Steven Walton
@swalton.ai
Ph.D. from University of Oregon
Visiting Scholar SHI Lab @ Georgia Tech
Opinions expressed are those of my cat
Visiting Scholar SHI Lab @ Georgia Tech
Opinions expressed are those of my cat
You need to determine if they're using actual logic or something else. Actually the stronger evidence is in their explanation, not their failure to get the right answer. I can even get them to get the right answer with bad logic
September 28, 2025 at 9:56 PM
You need to determine if they're using actual logic or something else. Actually the stronger evidence is in their explanation, not their failure to get the right answer. I can even get them to get the right answer with bad logic
Do you have strong evidence of such a claim? I think it'll be hard to prove tbh.
But we do have a lot of evidence of failures, which suggest we should doubt that claim. Failures like trivial math calculations.
Links in the image alts
GPT-5, Claude Opus 4.1, Gemini 2.5 Pro
But we do have a lot of evidence of failures, which suggest we should doubt that claim. Failures like trivial math calculations.
Links in the image alts
GPT-5, Claude Opus 4.1, Gemini 2.5 Pro
September 28, 2025 at 9:36 PM
Do you have strong evidence of such a claim? I think it'll be hard to prove tbh.
But we do have a lot of evidence of failures, which suggest we should doubt that claim. Failures like trivial math calculations.
Links in the image alts
GPT-5, Claude Opus 4.1, Gemini 2.5 Pro
But we do have a lot of evidence of failures, which suggest we should doubt that claim. Failures like trivial math calculations.
Links in the image alts
GPT-5, Claude Opus 4.1, Gemini 2.5 Pro
An example of the latter is you're forcing someone to type "I'm suicidal" (no correction) rather than "I'm going to commit suicide." (See screenshot)
Great job, you protected no one and made it harder for someone in crisis to reach out for help. At best you add a straw to a camel's back.
Great job, you protected no one and made it harder for someone in crisis to reach out for help. At best you add a straw to a camel's back.
September 28, 2025 at 8:35 PM
An example of the latter is you're forcing someone to type "I'm suicidal" (no correction) rather than "I'm going to commit suicide." (See screenshot)
Great job, you protected no one and made it harder for someone in crisis to reach out for help. At best you add a straw to a camel's back.
Great job, you protected no one and made it harder for someone in crisis to reach out for help. At best you add a straw to a camel's back.
I'm not sure I'll ever learn how to type correctly on an iPhone...
The phrase I want to type is "shoot themselves in the foot"
Apple please:
1) don't correct multiple words at once. It creates more work
2) censor words or phrases. You're not protecting anyone, you're creating more harm
The phrase I want to type is "shoot themselves in the foot"
Apple please:
1) don't correct multiple words at once. It creates more work
2) censor words or phrases. You're not protecting anyone, you're creating more harm
September 28, 2025 at 8:35 PM
I'm not sure I'll ever learn how to type correctly on an iPhone...
The phrase I want to type is "shoot themselves in the foot"
Apple please:
1) don't correct multiple words at once. It creates more work
2) censor words or phrases. You're not protecting anyone, you're creating more harm
The phrase I want to type is "shoot themselves in the foot"
Apple please:
1) don't correct multiple words at once. It creates more work
2) censor words or phrases. You're not protecting anyone, you're creating more harm
over-parameterization is
If you carefully look at the outputs you'll actually observe more subtle errors.
These models are wildly impressive, but we can't effectively use them nor improve them without recognizing this.
If you carefully look at the outputs you'll actually observe more subtle errors.
These models are wildly impressive, but we can't effectively use them nor improve them without recognizing this.
May 27, 2025 at 11:19 PM
over-parameterization is
If you carefully look at the outputs you'll actually observe more subtle errors.
These models are wildly impressive, but we can't effectively use them nor improve them without recognizing this.
If you carefully look at the outputs you'll actually observe more subtle errors.
These models are wildly impressive, but we can't effectively use them nor improve them without recognizing this.
Let the record show, on May 27th, 2025
GPT-o3, a model for advanced reasoning, still cannot figure out the trivial boat crossing problem without some extra help.
This came up because a discussion of how more parameters does not mean better generalization. It's the nuances that matter, that's what
GPT-o3, a model for advanced reasoning, still cannot figure out the trivial boat crossing problem without some extra help.
This came up because a discussion of how more parameters does not mean better generalization. It's the nuances that matter, that's what
May 27, 2025 at 10:59 PM
Let the record show, on May 27th, 2025
GPT-o3, a model for advanced reasoning, still cannot figure out the trivial boat crossing problem without some extra help.
This came up because a discussion of how more parameters does not mean better generalization. It's the nuances that matter, that's what
GPT-o3, a model for advanced reasoning, still cannot figure out the trivial boat crossing problem without some extra help.
This came up because a discussion of how more parameters does not mean better generalization. It's the nuances that matter, that's what
If you're wondering what the Founding Fathers would do, we don't have to guess.
This isn't a one off. They all wrote about it prolifically. Benjamin Franklin upped the ratio to 100 to 1.
Freedom isn't free and this is the price. Even for the worst of the worst, innocent UNTIL PROVEN guilty.
This isn't a one off. They all wrote about it prolifically. Benjamin Franklin upped the ratio to 100 to 1.
Freedom isn't free and this is the price. Even for the worst of the worst, innocent UNTIL PROVEN guilty.
April 18, 2025 at 12:30 AM
If you're wondering what the Founding Fathers would do, we don't have to guess.
This isn't a one off. They all wrote about it prolifically. Benjamin Franklin upped the ratio to 100 to 1.
Freedom isn't free and this is the price. Even for the worst of the worst, innocent UNTIL PROVEN guilty.
This isn't a one off. They all wrote about it prolifically. Benjamin Franklin upped the ratio to 100 to 1.
Freedom isn't free and this is the price. Even for the worst of the worst, innocent UNTIL PROVEN guilty.
One of the most influential philosophies to the founding of the US was the idea from Blackstone. It is something EVERY lawyer is well versed in and is the foundation of US law.
Innocent until PROVEN guilty.
I don't care who we're talking about. YOU ARE NOT A CRIMINAL UNTIL A COURT SAYS YOU ARE
Innocent until PROVEN guilty.
I don't care who we're talking about. YOU ARE NOT A CRIMINAL UNTIL A COURT SAYS YOU ARE
April 18, 2025 at 12:30 AM
One of the most influential philosophies to the founding of the US was the idea from Blackstone. It is something EVERY lawyer is well versed in and is the foundation of US law.
Innocent until PROVEN guilty.
I don't care who we're talking about. YOU ARE NOT A CRIMINAL UNTIL A COURT SAYS YOU ARE
Innocent until PROVEN guilty.
I don't care who we're talking about. YOU ARE NOT A CRIMINAL UNTIL A COURT SAYS YOU ARE
How do I know when my cat is fully charged?
April 12, 2025 at 9:53 PM
How do I know when my cat is fully charged?
In case anyone was wondering, yes, you can transcode 8 videos to AV1 WHILE playing Elden Ring at maximum settings on a NVIDIA 4080S. I would suggest keeping it at 7 though. Little extra wiggle room.
April 5, 2025 at 5:15 AM
In case anyone was wondering, yes, you can transcode 8 videos to AV1 WHILE playing Elden Ring at maximum settings on a NVIDIA 4080S. I would suggest keeping it at 7 though. Little extra wiggle room.
Also for fun, it is a shorter drive (distance) from Lisbon to Kyiv than it is from LA to DC
March 29, 2025 at 8:42 PM
Also for fun, it is a shorter drive (distance) from Lisbon to Kyiv than it is from LA to DC
For fun, here are all the states that voted Democrat in the 2024 election.
March 29, 2025 at 8:33 PM
For fun, here are all the states that voted Democrat in the 2024 election.
Filling out applications to #NVIDIA and I found this in the skills section.
I didn't know they had such an interest in camping! Rest assured NVIDIA, I have the skills to survive the robot apocalypse and live in the woods! I hope it doesn't come to that, but I'm ready!
I didn't know they had such an interest in camping! Rest assured NVIDIA, I have the skills to survive the robot apocalypse and live in the woods! I hope it doesn't come to that, but I'm ready!
March 27, 2025 at 7:52 PM
Filling out applications to #NVIDIA and I found this in the skills section.
I didn't know they had such an interest in camping! Rest assured NVIDIA, I have the skills to survive the robot apocalypse and live in the woods! I hope it doesn't come to that, but I'm ready!
I didn't know they had such an interest in camping! Rest assured NVIDIA, I have the skills to survive the robot apocalypse and live in the woods! I hope it doesn't come to that, but I'm ready!
I thought the whole purpose of generating human faces was because we ourselves are really good evaluators. So we can make up for where the mathematical models fail. Millions of years of evolution vs statistical distances.
The details matter. Be through crochet, eyes, reflections, or whatever.
The details matter. Be through crochet, eyes, reflections, or whatever.
March 27, 2025 at 6:47 PM
I thought the whole purpose of generating human faces was because we ourselves are really good evaluators. So we can make up for where the mathematical models fail. Millions of years of evolution vs statistical distances.
The details matter. Be through crochet, eyes, reflections, or whatever.
The details matter. Be through crochet, eyes, reflections, or whatever.
But eyes are similarly nuanced. In fact, they are nuance in the way that they require some world modeling. Consistency in what is not obviously in the scene.
The subtle effects for eyes are some of the reasons people tried to put transformers in GANs (like my own work StyleNAT). For long range
The subtle effects for eyes are some of the reasons people tried to put transformers in GANs (like my own work StyleNAT). For long range
March 27, 2025 at 6:47 PM
But eyes are similarly nuanced. In fact, they are nuance in the way that they require some world modeling. Consistency in what is not obviously in the scene.
The subtle effects for eyes are some of the reasons people tried to put transformers in GANs (like my own work StyleNAT). For long range
The subtle effects for eyes are some of the reasons people tried to put transformers in GANs (like my own work StyleNAT). For long range
Why don't eyes matter? They are part of the underlying physics of the scene. They are critical to understanding lighting. "gateway to the soul" to be poetic.
I do agree that crochet patters and thread based art are good tests. They often highlight models' fundamental flaws in lack of nuance.
I do agree that crochet patters and thread based art are good tests. They often highlight models' fundamental flaws in lack of nuance.
March 27, 2025 at 6:47 PM
Why don't eyes matter? They are part of the underlying physics of the scene. They are critical to understanding lighting. "gateway to the soul" to be poetic.
I do agree that crochet patters and thread based art are good tests. They often highlight models' fundamental flaws in lack of nuance.
I do agree that crochet patters and thread based art are good tests. They often highlight models' fundamental flaws in lack of nuance.
And they blocked me lol
A good reminder, some people aren't promoting the cause they claim to be. This person clearly isn't for privacy and their profile is an obvious farce.
I'll stand by what I said. Install @signal.org
Communication should be private by default
A good reminder, some people aren't promoting the cause they claim to be. This person clearly isn't for privacy and their profile is an obvious farce.
I'll stand by what I said. Install @signal.org
Communication should be private by default
March 25, 2025 at 1:55 AM
And they blocked me lol
A good reminder, some people aren't promoting the cause they claim to be. This person clearly isn't for privacy and their profile is an obvious farce.
I'll stand by what I said. Install @signal.org
Communication should be private by default
A good reminder, some people aren't promoting the cause they claim to be. This person clearly isn't for privacy and their profile is an obvious farce.
I'll stand by what I said. Install @signal.org
Communication should be private by default
I think you're being disingenuous and spreading unnecessary fear. Nothing is invulnerable to attacks like that.
It's the equivalent of tricking someone into giving you the password to their safe. That doesn't mean don't get a safe.
Or... not zipping up your bulletproof vest, as I already said
It's the equivalent of tricking someone into giving you the password to their safe. That doesn't mean don't get a safe.
Or... not zipping up your bulletproof vest, as I already said
March 25, 2025 at 1:35 AM
I think you're being disingenuous and spreading unnecessary fear. Nothing is invulnerable to attacks like that.
It's the equivalent of tricking someone into giving you the password to their safe. That doesn't mean don't get a safe.
Or... not zipping up your bulletproof vest, as I already said
It's the equivalent of tricking someone into giving you the password to their safe. That doesn't mean don't get a safe.
Or... not zipping up your bulletproof vest, as I already said
More idiotically, @signal.org has disappearing messages. If you're really concerned about leaks you can enable them.
Burn after reading.
This wouldn't have saved them there but it is a step that reduces risk of leaks. You lose your phone? Well good thing messages are deleted!
Burn after reading.
This wouldn't have saved them there but it is a step that reduces risk of leaks. You lose your phone? Well good thing messages are deleted!
March 25, 2025 at 1:22 AM
More idiotically, @signal.org has disappearing messages. If you're really concerned about leaks you can enable them.
Burn after reading.
This wouldn't have saved them there but it is a step that reduces risk of leaks. You lose your phone? Well good thing messages are deleted!
Burn after reading.
This wouldn't have saved them there but it is a step that reduces risk of leaks. You lose your phone? Well good thing messages are deleted!
Signal tells EVERY user in a group chat any time someone new is added. Not only that, but who added who.
Signal EXPLICITLY does this for security. To make it hard to sneak new members in that could spy on you.
They even warn you if contacts change numbers and is a common joke among users
Signal EXPLICITLY does this for security. To make it hard to sneak new members in that could spy on you.
They even warn you if contacts change numbers and is a common joke among users
March 25, 2025 at 1:22 AM
Signal tells EVERY user in a group chat any time someone new is added. Not only that, but who added who.
Signal EXPLICITLY does this for security. To make it hard to sneak new members in that could spy on you.
They even warn you if contacts change numbers and is a common joke among users
Signal EXPLICITLY does this for security. To make it hard to sneak new members in that could spy on you.
They even warn you if contacts change numbers and is a common joke among users
When you ask Siri:
"What month is it?"
This is a pretty big fail.
IME a lot of Apple ML is not highly robust. Needs more augmentation. Hopefully this isn't because lack of data harvesting (I believe it can be solved without that)
(I tried doing video but it cuts audio when talking to Siri)
"What month is it?"
This is a pretty big fail.
IME a lot of Apple ML is not highly robust. Needs more augmentation. Hopefully this isn't because lack of data harvesting (I believe it can be solved without that)
(I tried doing video but it cuts audio when talking to Siri)
March 21, 2025 at 6:31 PM
When you ask Siri:
"What month is it?"
This is a pretty big fail.
IME a lot of Apple ML is not highly robust. Needs more augmentation. Hopefully this isn't because lack of data harvesting (I believe it can be solved without that)
(I tried doing video but it cuts audio when talking to Siri)
"What month is it?"
This is a pretty big fail.
IME a lot of Apple ML is not highly robust. Needs more augmentation. Hopefully this isn't because lack of data harvesting (I believe it can be solved without that)
(I tried doing video but it cuts audio when talking to Siri)
The algorithms aren't alright
How do you serve a leech who has fewer views over the official source? The problem is this seems to be the norm, not a one off.
Anyways, some cool stuff with GTC. Watch here
www.youtube.com/live/_waPvOw...
How do you serve a leech who has fewer views over the official source? The problem is this seems to be the norm, not a one off.
Anyways, some cool stuff with GTC. Watch here
www.youtube.com/live/_waPvOw...
March 18, 2025 at 6:00 PM
The algorithms aren't alright
How do you serve a leech who has fewer views over the official source? The problem is this seems to be the norm, not a one off.
Anyways, some cool stuff with GTC. Watch here
www.youtube.com/live/_waPvOw...
How do you serve a leech who has fewer views over the official source? The problem is this seems to be the norm, not a one off.
Anyways, some cool stuff with GTC. Watch here
www.youtube.com/live/_waPvOw...
@andrewgwils.bsky.social, you used "we" on a single author paper. You know what you have to do...
en.wikipedia.org/wiki/F._D._C...
This is a great paper and you should read it even if there is no cat coauthor.
en.wikipedia.org/wiki/F._D._C...
This is a great paper and you should read it even if there is no cat coauthor.
March 17, 2025 at 10:31 PM
@andrewgwils.bsky.social, you used "we" on a single author paper. You know what you have to do...
en.wikipedia.org/wiki/F._D._C...
This is a great paper and you should read it even if there is no cat coauthor.
en.wikipedia.org/wiki/F._D._C...
This is a great paper and you should read it even if there is no cat coauthor.
Crows are extremely intelligent animals and can learn complex behaviors.
Crows were the first to invent calculus, beating Newton by 200 years!
Crows also like food
Crows were the first to invent calculus, beating Newton by 200 years!
Crows also like food
March 13, 2025 at 6:33 PM
Crows are extremely intelligent animals and can learn complex behaviors.
Crows were the first to invent calculus, beating Newton by 200 years!
Crows also like food
Crows were the first to invent calculus, beating Newton by 200 years!
Crows also like food