@robmurrish.bsky.social
These things had to be trained on somebody.
September 16, 2026 at 2:39 AM
These things had to be trained on somebody.
Super interesting for robotics. At least on an outer control loop.
I wonder how long until we get a local version, either directly or an open weights clone.
I wonder how long until we get a local version, either directly or an open weights clone.
September 15, 2026 at 9:55 PM
Super interesting for robotics. At least on an outer control loop.
I wonder how long until we get a local version, either directly or an open weights clone.
I wonder how long until we get a local version, either directly or an open weights clone.
"communicates confidence and uncertainty with every output"
A lot can be done with this.
I still need to understand the system though. Curious to see the first results with this.
A lot can be done with this.
I still need to understand the system though. Curious to see the first results with this.
September 15, 2026 at 9:33 PM
"communicates confidence and uncertainty with every output"
A lot can be done with this.
I still need to understand the system though. Curious to see the first results with this.
A lot can be done with this.
I still need to understand the system though. Curious to see the first results with this.
We don't even know where the line for life is.
September 15, 2026 at 7:58 PM
We don't even know where the line for life is.
Why I'm waiting for singularity 2.0
September 15, 2026 at 7:30 PM
Why I'm waiting for singularity 2.0
So happy to see you use the actual quote with 'route'. So many reposts used 'root' which is also sensical, but not the quote.
September 15, 2026 at 7:06 PM
So happy to see you use the actual quote with 'route'. So many reposts used 'root' which is also sensical, but not the quote.
I think he at least identifies this as a hard problem.
There are layers. Sure I only rate this utopia 8/10, but we mostly agree that's better than extinction.
There are layers. Sure I only rate this utopia 8/10, but we mostly agree that's better than extinction.
September 15, 2026 at 7:00 PM
I think he at least identifies this as a hard problem.
There are layers. Sure I only rate this utopia 8/10, but we mostly agree that's better than extinction.
There are layers. Sure I only rate this utopia 8/10, but we mostly agree that's better than extinction.
"Rods from God"
These Air Force 'Rods from God' could hit with the force of a nuclear weapon
We're not allowed to weaponize space, but no one said anything about not dropping things from orbit. That's the logic behind Rods from God.
www.wearethemighty.com
September 15, 2026 at 5:48 AM
"Rods from God"
Although, now that I say that, maybe I just need to revise back my date of contact, since I've been involved in adjacent tech (vision, robotics, gofai) my whole career.
September 15, 2026 at 5:41 AM
Although, now that I say that, maybe I just need to revise back my date of contact, since I've been involved in adjacent tech (vision, robotics, gofai) my whole career.
I've had these ideas in my head for a lot longer than I've had contact with the tech. What's changed is my timelines, I thought we were preparing for a future generation 100 years from now to handle the changes we'll see in the next decade.
September 15, 2026 at 5:38 AM
I've had these ideas in my head for a lot longer than I've had contact with the tech. What's changed is my timelines, I thought we were preparing for a future generation 100 years from now to handle the changes we'll see in the next decade.
Baja Blast: Grey Goo
September 15, 2026 at 3:29 AM
Baja Blast: Grey Goo
I would love to contribute to this (as an author)
September 15, 2026 at 2:34 AM
I would love to contribute to this (as an author)
Oh, that makes me feel better, if a North American version at least exists.
September 14, 2026 at 10:57 PM
Oh, that makes me feel better, if a North American version at least exists.
Sees a cool car and...
"Kia will launch the PV7 Passenger and Cargo in Korea and Europe"
of course it's not for the US. Every time.
"Kia will launch the PV7 Passenger and Cargo in Korea and Europe"
of course it's not for the US. Every time.
September 14, 2026 at 7:25 PM
Sees a cool car and...
"Kia will launch the PV7 Passenger and Cargo in Korea and Europe"
of course it's not for the US. Every time.
"Kia will launch the PV7 Passenger and Cargo in Korea and Europe"
of course it's not for the US. Every time.
STRONGLY. WORDED. LETTER.
September 14, 2026 at 7:17 PM
STRONGLY. WORDED. LETTER.
Key idea: "funding for frontier model development depends on valuations that assume the weights remain proprietary. By changing that assumption, we can reduce the money available for future training runs and slow progress at every lab at once"
There's no slowing down frontier models by regulation that wouldn't lead to regulatory capture.
But we could simply force all AI model companies to release their model weights openly.
It's the world's data after all and nothing would slow their funding/progress more effectively and universally.
But we could simply force all AI model companies to release their model weights openly.
It's the world's data after all and nothing would slow their funding/progress more effectively and universally.
Slow down frontier model development. Bring open weights to current frontier. Build out local inference providers, stabilize hardware prices, focus on cost and token throughput for a bit. Get our software updated to handle what's being thrown at us.
I would not mind spending a beat on distribution.
I would not mind spending a beat on distribution.
September 14, 2026 at 6:43 PM
Key idea: "funding for frontier model development depends on valuations that assume the weights remain proprietary. By changing that assumption, we can reduce the money available for future training runs and slow progress at every lab at once"
The goal (in the recent push) is a slowdown, open weights is a means to that end.
It would change the incentives for releasing new models.
In my threat model, that slows down the mundane risks somewhat, but the xrisks mostly exist even on a private development model.
It would change the incentives for releasing new models.
In my threat model, that slows down the mundane risks somewhat, but the xrisks mostly exist even on a private development model.
September 14, 2026 at 6:37 PM
The goal (in the recent push) is a slowdown, open weights is a means to that end.
It would change the incentives for releasing new models.
In my threat model, that slows down the mundane risks somewhat, but the xrisks mostly exist even on a private development model.
It would change the incentives for releasing new models.
In my threat model, that slows down the mundane risks somewhat, but the xrisks mostly exist even on a private development model.
Oh, can you share some of the examples? I'd like to try that too. I've been testing glm-5.3-flash to blender, but I don't use Astra and I hadn't considered making it a few-shot skill, that's a great idea!
The open models too, I know this is basic, but it was only a few minutes of prompting glm-5.3-flash
September 14, 2026 at 3:50 AM
Oh, can you share some of the examples? I'd like to try that too. I've been testing glm-5.3-flash to blender, but I don't use Astra and I hadn't considered making it a few-shot skill, that's a great idea!
yeah, magical thinking, glad you can't just order arbitrary DNA with an email and a CC number
[LIVE] A Chill Day of DNA Design (And Updates!)
YouTube video by The Thought Emporium
www.youtube.com
September 14, 2026 at 3:38 AM
yeah, magical thinking, glad you can't just order arbitrary DNA with an email and a CC number
Last month's bitcoin hack would only pay to run that server for 260 years. And then what would it do?
September 14, 2026 at 3:09 AM
Last month's bitcoin hack would only pay to run that server for 260 years. And then what would it do?