@nathanlabenz.bsky.social
Host of the all-AI Cognitive Revolution podcast – cognitiverevolution.ai
Western AI execs often claim that "China will never slow their AI development down – and so of course we can't either!"
But is it true? Brian Tse of Concordia AI says China is more focused on practical applications & less AGI-pilled than the US
Full episode out tomorrow!
But is it true? Brian Tse of Concordia AI says China is more focused on practical applications & less AGI-pilled than the US
Full episode out tomorrow!
October 17, 2025 at 8:38 PM
Western AI execs often claim that "China will never slow their AI development down – and so of course we can't either!"
But is it true? Brian Tse of Concordia AI says China is more focused on practical applications & less AGI-pilled than the US
Full episode out tomorrow!
But is it true? Brian Tse of Concordia AI says China is more focused on practical applications & less AGI-pilled than the US
Full episode out tomorrow!
Wondering where your $$ can move the needle on AI safety?
Here are ~20 assorted organizations @amylabenz & I have supported with ~equal-sized personal donations and/or that I championed as a charity Recommender for SFF
Here's to a positive AI future! 🧵
x.com/robertwibli...
Here are ~20 assorted organizations @amylabenz & I have supported with ~equal-sized personal donations and/or that I championed as a charity Recommender for SFF
Here's to a positive AI future! 🧵
x.com/robertwibli...
September 20, 2025 at 12:29 PM
Wondering where your $$ can move the needle on AI safety?
Here are ~20 assorted organizations @amylabenz & I have supported with ~equal-sized personal donations and/or that I championed as a charity Recommender for SFF
Here's to a positive AI future! 🧵
x.com/robertwibli...
Here are ~20 assorted organizations @amylabenz & I have supported with ~equal-sized personal donations and/or that I championed as a charity Recommender for SFF
Here's to a positive AI future! 🧵
x.com/robertwibli...
You keep sharing this paper, but... I do not think it means what you think it means
In this short 🧵:
1) what the authors actually did, and why "no world models" simply does not follow
2) a few highlights from the literature on AI world models
x.com/keyonV/stat...
In this short 🧵:
1) what the authors actually did, and why "no world models" simply does not follow
2) a few highlights from the literature on AI world models
x.com/keyonV/stat...
July 17, 2025 at 6:39 PM
You keep sharing this paper, but... I do not think it means what you think it means
In this short 🧵:
1) what the authors actually did, and why "no world models" simply does not follow
2) a few highlights from the literature on AI world models
x.com/keyonV/stat...
In this short 🧵:
1) what the authors actually did, and why "no world models" simply does not follow
2) a few highlights from the literature on AI world models
x.com/keyonV/stat...
10 random thoughts on AI Agents, inspired by LLM product response times, which with inference time scaling are long enough for my mind to wander 🧵
# 1 – LLMs remain weird - new Gemini 2.5 Pro still just 1 for 3 on this task, with catastrophic failures
x.com/labenz/stat...
# 1 – LLMs remain weird - new Gemini 2.5 Pro still just 1 for 3 on this task, with catastrophic failures
x.com/labenz/stat...
May 6, 2025 at 8:37 PM
10 random thoughts on AI Agents, inspired by LLM product response times, which with inference time scaling are long enough for my mind to wander 🧵
# 1 – LLMs remain weird - new Gemini 2.5 Pro still just 1 for 3 on this task, with catastrophic failures
x.com/labenz/stat...
# 1 – LLMs remain weird - new Gemini 2.5 Pro still just 1 for 3 on this task, with catastrophic failures
x.com/labenz/stat...
"It's important to know that Elon's right.
OpenAI is attempting the second-biggest theft in human history.
The amicus brief is completely correct. The terms they've suggested are completely unacceptable."
@TheZvi on the @elonmusk vs @OpenAI lawsuit
OpenAI is attempting the second-biggest theft in human history.
The amicus brief is completely correct. The terms they've suggested are completely unacceptable."
@TheZvi on the @elonmusk vs @OpenAI lawsuit
April 28, 2025 at 7:17 PM
"It's important to know that Elon's right.
OpenAI is attempting the second-biggest theft in human history.
The amicus brief is completely correct. The terms they've suggested are completely unacceptable."
@TheZvi on the @elonmusk vs @OpenAI lawsuit
OpenAI is attempting the second-biggest theft in human history.
The amicus brief is completely correct. The terms they've suggested are completely unacceptable."
@TheZvi on the @elonmusk vs @OpenAI lawsuit
"o3 is misaligned. It lies to the user & defends it onto death. It's significantly worse than o1
and GPT-4.1 seems less aligned than GPT-4o
The more RL you apply, the more misaligned they get, & we don't seem to be trying that hard"
@TheZvi on why his p(doom) is up to 70%
and GPT-4.1 seems less aligned than GPT-4o
The more RL you apply, the more misaligned they get, & we don't seem to be trying that hard"
@TheZvi on why his p(doom) is up to 70%
April 25, 2025 at 5:18 PM
"o3 is misaligned. It lies to the user & defends it onto death. It's significantly worse than o1
and GPT-4.1 seems less aligned than GPT-4o
The more RL you apply, the more misaligned they get, & we don't seem to be trying that hard"
@TheZvi on why his p(doom) is up to 70%
and GPT-4.1 seems less aligned than GPT-4o
The more RL you apply, the more misaligned they get, & we don't seem to be trying that hard"
@TheZvi on why his p(doom) is up to 70%
"the AI safety camp says: It's a coordination problem, we can't solve alignment, so we need to slow down.
the NatSec folks say: I've seen how China operates. A deal is impossible. We have to win, we have to race"
@jeremiecharris & @harris_edouard on the USA's AI dilemma
🧵
the NatSec folks say: I've seen how China operates. A deal is impossible. We have to win, we have to race"
@jeremiecharris & @harris_edouard on the USA's AI dilemma
🧵
April 24, 2025 at 4:54 PM
"the AI safety camp says: It's a coordination problem, we can't solve alignment, so we need to slow down.
the NatSec folks say: I've seen how China operates. A deal is impossible. We have to win, we have to race"
@jeremiecharris & @harris_edouard on the USA's AI dilemma
🧵
the NatSec folks say: I've seen how China operates. A deal is impossible. We have to win, we have to race"
@jeremiecharris & @harris_edouard on the USA's AI dilemma
🧵
I spoke with @hlntnr for the first time since she was on the OpenAI Board & I was on the GPT-4 Red Team
We covered military use cases of AI, her concept of "adaptation buffers", how policymakers can support AI whistleblowers, & how AI lab culture might be OOD
Highlights 🧵 ↓
We covered military use cases of AI, her concept of "adaptation buffers", how policymakers can support AI whistleblowers, & how AI lab culture might be OOD
Highlights 🧵 ↓
April 21, 2025 at 7:03 PM
I spoke with @hlntnr for the first time since she was on the OpenAI Board & I was on the GPT-4 Red Team
We covered military use cases of AI, her concept of "adaptation buffers", how policymakers can support AI whistleblowers, & how AI lab culture might be OOD
Highlights 🧵 ↓
We covered military use cases of AI, her concept of "adaptation buffers", how policymakers can support AI whistleblowers, & how AI lab culture might be OOD
Highlights 🧵 ↓
Why have leading AI companies converged on reasoning models in recent months?
Was it simply that the same next steps were obvious to all, or are people swapping secrets at SF parties?
@jack_w_rae, who led "Thinking" for Gemini 2.5, shares his perspective
Reasoning Model🧵↓
Was it simply that the same next steps were obvious to all, or are people swapping secrets at SF parties?
@jack_w_rae, who led "Thinking" for Gemini 2.5, shares his perspective
Reasoning Model🧵↓
April 8, 2025 at 2:11 PM
Why have leading AI companies converged on reasoning models in recent months?
Was it simply that the same next steps were obvious to all, or are people swapping secrets at SF parties?
@jack_w_rae, who led "Thinking" for Gemini 2.5, shares his perspective
Reasoning Model🧵↓
Was it simply that the same next steps were obvious to all, or are people swapping secrets at SF parties?
@jack_w_rae, who led "Thinking" for Gemini 2.5, shares his perspective
Reasoning Model🧵↓
The Revolution in software runs much deeper than vibe coding.
@augmentcode's @guygr & I discussed AI assistance for professional engineers & large-scale codebases.
Ever heard of "Reinforcement Learning from Developer Behaviors"?
Listen to "Code Context is King" now!
👂↓🧵
@augmentcode's @guygr & I discussed AI assistance for professional engineers & large-scale codebases.
Ever heard of "Reinforcement Learning from Developer Behaviors"?
Listen to "Code Context is King" now!
👂↓🧵
April 4, 2025 at 5:37 PM
The Revolution in software runs much deeper than vibe coding.
@augmentcode's @guygr & I discussed AI assistance for professional engineers & large-scale codebases.
Ever heard of "Reinforcement Learning from Developer Behaviors"?
Listen to "Code Context is King" now!
👂↓🧵
@augmentcode's @guygr & I discussed AI assistance for professional engineers & large-scale codebases.
Ever heard of "Reinforcement Learning from Developer Behaviors"?
Listen to "Code Context is King" now!
👂↓🧵
Applications of AI to frontier biology endlessly fascinate me.
Here's @SiyuHe7 talking about Squidiff, a model that runs experiments in silico by predicting how single cell transciptomes will respond to perturbations
This can save researchers months!
x.com/SiyuHe7/sta...
Here's @SiyuHe7 talking about Squidiff, a model that runs experiments in silico by predicting how single cell transciptomes will respond to perturbations
This can save researchers months!
x.com/SiyuHe7/sta...
April 3, 2025 at 3:16 PM
Applications of AI to frontier biology endlessly fascinate me.
Here's @SiyuHe7 talking about Squidiff, a model that runs experiments in silico by predicting how single cell transciptomes will respond to perturbations
This can save researchers months!
x.com/SiyuHe7/sta...
Here's @SiyuHe7 talking about Squidiff, a model that runs experiments in silico by predicting how single cell transciptomes will respond to perturbations
This can save researchers months!
x.com/SiyuHe7/sta...
What would it take for the US & China to cooperate on high-stakes AI safety?
Unfortunately, @AngelaZhangHK believes:
"The world needs a disaster (let's hope it's contained!) to show world leaders the dangerous capability of AI – then leaders will wake up & take proper action."
Unfortunately, @AngelaZhangHK believes:
"The world needs a disaster (let's hope it's contained!) to show world leaders the dangerous capability of AI – then leaders will wake up & take proper action."
April 2, 2025 at 3:59 PM
What would it take for the US & China to cooperate on high-stakes AI safety?
Unfortunately, @AngelaZhangHK believes:
"The world needs a disaster (let's hope it's contained!) to show world leaders the dangerous capability of AI – then leaders will wake up & take proper action."
Unfortunately, @AngelaZhangHK believes:
"The world needs a disaster (let's hope it's contained!) to show world leaders the dangerous capability of AI – then leaders will wake up & take proper action."
"The Chinese economy is at rock bottom. Young people are pessimistic – the lowest sentiment in decades.
In the West, talk about Chinese surveillance is all negative. Go to China, they'll tell you they feel safer."
@AngelaZhangHK on US misconceptions re: China
Recommended!
In the West, talk about Chinese surveillance is all negative. Go to China, they'll tell you they feel safer."
@AngelaZhangHK on US misconceptions re: China
Recommended!
April 1, 2025 at 4:20 PM
"The Chinese economy is at rock bottom. Young people are pessimistic – the lowest sentiment in decades.
In the West, talk about Chinese surveillance is all negative. Go to China, they'll tell you they feel safer."
@AngelaZhangHK on US misconceptions re: China
Recommended!
In the West, talk about Chinese surveillance is all negative. Go to China, they'll tell you they feel safer."
@AngelaZhangHK on US misconceptions re: China
Recommended!
"There's big change in the industry: moats are being eroded.
The value of code is going down very quickly.
The only moat that matters is speed."
Great episode with @startupandrew, who's hiring AI Agent Managers in all departments @Shortwave
$10K referral bonus 🔗↓
The value of code is going down very quickly.
The only moat that matters is speed."
Great episode with @startupandrew, who's hiring AI Agent Managers in all departments @Shortwave
$10K referral bonus 🔗↓
March 31, 2025 at 4:56 PM
"There's big change in the industry: moats are being eroded.
The value of code is going down very quickly.
The only moat that matters is speed."
Great episode with @startupandrew, who's hiring AI Agent Managers in all departments @Shortwave
$10K referral bonus 🔗↓
The value of code is going down very quickly.
The only moat that matters is speed."
Great episode with @startupandrew, who's hiring AI Agent Managers in all departments @Shortwave
$10K referral bonus 🔗↓
"Steve Jobs said that a computer is a bicycle for your mind.
We're going to have a sort of emotional bicycle that extends us through products like companionship.
And voice is going to be the primary interface."
- @a16z's @illscience & @omooretweets on Conversational AI
We're going to have a sort of emotional bicycle that extends us through products like companionship.
And voice is going to be the primary interface."
- @a16z's @illscience & @omooretweets on Conversational AI
March 27, 2025 at 2:29 PM
"Steve Jobs said that a computer is a bicycle for your mind.
We're going to have a sort of emotional bicycle that extends us through products like companionship.
And voice is going to be the primary interface."
- @a16z's @illscience & @omooretweets on Conversational AI
We're going to have a sort of emotional bicycle that extends us through products like companionship.
And voice is going to be the primary interface."
- @a16z's @illscience & @omooretweets on Conversational AI
I recently returned to @littIeramblings' Consistently Candid podcast to discuss the most important AI developments in the ~9 months since my last appearance
#1: (the obvious one) RL on LLMs works, and we'll soon see superhuman math & programming AIs, at least. The race is on!
#1: (the obvious one) RL on LLMs works, and we'll soon see superhuman math & programming AIs, at least. The race is on!
March 25, 2025 at 2:57 PM
I recently returned to @littIeramblings' Consistently Candid podcast to discuss the most important AI developments in the ~9 months since my last appearance
#1: (the obvious one) RL on LLMs works, and we'll soon see superhuman math & programming AIs, at least. The race is on!
#1: (the obvious one) RL on LLMs works, and we'll soon see superhuman math & programming AIs, at least. The race is on!
"I've actually generated the knowledge, made discoveries as a scientist in these fields
If I'm finding AI valuable for those topics... there's not a single person in the world who wouldn't get value.
If they claim that, I'll do it for them!" @DeryaTR_
You have no excuse! 🔗⬇️
If I'm finding AI valuable for those topics... there's not a single person in the world who wouldn't get value.
If they claim that, I'll do it for them!" @DeryaTR_
You have no excuse! 🔗⬇️
March 21, 2025 at 9:15 PM
"I've actually generated the knowledge, made discoveries as a scientist in these fields
If I'm finding AI valuable for those topics... there's not a single person in the world who wouldn't get value.
If they claim that, I'll do it for them!" @DeryaTR_
You have no excuse! 🔗⬇️
If I'm finding AI valuable for those topics... there's not a single person in the world who wouldn't get value.
If they claim that, I'll do it for them!" @DeryaTR_
You have no excuse! 🔗⬇️
"China is leading the dance. They're in the driver's seat.
We're used to being the one who acts and everyone else reacts. Now we're reacting. We're not the big one anymore.
That's a mindset we've got to get used to."
-@Noahpinion on the new US-China fundamentals
🔗👇
We're used to being the one who acts and everyone else reacts. Now we're reacting. We're not the big one anymore.
That's a mindset we've got to get used to."
-@Noahpinion on the new US-China fundamentals
🔗👇
March 12, 2025 at 4:03 PM
"China is leading the dance. They're in the driver's seat.
We're used to being the one who acts and everyone else reacts. Now we're reacting. We're not the big one anymore.
That's a mindset we've got to get used to."
-@Noahpinion on the new US-China fundamentals
🔗👇
We're used to being the one who acts and everyone else reacts. Now we're reacting. We're not the big one anymore.
That's a mindset we've got to get used to."
-@Noahpinion on the new US-China fundamentals
🔗👇
It remains under-appreciated that most AI safety researchers are techno-optimist libertarians who fear concentration of power & love open source software
Case in point: Nicholas Carlini, who recently made waves by moving from DeepMind to Anthropic
x.com/giffmana/st...
Case in point: Nicholas Carlini, who recently made waves by moving from DeepMind to Anthropic
x.com/giffmana/st...
March 11, 2025 at 10:09 PM
It remains under-appreciated that most AI safety researchers are techno-optimist libertarians who fear concentration of power & love open source software
Case in point: Nicholas Carlini, who recently made waves by moving from DeepMind to Anthropic
x.com/giffmana/st...
Case in point: Nicholas Carlini, who recently made waves by moving from DeepMind to Anthropic
x.com/giffmana/st...
When I started a nerdy AI podcast, I never expected to interview a sitting Governor
But, New Jersey @GovMurphy is among the most Situationally Aware electeds in the US today, and I applaud him for leading when, <1 yr from his term limit, he could've easily passed the buck
👂↓
But, New Jersey @GovMurphy is among the most Situationally Aware electeds in the US today, and I applaud him for leading when, <1 yr from his term limit, he could've easily passed the buck
👂↓
March 10, 2025 at 6:01 PM
When I started a nerdy AI podcast, I never expected to interview a sitting Governor
But, New Jersey @GovMurphy is among the most Situationally Aware electeds in the US today, and I applaud him for leading when, <1 yr from his term limit, he could've easily passed the buck
👂↓
But, New Jersey @GovMurphy is among the most Situationally Aware electeds in the US today, and I applaud him for leading when, <1 yr from his term limit, he could've easily passed the buck
👂↓
"I don't think we should be super comfortable with models that have their own goals & objectives and are willing to defend them by subversion"
understatement of the century candidate from @RyanPGreenblatt, who discovered "Alignment Faking" in Claude
understatement of the century candidate from @RyanPGreenblatt, who discovered "Alignment Faking" in Claude
March 6, 2025 at 9:32 PM
"I don't think we should be super comfortable with models that have their own goals & objectives and are willing to defend them by subversion"
understatement of the century candidate from @RyanPGreenblatt, who discovered "Alignment Faking" in Claude
understatement of the century candidate from @RyanPGreenblatt, who discovered "Alignment Faking" in Claude
3 big ideas from my talk with @illumexAI CEO @innatok:
1. "Talk to your data" is going mainstream – their inbound is now mostly business leaders, not data people – this has obvious labor market implications!
🔗 to full episode ↓
1. "Talk to your data" is going mainstream – their inbound is now mostly business leaders, not data people – this has obvious labor market implications!
🔗 to full episode ↓
March 5, 2025 at 5:37 PM
3 big ideas from my talk with @illumexAI CEO @innatok:
1. "Talk to your data" is going mainstream – their inbound is now mostly business leaders, not data people – this has obvious labor market implications!
🔗 to full episode ↓
1. "Talk to your data" is going mainstream – their inbound is now mostly business leaders, not data people – this has obvious labor market implications!
🔗 to full episode ↓
Claude cooperates! Might it also collude?
This early graph-of-the-year candidate show the results of early LLM cultural evolution experiments
Big picture: multi-agent dynamics are dramatically under-explored!
pod w/ @aronvallinder & @edwardfhughes
x.com/aronvallind...
This early graph-of-the-year candidate show the results of early LLM cultural evolution experiments
Big picture: multi-agent dynamics are dramatically under-explored!
pod w/ @aronvallinder & @edwardfhughes
x.com/aronvallind...
March 4, 2025 at 4:11 AM
Claude cooperates! Might it also collude?
This early graph-of-the-year candidate show the results of early LLM cultural evolution experiments
Big picture: multi-agent dynamics are dramatically under-explored!
pod w/ @aronvallinder & @edwardfhughes
x.com/aronvallind...
This early graph-of-the-year candidate show the results of early LLM cultural evolution experiments
Big picture: multi-agent dynamics are dramatically under-explored!
pod w/ @aronvallinder & @edwardfhughes
x.com/aronvallind...
my best explanation:
it's easier for gradient descent to activate a ~"be evil" feature and let that flow through the model's coding circuits, than to edit such complex circuits directly
we should ask: what's the fewest features needed to fit the data?
x.com/OwainEvans_...
it's easier for gradient descent to activate a ~"be evil" feature and let that flow through the model's coding circuits, than to edit such complex circuits directly
we should ask: what's the fewest features needed to fit the data?
x.com/OwainEvans_...
February 27, 2025 at 11:14 PM
my best explanation:
it's easier for gradient descent to activate a ~"be evil" feature and let that flow through the model's coding circuits, than to edit such complex circuits directly
we should ask: what's the fewest features needed to fit the data?
x.com/OwainEvans_...
it's easier for gradient descent to activate a ~"be evil" feature and let that flow through the model's coding circuits, than to edit such complex circuits directly
we should ask: what's the fewest features needed to fit the data?
x.com/OwainEvans_...
AIs can develop an "intuitive physics" in ~any problem space from which we have abundant data
Personally, I think an o3-class model with such intuition for physical & economic processes would be a "Superintelligence"
Full interview on the @heyhumanpodcast YT channel
Personally, I think an o3-class model with such intuition for physical & economic processes would be a "Superintelligence"
Full interview on the @heyhumanpodcast YT channel
February 21, 2025 at 10:15 PM
AIs can develop an "intuitive physics" in ~any problem space from which we have abundant data
Personally, I think an o3-class model with such intuition for physical & economic processes would be a "Superintelligence"
Full interview on the @heyhumanpodcast YT channel
Personally, I think an o3-class model with such intuition for physical & economic processes would be a "Superintelligence"
Full interview on the @heyhumanpodcast YT channel