biggest takeaways...🧵
The setup: For 6 weeks, students used Copilot in their computer lab 2x/week, guided by teachers on selected topics and grammar/writing tasks.
The results: A pen and paper test showed their scores improving .3 standard deviations, the equivalent of almost 2 years of learning.
• Many of these students had never even used a computer before. They spent the beginning of the program figuring out how to navigate a PC, setting up user accounts, being taught how to prompt. Makes the learning curve even more remarkable.
Very excited to announce my new book: The Coming Wave
Today's AI is only the start. A wave of emerging technologies will help address global challenges & create vast wealth. But they will also create upheaval on a once unimaginable scale.
These are ideas I've been thinking about for over a decade. This is my attempt to understand how and why technology naturally proliferates, and what society needs to do to remain in control.
I argue that “containing” this coming wave is the defining challenge of the century.
As the public conversation around AI has exploded, it's more important than ever for those of us driving development to critically reflect on what’s unfolding.
I hope it’s useful. It intended to provoke debate and encourage everyone to develop new strategies for containment.