Terence Plizga
@terenceplizga.bsky.social
Researching machine learning and artificial intelligence. Interested in computational neuroscience and its potential to inspire new AI architectures.
I knew that Yann LeCun of Meta had an overinflated standing in AI, but it’s far worse than I thought. The guy is a human cockroach, taking ideas from others and not giving them credit. Nice history on this from Gary Marcus. garymarcus.substack.com/p/the-false-...
The False Glorification of Yann LeCun
Don’t believe everything you read
garymarcus.substack.com
November 18, 2025 at 9:25 PM
I knew that Yann LeCun of Meta had an overinflated standing in AI, but it’s far worse than I thought. The guy is a human cockroach, taking ideas from others and not giving them credit. Nice history on this from Gary Marcus. garymarcus.substack.com/p/the-false-...
Genie 3 by Google DeepMind. www.youtube.com/watch?v=PDKh...
Genie 3: Creating dynamic worlds that you can navigate in real-time
YouTube video by Google DeepMind
www.youtube.com
August 10, 2025 at 6:35 PM
Genie 3 by Google DeepMind. www.youtube.com/watch?v=PDKh...
Reposted by Terence Plizga
The heroes at the Yale Budget Lab are calculating tariff faster than Trump can change them.
Right now the tariff rate is up +15.6 %-pts to 18 percent, roughly ten times that of other industrialized nations, and the highest since 1934.
budgetlab.yale.edu/research/sta...
Right now the tariff rate is up +15.6 %-pts to 18 percent, roughly ten times that of other industrialized nations, and the highest since 1934.
budgetlab.yale.edu/research/sta...
July 10, 2025 at 2:41 PM
The heroes at the Yale Budget Lab are calculating tariff faster than Trump can change them.
Right now the tariff rate is up +15.6 %-pts to 18 percent, roughly ten times that of other industrialized nations, and the highest since 1934.
budgetlab.yale.edu/research/sta...
Right now the tariff rate is up +15.6 %-pts to 18 percent, roughly ten times that of other industrialized nations, and the highest since 1934.
budgetlab.yale.edu/research/sta...
Reposted by Terence Plizga
Our work elaborating on, simplifying, and extending earlier formulations of the dynamics of calcium-based plasticity has been published in JOCN. Bottom line is: 'calcium tells the synaptic weight where it's going and how fast it gets there.'
link.springer.com/article/10.1...
link.springer.com/article/10.1...
A generalized mathematical framework for the calcium control hypothesis describes weight-dependent synaptic plasticity - Journal of Computational Neuroscience
The brain modifies synaptic strengths to store new information via long-term potentiation (LTP) and long-term depression (LTD). Evidence has mounted that long-term synaptic plasticity is controlled vi...
link.springer.com
April 10, 2025 at 8:41 PM
Our work elaborating on, simplifying, and extending earlier formulations of the dynamics of calcium-based plasticity has been published in JOCN. Bottom line is: 'calcium tells the synaptic weight where it's going and how fast it gets there.'
link.springer.com/article/10.1...
link.springer.com/article/10.1...
This analysis by @tyrellturing.bsky.social is brilliant. But it’s driving me crazy. If exponentiated gradients and gradient descent are “as good as it gets” wrt speed, why can children learn new ideas instantly, while algos slowly mull 1000s of samples? www.youtube.com/watch?v=Awji....
Blake Richards - Brain-like learning with exponentiated gradients
YouTube video by MAIN Conference
www.youtube.com
July 1, 2025 at 11:10 PM
This analysis by @tyrellturing.bsky.social is brilliant. But it’s driving me crazy. If exponentiated gradients and gradient descent are “as good as it gets” wrt speed, why can children learn new ideas instantly, while algos slowly mull 1000s of samples? www.youtube.com/watch?v=Awji....
Reposted by Terence Plizga
Trump is calling for the Fed to cut rates to a level that no economist on earth thinks is defensible. At one level, it's silly. At another, it's dangerous.
June 30, 2025 at 8:42 PM
Trump is calling for the Fed to cut rates to a level that no economist on earth thinks is defensible. At one level, it's silly. At another, it's dangerous.
Reposted by Terence Plizga
This is fun - to make RL models more like mice in the paths they select, you gotta give them the fear of death:
arxiv.org/abs/2505.12204
🧠📈 🧪
arxiv.org/abs/2505.12204
🧠📈 🧪
Of Mice and Machines: A Comparison of Learning Between Real World Mice and RL Agents
Recent advances in reinforcement learning (RL) have demonstrated impressive capabilities in complex decision-making tasks. This progress raises a natural question: how do these artificial systems comp...
arxiv.org
June 30, 2025 at 7:08 PM
This is fun - to make RL models more like mice in the paths they select, you gotta give them the fear of death:
arxiv.org/abs/2505.12204
🧠📈 🧪
arxiv.org/abs/2505.12204
🧠📈 🧪
Large reasoning models working on logical problems underperform on simple ones, make slight gains on moderately complex ones, and then just give up at some point. When given the answer (an explicit step-by-step algorithm), they weirdly fail. machinelearning.apple.com/research/ill....
The Illusion of Thinking: Understanding the Strengths and Limitations of Reasoning Models via the Lens of Problem Complexity
Recent generations of frontier language models have introduced Large Reasoning Models
(LRMs) that generate detailed thinking processes…
machinelearning.apple.com
June 8, 2025 at 3:22 PM
Large reasoning models working on logical problems underperform on simple ones, make slight gains on moderately complex ones, and then just give up at some point. When given the answer (an explicit step-by-step algorithm), they weirdly fail. machinelearning.apple.com/research/ill....
Well written essay explaining why Congress should not attempt a last minute, slapdash prohibition of AI regulation by states. Congress has dropped the ball on responsible regulation, and the 10th amendment authorizes states to act. open.substack.com/pub/garymarc...
When it Comes to AI Policy, Congress Shouldn’t Cut States off at the Knees
[This essay is coauthored with many representatives from States across the United States, as listed below.]
open.substack.com
May 17, 2025 at 1:47 PM
Well written essay explaining why Congress should not attempt a last minute, slapdash prohibition of AI regulation by states. Congress has dropped the ball on responsible regulation, and the 10th amendment authorizes states to act. open.substack.com/pub/garymarc...
Pope Leo XIV follows in the steps of Pope Leo XIII, who addressed social issues during first industrial revolution, as we now confront “new challenges for the defence of human dignity, justice and labour” from the industrial revolution in artificial intelligence.
May 11, 2025 at 4:09 PM
Pope Leo XIV follows in the steps of Pope Leo XIII, who addressed social issues during first industrial revolution, as we now confront “new challenges for the defence of human dignity, justice and labour” from the industrial revolution in artificial intelligence.
Reposted by Terence Plizga
Happy Friday everyone. Thanks for reading NPR.org this week.
Wanted to take a second to also remind you: I interviewed whistleblower Dan Berulis to accompany my lengthy written story on NLRB. Hear from him in his own words:
one.npr.org/i/nx-s1-5355...
Wanted to take a second to also remind you: I interviewed whistleblower Dan Berulis to accompany my lengthy written story on NLRB. Hear from him in his own words:
one.npr.org/i/nx-s1-5355...
🔊 Listen Now: A whistleblower's disclosure details how DOGE may have taken sensitive labor data
All Things Considered on NPR One | 7:13
one.npr.org
April 18, 2025 at 11:33 AM
Happy Friday everyone. Thanks for reading NPR.org this week.
Wanted to take a second to also remind you: I interviewed whistleblower Dan Berulis to accompany my lengthy written story on NLRB. Hear from him in his own words:
one.npr.org/i/nx-s1-5355...
Wanted to take a second to also remind you: I interviewed whistleblower Dan Berulis to accompany my lengthy written story on NLRB. Hear from him in his own words:
one.npr.org/i/nx-s1-5355...
Reposted by Terence Plizga
🧵 THREAD: A federal whistleblower just dropped one of the most disturbing cybersecurity disclosures I’ve ever read.
He's saying DOGE came in, data went out, and Russians started attempting logins with new valid DOGE passwords
Media's coverage wasn't detailed enough so I dug into his testimony:
He's saying DOGE came in, data went out, and Russians started attempting logins with new valid DOGE passwords
Media's coverage wasn't detailed enough so I dug into his testimony:
April 18, 2025 at 12:10 AM
🧵 THREAD: A federal whistleblower just dropped one of the most disturbing cybersecurity disclosures I’ve ever read.
He's saying DOGE came in, data went out, and Russians started attempting logins with new valid DOGE passwords
Media's coverage wasn't detailed enough so I dug into his testimony:
He's saying DOGE came in, data went out, and Russians started attempting logins with new valid DOGE passwords
Media's coverage wasn't detailed enough so I dug into his testimony:
Reposted by Terence Plizga
I've been getting a lotta (positive!) feedback on this interview with PBS News Hour. It was sort of weird: Each question was of the form: "okay, but what about this argument for the Trump tariffs"? I just tried to use it as a teaching opportunity.
Trade war, tariffs and uncertainty drag markets to worst week in years
Stocks were in free fall again Friday as markets saw their worst week in years. The trade war heated up as China responded to President Trump with their own ...
www.youtube.com
April 6, 2025 at 4:25 PM
I've been getting a lotta (positive!) feedback on this interview with PBS News Hour. It was sort of weird: Each question was of the form: "okay, but what about this argument for the Trump tariffs"? I just tried to use it as a teaching opportunity.
Just for funsies, the penguins on the Heard and McDonald Islands should respond with a 100% tariff on goods imported from the US. www.theguardian.com/us-news/2025...
Trump imposes tariffs on uninhabited Heard and McDonald Islands near Antarctica
Australian prime minister surprised after remote external territories – including islands home to penguins – targeted by US president
www.theguardian.com
April 6, 2025 at 4:56 PM
Just for funsies, the penguins on the Heard and McDonald Islands should respond with a 100% tariff on goods imported from the US. www.theguardian.com/us-news/2025...
Tesla is dying. Shares down 40+%. Plummeting sales will cancel its lucrative regulatory emissions credit business, responsible for 30% of total income. Insiders are dumping stock. Prices of used Teslas fall, with too many people selling.
March 10, 2025 at 3:12 PM
Tesla is dying. Shares down 40+%. Plummeting sales will cancel its lucrative regulatory emissions credit business, responsible for 30% of total income. Insiders are dumping stock. Prices of used Teslas fall, with too many people selling.
Well written essay describing how Elon Musk’s reputation has gone into free fall: garymarcus.substack.com/p/elon-musks....
Elon Musk's inability to listen to others is torching almost everything he touches
Last July, when I was still a regular user of X, I warned, not entirely in jest, that Elon Musk was taking a flamethrower to his own reputation.
garymarcus.substack.com
February 26, 2025 at 4:34 AM
Well written essay describing how Elon Musk’s reputation has gone into free fall: garymarcus.substack.com/p/elon-musks....
Reposted by Terence Plizga
Cool paper from Adam Kepecs' group, showing cortical microcircuits paired with ACh inputs could implement region-level credit assignment:
www.biorxiv.org/content/10.1...
I'm pumped by this, cause I've been saying for years that these circuits are prob involved in credit assignment! 😛
🧠📈 🧪
www.biorxiv.org/content/10.1...
I'm pumped by this, cause I've been saying for years that these circuits are prob involved in credit assignment! 😛
🧠📈 🧪
A Cortical Microcircuit for Region-Specific Credit Assignment in Reinforcement Learning
The distributed architecture of the cortex poses a fundamental challenge for reinforcement learning: how to assign credit specifically to regions that contribute to successful behavior? Cortical neuro...
www.biorxiv.org
October 18, 2024 at 2:32 PM
Cool paper from Adam Kepecs' group, showing cortical microcircuits paired with ACh inputs could implement region-level credit assignment:
www.biorxiv.org/content/10.1...
I'm pumped by this, cause I've been saying for years that these circuits are prob involved in credit assignment! 😛
🧠📈 🧪
www.biorxiv.org/content/10.1...
I'm pumped by this, cause I've been saying for years that these circuits are prob involved in credit assignment! 😛
🧠📈 🧪
New 20 hour bootcamp on Probability & Statistics from Steven Brunton: www.youtube.com/watch?v=sQqn.... Videos are released weekly, but the full playlist is available.
Probability and Statistics: Overview
YouTube video by Steve Brunton
www.youtube.com
October 18, 2024 at 7:43 PM
New 20 hour bootcamp on Probability & Statistics from Steven Brunton: www.youtube.com/watch?v=sQqn.... Videos are released weekly, but the full playlist is available.
When it comes to security, LLMs are like Swiss cheese — and that’s going to cause huge problems: open.substack.com/pub/garymarc...
When it comes to security, LLMs are like Swiss cheese — and that’s going to cause huge problems
Two new examples that are quite disconcerting
open.substack.com
October 18, 2024 at 3:36 PM
When it comes to security, LLMs are like Swiss cheese — and that’s going to cause huge problems: open.substack.com/pub/garymarc...
The predictions by @elonmusk re scientific progress are usually (always?) incorrect. He said he would solve fully self-driving vehicles by 2019. Didn’t happen. Even worse, his erratic “leadership” may have made things worse. www.washingtonpost.com/technology/2...
How Elon Musk knocked Tesla’s ‘Full Self-Driving’ off course
Before he took over Twitter he wanted to make Tesla cars drive themselves.
www.washingtonpost.com
April 9, 2024 at 11:58 PM
The predictions by @elonmusk re scientific progress are usually (always?) incorrect. He said he would solve fully self-driving vehicles by 2019. Didn’t happen. Even worse, his erratic “leadership” may have made things worse. www.washingtonpost.com/technology/2...
Attention is NOT everything. Linear performance gains require exponentially more data. Zero shot training is a mirage. Current AI architectures will never achieve the generalization required for AGI. arxiv.org/abs/2404.04125
April 8, 2024 at 11:54 PM
Attention is NOT everything. Linear performance gains require exponentially more data. Zero shot training is a mirage. Current AI architectures will never achieve the generalization required for AGI. arxiv.org/abs/2404.04125
Three researchers used machine learning and neural networks to read a 2,000-year-old scroll that had been scorched in the Mount Vesuvius eruption, winning a $700,000 prize. www.cbsnews.com/news/scroll-...
February 15, 2024 at 5:13 PM
Three researchers used machine learning and neural networks to read a 2,000-year-old scroll that had been scorched in the Mount Vesuvius eruption, winning a $700,000 prize. www.cbsnews.com/news/scroll-...
Foundational AI is not very foundational. It has no clue how the world works. If it detects a series of words or images that show up frequently, it regurgitates that as the answer to your question, even if that “answer” is completely wrong. garymarcus.substack.com/p/statistics...
Statistics versus Understanding: The Essence of What Ails Generative AI
The Foundation Remains Shaky
garymarcus.substack.com
February 14, 2024 at 7:55 PM
Foundational AI is not very foundational. It has no clue how the world works. If it detects a series of words or images that show up frequently, it regurgitates that as the answer to your question, even if that “answer” is completely wrong. garymarcus.substack.com/p/statistics...
Hmmm, I need to find some people to follow.
February 13, 2024 at 10:55 PM
Hmmm, I need to find some people to follow.
Testing … testing … 1 … 2 … 3 …
Hello Blue Sky denizens!
Hello Blue Sky denizens!
February 13, 2024 at 10:52 PM
Testing … testing … 1 … 2 … 3 …
Hello Blue Sky denizens!
Hello Blue Sky denizens!