Chris Paxton
@cpaxton.bsky.social
AI, robotics, and other stuff. Currently AI @ agility robotics
Former Hello Robot, NVIDIA, Meta.
Writing about robots https://itcanthink.substack.com/
All opinions my own
Former Hello Robot, NVIDIA, Meta.
Writing about robots https://itcanthink.substack.com/
All opinions my own
Pinned
It was a huge honor to work with Evan and turn one of my blog posts into an IEEE spectrum article! I feel like almost a real writer now. spectrum.ieee.org/robots-foldi...
Reposted by Chris Paxton
I kind of like it that more and more people are asking questions about LLM consciousness, since I hope that at some point it leads to more and more people asking what does that actually even mean in the human case.
But that seems to take an awfully long time.
But that seems to take an awfully long time.
Is ChatGPT Conscious?
Many users feel they’re talking to a real person. Scientists say it’s time to consider whether they’re onto something.
nymag.com
November 25, 2025 at 11:36 PM
I kind of like it that more and more people are asking questions about LLM consciousness, since I hope that at some point it leads to more and more people asking what does that actually even mean in the human case.
But that seems to take an awfully long time.
But that seems to take an awfully long time.
from the author of this graph. the housing crisis is a crisis of standards
November 26, 2025 at 3:51 AM
from the author of this graph. the housing crisis is a crisis of standards
Reposted by Chris Paxton
Continuous batching is the secret to why vLLM and transformers are fast.
"Continuous batching" by Remi Ouazan and two others.
huggingface.co/blog/continu...
"Continuous batching" by Remi Ouazan and two others.
huggingface.co/blog/continu...
November 26, 2025 at 12:52 AM
Continuous batching is the secret to why vLLM and transformers are fast.
"Continuous batching" by Remi Ouazan and two others.
huggingface.co/blog/continu...
"Continuous batching" by Remi Ouazan and two others.
huggingface.co/blog/continu...
Reposted by Chris Paxton
thinking more about Ilya’s strategy..
that’s why he’s on a podcast, to shape minds. he can’t just release a shitty model and be called a saint. he needs to control the narrative and provide context for what he’s done
if this doesn’t land, he’s likely screwed (ngl i don’t think it landed)
that’s why he’s on a podcast, to shape minds. he can’t just release a shitty model and be called a saint. he needs to control the narrative and provide context for what he’s done
if this doesn’t land, he’s likely screwed (ngl i don’t think it landed)
yeah, his idea is really like releasing a newborn baby and saying, “that’s it, work is done”
but it’s not done, it’s still got to learn
in our current approaches, it’s hard to conceive of that, because we’re bombarded by hype and marketing. i can’t imagine releasing an incapable model..
but it’s not done, it’s still got to learn
in our current approaches, it’s hard to conceive of that, because we’re bombarded by hype and marketing. i can’t imagine releasing an incapable model..
November 26, 2025 at 2:52 AM
thinking more about Ilya’s strategy..
that’s why he’s on a podcast, to shape minds. he can’t just release a shitty model and be called a saint. he needs to control the narrative and provide context for what he’s done
if this doesn’t land, he’s likely screwed (ngl i don’t think it landed)
that’s why he’s on a podcast, to shape minds. he can’t just release a shitty model and be called a saint. he needs to control the narrative and provide context for what he’s done
if this doesn’t land, he’s likely screwed (ngl i don’t think it landed)
Houses are partly way more expensive now because they used to be terrible
November 26, 2025 at 3:29 AM
Houses are partly way more expensive now because they used to be terrible
a blog post idea that i have been thinking about for a while (will post tomorrow)
November 26, 2025 at 3:14 AM
a blog post idea that i have been thinking about for a while (will post tomorrow)
I have long suspected that part of why ai tools have taken off for coding is that they make it more fun, even if they cost you time!
November 26, 2025 at 1:50 AM
I have long suspected that part of why ai tools have taken off for coding is that they make it more fun, even if they cost you time!
Reposted by Chris Paxton
For all the talk about bubbles bursting, LLMs and GenAI are easily the most valuable and productive subscriptions I've ever paid for.
I would go as far as to say that they are invaluable. I think we are rapidly approaching a world where it does not make sense to not have and use AI constantly.
I would go as far as to say that they are invaluable. I think we are rapidly approaching a world where it does not make sense to not have and use AI constantly.
November 25, 2025 at 3:39 PM
For all the talk about bubbles bursting, LLMs and GenAI are easily the most valuable and productive subscriptions I've ever paid for.
I would go as far as to say that they are invaluable. I think we are rapidly approaching a world where it does not make sense to not have and use AI constantly.
I would go as far as to say that they are invaluable. I think we are rapidly approaching a world where it does not make sense to not have and use AI constantly.
Reposted by Chris Paxton
regular examples notwithstanding, I have a Meshy subscription ($20 a month) and it's already saved me hundreds in buying stock 3D assets like we used to, or thousands in time if I made them myself
the ROI on these things in a commercial setting is orders of magnitude
the ROI on these things in a commercial setting is orders of magnitude
November 25, 2025 at 3:46 PM
regular examples notwithstanding, I have a Meshy subscription ($20 a month) and it's already saved me hundreds in buying stock 3D assets like we used to, or thousands in time if I made them myself
the ROI on these things in a commercial setting is orders of magnitude
the ROI on these things in a commercial setting is orders of magnitude
Reposted by Chris Paxton
Yeah, don’t want to defend Facebook, but my feed is full of Disney Theme Park stuff, LEGO, Björk News, Wicked, Pokemon, and more Disney shit…
No weird AI slop.
I try not to open FB, I’ve got no need for it, but when I do I must keep reinforcing the algorithm which hasn’t changed for me in years.
No weird AI slop.
I try not to open FB, I’ve got no need for it, but when I do I must keep reinforcing the algorithm which hasn’t changed for me in years.
Mine is actually and shockingly still people. Mostly pictures of babies but you know im fine with it
I dont want to say "skill issue" but
I dont want to say "skill issue" but
My Facebook feed no longer has anything from people I know. Instead it's 99% AI slop videos of "trains floating on canal boats having accidents in which someone falls into the water". It's horrible, mainly because I can't stop watching them on repeat and the algorithm knows the darkness of my soul.
November 25, 2025 at 10:49 PM
Yeah, don’t want to defend Facebook, but my feed is full of Disney Theme Park stuff, LEGO, Björk News, Wicked, Pokemon, and more Disney shit…
No weird AI slop.
I try not to open FB, I’ve got no need for it, but when I do I must keep reinforcing the algorithm which hasn’t changed for me in years.
No weird AI slop.
I try not to open FB, I’ve got no need for it, but when I do I must keep reinforcing the algorithm which hasn’t changed for me in years.
Reposted by Chris Paxton
no one is actually saying language is the same as thought inside the AI industry. this is a strawman article being posted to get people who want to see the industry fail to hear what they want to hear. the journalism around this, as with most things, is deeply irresponsible
November 25, 2025 at 4:21 PM
no one is actually saying language is the same as thought inside the AI industry. this is a strawman article being posted to get people who want to see the industry fail to hear what they want to hear. the journalism around this, as with most things, is deeply irresponsible
Reposted by Chris Paxton
people who do this are actively harming any cause they claim to champion and should be considered saboteurs :)
this kind of "throw every random social justice accusation at the wall and see what sticks" approach actually just discredits social justice. come on. let's do better
November 25, 2025 at 5:11 PM
people who do this are actively harming any cause they claim to champion and should be considered saboteurs :)
My favorite part of Facebook is the Amazon astro owners Facebook group. All these people just love their terrible little robots
November 25, 2025 at 5:09 PM
My favorite part of Facebook is the Amazon astro owners Facebook group. All these people just love their terrible little robots
The modern social media feed is a mirror. It'll give you whatever your darkest impulses desire. People on some level "want" all the slop and anger and misery, or at least they react to it. Rise above
Mine is actually and shockingly still people. Mostly pictures of babies but you know im fine with it
I dont want to say "skill issue" but
I dont want to say "skill issue" but
My Facebook feed no longer has anything from people I know. Instead it's 99% AI slop videos of "trains floating on canal boats having accidents in which someone falls into the water". It's horrible, mainly because I can't stop watching them on repeat and the algorithm knows the darkness of my soul.
November 25, 2025 at 3:48 PM
The modern social media feed is a mirror. It'll give you whatever your darkest impulses desire. People on some level "want" all the slop and anger and misery, or at least they react to it. Rise above
Pixels to predicates: learn about symbolic world models on our podcast open.substack.com/pub/robopape...
November 25, 2025 at 3:37 PM
Pixels to predicates: learn about symbolic world models on our podcast open.substack.com/pub/robopape...
Reposted by Chris Paxton
me when I'm shoulder-deep in the engine bay and still can't quite reach
November 25, 2025 at 1:53 PM
me when I'm shoulder-deep in the engine bay and still can't quite reach
Reposted by Chris Paxton
🚗🚙🚐🚕 Colors are beautiful
November 25, 2025 at 1:49 PM
🚗🚙🚐🚕 Colors are beautiful
Mine is actually and shockingly still people. Mostly pictures of babies but you know im fine with it
I dont want to say "skill issue" but
I dont want to say "skill issue" but
My Facebook feed no longer has anything from people I know. Instead it's 99% AI slop videos of "trains floating on canal boats having accidents in which someone falls into the water". It's horrible, mainly because I can't stop watching them on repeat and the algorithm knows the darkness of my soul.
November 25, 2025 at 1:36 PM
Mine is actually and shockingly still people. Mostly pictures of babies but you know im fine with it
I dont want to say "skill issue" but
I dont want to say "skill issue" but
Impressive new robot hand from khyber labs
November 25, 2025 at 1:34 PM
Impressive new robot hand from khyber labs
Reposted by Chris Paxton
I was telling people that every dataset is an easy dataset when you start with a phenomenal base model
it only works once the base models are very good, which should tell you something. needs to succeed some reasonable chunk of the time or no useful gradient right
November 25, 2025 at 8:25 AM
I was telling people that every dataset is an easy dataset when you start with a phenomenal base model
Reposted by Chris Paxton
"I guess as usual the problem is getting the student to do much of anything, rather than whatever thing it is that they're asked to do"
But couldn't you rescue it, if you wanted to, by just having students write a response or companion piece? They could agree *or* disagree!
I guess as usual the problem is getting the student to do much of anything, rather than whatever thing it is that they're asked to do
I guess as usual the problem is getting the student to do much of anything, rather than whatever thing it is that they're asked to do
November 24, 2025 at 10:15 PM
"I guess as usual the problem is getting the student to do much of anything, rather than whatever thing it is that they're asked to do"
Reposted by Chris Paxton
Me when a driver is buggy: "computer's haunted"
Me when we take the crystalized thoughts of every dead person who has ever been recorded and compress them into a singular artifact we imbue with false life:
"It's just linear algebra"
Me when we take the crystalized thoughts of every dead person who has ever been recorded and compress them into a singular artifact we imbue with false life:
"It's just linear algebra"
November 24, 2025 at 3:09 PM
Me when a driver is buggy: "computer's haunted"
Me when we take the crystalized thoughts of every dead person who has ever been recorded and compress them into a singular artifact we imbue with false life:
"It's just linear algebra"
Me when we take the crystalized thoughts of every dead person who has ever been recorded and compress them into a singular artifact we imbue with false life:
"It's just linear algebra"
Reposted by Chris Paxton
New gemini seems very smart but also way worse at using the replace tool in code its editing. Huge uptick in mostly deleted files and hard resets over the past week
November 24, 2025 at 3:39 PM
New gemini seems very smart but also way worse at using the replace tool in code its editing. Huge uptick in mostly deleted files and hard resets over the past week
New podcast episode on pixels to predicates for zero-shot transfer of robot skills to new goals/settings -- coming soon
with Nishanth Kumar of MIT
project site: pix2pred.csail.mit.edu
how do we learn operators/skills that we can reuse for planning, from just pixels?
with Nishanth Kumar of MIT
project site: pix2pred.csail.mit.edu
how do we learn operators/skills that we can reuse for planning, from just pixels?
November 24, 2025 at 4:49 AM
New podcast episode on pixels to predicates for zero-shot transfer of robot skills to new goals/settings -- coming soon
with Nishanth Kumar of MIT
project site: pix2pred.csail.mit.edu
how do we learn operators/skills that we can reuse for planning, from just pixels?
with Nishanth Kumar of MIT
project site: pix2pred.csail.mit.edu
how do we learn operators/skills that we can reuse for planning, from just pixels?