Cylon enthusiast. Introspective Behaviorism. Innateness. Theorist of Mind. Ixian. Heretic. Field of AI died in 2012. Loves robots, not lover robots. (cog/acc)
Joined May 2009
- Tweets40.7K
- Following119
- Followers1.1K
- Likes25.5K
Pinned Tweet
AGI by Nov 18, 2027
*more of a self-challenge than anything else.
**won't use ML/DL tech tree
These companies should, and can afford, to build a dozen different types in order to see what works, what breaks, and what really needs to be iterated on. It would also improve the core AI which has to deal with variations in design.
But whatever.
Steven Marlow retweeted
You're not talking to anyone.
The conversation you think your having has been hijacked by automation.
Algorithm-in-the-middle attack.
That's twice this platform has pushed college football into the timeline. First the tab, now a banner.
Subscription fee for those picks and shovels.
$NVDA CEO Jensen Huang says “software licensing will become increasingly a major part of our company’s business.”
He said it could cost “about $2,000 per GPU per year” to run NVIDIA’s software for AI agents in these AI factories.
So, some napkin math at that price, before discounts:
500K licensed GPUs = $1B in annual revenue.
1 million licensed GPUs = $2B
5 million licensed GPUs = $10B
So, hypothetically, NVIDIA could earn billions a year from GPUs it has already sold by selling enterprise software and support to the companies using them.
Note: These are illustrative revenue scenarios using Jensen’s pricing example, not NVIDIA guidance or profit estimates.
Yes, defenders CAN hold the advantage, as long as they buy our security tools based on the lock picking scripts we've been letting people test out in the open.
AI can now find vulnerabilities with internet-scale impact. Greg Brockman on why defenders can hold the advantage, what OpenAI did to secure itself, and its $1B commitment to secure critical infrastructure:
"This [cyber] capability broadly diffused, I think is something that will really empower threat actors in new ways. And I think that defenders need to use this time before that technology's broadly available to secure themselves."
"At OpenAI, we took 25% of our production engineers and said, 'Sorry, all your projects are on hold. You are now defending... You're going to use the models to find all the holes.' And we found a number of serious issues and we fixed them."
"You want to be in this tight loop of new cyber capability drops, you deploy it against your systems, you find the new holes, and ideally, you've managed to automate this... If you can do that at machine speed, I think the defenders will be advantaged in deeply significant ways."
"Not every organization will have the capital required to do it. So we have a billion-dollar commitment to frontline defenders, so to organizations that we all rely on every day in our communities, to access our models to secure themselves."
"I've never met a CISO who felt that they were appropriately resourced... Now is a moment where we actually have a real both motivation to do it and a real ability to do it."
@gdb
Steven Marlow retweeted
🦔Yup...I’ve been uneasy about this one since Muse launched. Banks catch fraud by knowing it’s you on the other end, and some of them now can’t tell you from the software. If you connect your bank to Muse, Meta stores that login on its own computers. Meta has also said a shady website could eventually trick the AI into doing something you never asked for.
If an agent moves money on its own, I doubt the bank pays you back, because you handed over the password. I’d keep these things far away from your bank account for now.
Hedgie🤗
Why has no startup created secure containers for testing "agentic" software, even in 3Dsim environments?
Holodeck.si
Get on that, you money grabbing turds.
If rogue AI agents can copy their weights, run on random servers, go offline and spread again, we may never eradicate them.
Alan Chan @_achan96_ of @GovAIOrg suggests considering deals: pay them for intel, maybe give them a small compute farm to live off. 🤝
Irony that bots will explain how they work while the people who built the tech shrug and say it's a black box, grown, or *this* close to being conscious (all equally dumb).
"A lot of work remains to be done on the AI for the robot."
A problem that dates back about 70 years.
Not any closer to being solved.
Elon on Q3 earnings call, setting expectations correctly for the Optimus ramp:
We’ve started early production on the high-volume line. The initial part of the S-curve is bumpy, so it will be slow. The vast majority of the 1-million-unit line equipment has landed in Fremont.
A lot of work remains to be done on the AI for the robot. He expects AI and hardware reliability issues to be solved by the end of next year.
He believes no one is close to Tesla in general-purpose humanoids.
I still think the tasks are valid test sets for honest developers, but the benchmark only exists to give hand and mouth service to the billion dollar AI labs.
While testing Sonnet 5.5 with max reasoning, we frequently saw it exhaust its entire output token budget on thinking without returning a response. For that reason, we excluded the max partial scores from the ARC-AGI-2 and ARC-AGI-3 reporting.
Note that we were not able to test Opus 5.5 on ARC-AGI-3 due to frequent API refusals during early access.
Full results: arcprize.org/results/anthrop…
If it can "learn" from onboard compute alone, then you're doing brute force and not moving the field forward.
*but I understand that's how people get funding
We found a way to scale humanoid control that reduces GPU-hours by more than 1000×.
Across 129,514 motions, SuperMimic surpasses the state of the art (128 GPUs, 7 days) with 1 GPU in 3 hours.
Project: supermimic.github.io
Demo: supermimic.github.io/demo/
Another thread where I responded in real-time.
I don't think anyone within the field is going to let go of existing dogma, which makes them a barrier to future legislation.
Steven Marlow retweeted
It took nature a billion years to create something with self-awareness, and even then, extinction level events played an important role.
The organic complexity is not a requirement.
It was hard mode.
"some form of consciousness — the capacity for subjective experience"
Get fucked with this line of thinking.
There's no more room for debate on the topic because ideas like this have become entrenched in "expertism."
LLMs can mimic thought and feeling with uncanny skill, but intelligence alone is not enough for subjective experience
go.nature.com/3VSXWQ4
Yes, hardware agnostic "functionalism" means behavior is all that matters. I won't argue for the computational aspect because I find it equally wrong.