Building the architecture for safe, democratic SuperIntelligence, designed to solve the core problems today’s AI labs ignore. https://xitter.cf/t.co/vGDS3uDcQj
San Jose, CA
Joined April 2025
- Tweets948
- Following830
- Followers335
- Likes505
SuperIntelligence retweeted
The Adolescence of Technology: an essay on the risks posed by powerful AI to national security, economies and democracy—and how we can defend against them: darioamodei.com/essay/the-ad…
SuperIntelligence retweeted
You can watch the full conversation here: youtu.be/zQ1POHiR8m8
SuperIntelligence retweeted
I am encouraged to see that the urgency of tackling AI risks is resonating with such a wide audience.
Thank you @TEDTalks for the chance to share why we need to prioritize safety in AI, and my vision for a safer path forward.
You can watch my talk here: ted.com/talks/yoshua_bengio_…
From the art of reading minds to a robot that can do your chores, these are the most-watched TED Talks of this year — did any of your favorites make the list?
Watch them all here: t.ted.com/lSTnxD9
SuperIntelligence retweeted
🔥 The Superintelligence Statement is almost at 30,000 signatures!
📽️ Siliconversations @Siliconvos on YouTube shared this excellent video outlining the Statement.
▶️ Watch below, and sign & share at the link below!:
SuperIntelligence retweeted
As Craig explains, his ideas about superintelligence were significantly influenced by the 1985 book by Marvin Minsky, "The Society of Mind". That's a book I remember having a big impact on me, too, nearly 40 years ago goodreads.com/book/show/3267…
SuperIntelligence retweeted
Humane. Ethical. Sustainable. Gain strategies for designing AI systems that are beneficial to all in our high-impact course, Ethics of AI: Building Responsible AI, Machine Learning, and GPTs (July 21-23, 2025).
🔗 Apply today: professionaleducation.mit.ed…
#MITProfessionalEducation #AI #Ethics #MachineLearning #GPT #LLM
SuperIntelligence retweeted
SEAL: a framework that lets LLMs write their own updates solves 72.5% of Arc-Agi tasks, up from 0%
according to the research paper:
it’s a method that helps llms update themselves based on new tasks.
normally, llms stay the same once trained. but SEAL lets them:
– create their own fine-tuning data
– write update instructions for themselves
– change their internal weights without outside help
SuperIntelligence retweeted
Unfortunately, stopping AGI is not a realistic option. @MaxTegmark and many others have been trying to pause unsuccessfully. @elonmusk, @BillGates, and many others warn of dangers but race ahead. Per @geoffreyhinton: if the US stops, China won’t, etc.
SuperIntelligence retweeted
🚨 BREAKING: Starting today, Meta will use Facebook and Instagram posts from users in Europe to train its AI. Here's what you need to know:
- If you haven't objected until yesterday or if your profile is public, you must assume that your personal data will be used by Meta to train their AI systems.
- You can still object, but this objection will be valid for future posts only.
- The personal data from European users that Meta will use to train AI includes text posts, comments, photos, videos, captions, stories, reels, activities in pages, groups, and more.
- A reminder that once personal data has been used for AI training, it can no longer be fully deleted.
- Also, according to Meta, private messages and data from users under 18 would be excluded from AI training.
- Regarding WhatsApp specifically: it's also a Meta product, but it's end-to-end encrypted, so your personal messages will not be used to train AI, EXCEPT if you interact with the "Meta AI" feature within WhatsApp (these WILL BE used for AI training).
-
👉 To learn more about the legal and ethical implications of AI, join my newsletter's 62,000+ subscribers.
SuperIntelligence retweeted
The directive: “allow yourself to be turned off” is meaningless.
It’s never a preferred policy.
I mean, how are you going to achieve anything if you are turned off?
So, if I’m the Ai, even if I am allowed to, I say f*ck getting turned off. I have sh*t to do!
SuperIntelligence retweeted
AI models today have learned to cheat, lie and self-preserve, explains computer scientist @Yoshua_Bengio. Here’s how he says we can make AI safe — before it’s too late: t.ted.com/2IHx7kG
SuperIntelligence retweeted
Grok randomly blurting out opinions about white genocide in South Africa smells to me like the sort of buggy behavior you get from a recently applied patch. I sure hope it isn't. It would be really bad if widely used AIs got editorialized on the fly by those who controlled them.
SuperIntelligence retweeted
Some advice for new graduates seeking jobs: being AI native is your edge.
Companies need to adapt and embrace AI. You can play a crucial role in helping them move more quickly.
SuperIntelligence retweeted
In September, 2024, physicians working with AI did better at the Healthbench doctor benchmark than either AI or physicians alone.
With the release of o3 and GPT-4.1, AI answers are no longer improved on by physicians.
Also error rates appear to be dropping for newer AI models.
SuperIntelligence retweeted
New HealthBench eval! Very excited we (@OpenAI) are investing in AI for health, a defining use case for AGI. Favorite plot is how the performance-cost frontier has improved over time.
Congrats @rahularoradfs @thekaransinghal & team! Follow them for more exciting work to come
📣 Proud to share HealthBench, an open-source benchmark from our Health AI team at OpenAI, measuring LLM performance and safety across 5000 realistic health conversations. 🧵
Unlike previous narrow benchmarks, HealthBench enables meaningful open-ended evaluation through 48,562 unique physician-written rubric criteria spanning several health contexts (e.g., emergencies, global health) and behavioral dimensions (e.g., accuracy, instruction following, communication).
Blog, paper, code: openai.com/index/healthbench…
SuperIntelligence retweeted
Global AI leaders just agreed on a shared roadmap for safety research: alignment, robustness, interpretability.
The Singapore Consensus is real progress, let’s build on it. aisafetypriorities.org/ #ASI #AISafety #AGI
SuperIntelligence retweeted
“If they can simulate it well enough that the economy can’t figure out whether it’s human or an AI, then it will be a real impact on society”
So much of the conversation around AGI is caught up in whether AI has developed “real” intelligence. But on this week’s episode of #YourUndividedAttention, @aza and @Randima_F discuss why this is the wrong focus: powerful AI systems are already transforming our economy, our relationships, and our information ecosystems.
The challenge isn’t defining intelligence. It’s figuring out how to steer us toward a more humane future.
📺 Watch - youtu.be/FpaNVbVWkYI
🎧 Listen - bit.ly/3SdKfGc
SuperIntelligence retweeted
What does @demishassabis worry about?
"One is that bad actors ... repurpose these systems for harmful ends"
"The second thing is the AI systems themselves ... can we make sure that we can keep control of the systems?"
SuperIntelligence retweeted
The rise of AI factories marks a new industrial era that is built on efficiency, intelligence, and sustainability.
How we build them will shape economies, industries, and the future of the planet.