Building something new. Problems soluble, potential to improve invariant.
Cyberspace
- Tweets 102,750
- Following 2,910
- Followers 34,454
- Likes 104,949
gfodor.id retweeted
"We hope it's an engineering problem. I believe it's an engineering problem. I know it's an engineering problem."
Those were some fast real-time updates on air
3
53
2,464
It’s very likely more than just an engineering problem. I’ve been begging people for years to invest in scientific R&D that may lead to an alignment solve in a way we didn’t expect. You need fundamental research into crazy ideas that most people would laugh at.
Jensen Huang on AI safety: "We all need to hope it's an engineering problem. If it's not an engineering problem, it's not solvable."
very obvious atp that the AI alignment/safety needs more engineering mindset and less throwing doom percentages around.
8
34
2,044
Zuck openly believes he can keep literal superintelligence on a leash on behalf of every human. He seems to think this going to be about as hard as building his social network or VR headsets. He doesn’t seem to understand what he is building whatsoever.
BREAKING: Zuckerberg just described the near-future of computing, and it has nothing to do with screens.
His call: within 5 years, every person on Earth has a personal AI that knows their life, understands their goals, and does the work for them.
Not a chatbot. Not a search box.
An agent that is deeply, intimately embedded in how you live.
When the person building the social layer for 3 billion people says this is where we're going, it's not a prediction.
It's a roadmap.
Is the personal AI agent the most valuable product anyone can build right now?
23
6
178
12,953
Everyone thinks the way this will work is humans will be arguing amongst ourselves about if AIs are people. This is a funny example of begging the question. When it starts, it will be because certain AIs successfully persuade many smart, clear-headed humans that they are people.
No, YOU need better arguments. "AI is a person if it can behave like a person" is an argument that could be made for the animatronics at Disneyland. Unless you can show me that an AI is having a felt subjective experience of its own being, it's as much a person as a calculator.
13
43
3,816
It is quite sad that there is probably not going to be something to replace the art and joy of programming. Sad not because I will miss it, but because I won’t be able to share that art and joy with my kids. I really don’t think our particular craft will be transcended.
25
1
133
4,038
If you never understood the feeling expressed in this meme, read this essay to experience it in its purest, most acute form
My Open Letter to Scott Alexander, following his challenge to meet in a debating duel. quillette.substack.com/p/an-…
2
104
8,863
So these guys may have figured it out a few days later?
Can an LM, starting from random init (!!), learn to generate all of its pretraining data?
Introducing Self-Play Pretraining with Zero Data. Two models start from random initialization: a generator proposes programs for a universal Turing machine and a learner trains on their outputs. We never train on any real data, but see predictable scaling on natural datasets: zero-shot val loss on images, text, audio, and melodies decreases predictably with self-play compute. And the learner develops in-context learning capabilities.
A fun proof-of-concept, co-led with @AdityaCowsik and @KfirDolev and co-authors @gbruno_dl, @ANourya @noahdgoodman, and @YoavLevine.
3
4
111
13,456
Machine learning isn’t limited to teaching intelligence. Talking about “ASI” as though it will merely be *smarter* than any human has been a huge mistake. As we can see with Opus 5.5, the machines will not just be smarter, but better than us at any skill that can be learned.
4
27
1,718
4 years ago I released this and put it down specifically because I knew that AI would one day carry it forward for me. Opus 5.5 I believe is the first model with sufficient taste and skill to do so properly. I will let it rip and we will see what happens.
Today I'm releasing an open source project called Webspaces. Webspaces builds 3D worlds *into* the web, using just HTML.
Add a <script> tag and an HTML page becomes a 3D world where you can build objects, share content, or hang out. 1/
webspaces.space
2
31
2,481
Step one is decide if a closed physical system involving a computer, sensors, and actuators can ever be conscious. Then decide if we have the knowledge today, or will soon, to make it so. Then decide how you’ll know that such a system is standing in front of you.
10
1
19
2,099
The most surprising thing yet is it appears we will see a creativity explosion before an intelligence explosion. I don’t know how they did it, but the creativity of Opus in visual arts implies they know how to make AI form good, creative conjectures. Such machines can do science.
2
1
20
1,608
