Mitch Harris
@maharrit.bsky.social
NLP/ML/AI/Data Science, Teaching/Consulting/Advising
Reposted by Mitch Harris
Reposting with alt text
November 18, 2025 at 9:34 AM
Reposting with alt text
Reposted by Mitch Harris
Vous assistez, quasiment en direct à la formation d'un cratère sur la Lune, avec un impact filmé ce matin !
November 16, 2025 at 3:26 PM
Vous assistez, quasiment en direct à la formation d'un cratère sur la Lune, avec un impact filmé ce matin !
Reposted by Mitch Harris
WeRateDogs turns 10 years old today! A very silly idea to numerically objectify dogs has become one of the kindest and most powerful online communities ever. Thank you for joining us over the last decade. I never knew what this could be until you all showed me. Here's to many more good dogs ❤️
November 15, 2025 at 3:42 PM
WeRateDogs turns 10 years old today! A very silly idea to numerically objectify dogs has become one of the kindest and most powerful online communities ever. Thank you for joining us over the last decade. I never knew what this could be until you all showed me. Here's to many more good dogs ❤️
Reposted by Mitch Harris
There was never any point to having reference letters. That's why we've all started using AI to do this nonesense task.
References should only be used for short-listed candidates for important positions/awards, and ideally, be done via a call to get the most honest opinion possible.
References should only be used for short-listed candidates for important positions/awards, and ideally, be done via a call to get the most honest opinion possible.
From my discussions with other faculty, the use of generative AI I hear about the most is writing reference letters.
What's the point of having reference letters anymore if everyone is just having them written by machine?
What's the point of having reference letters anymore if everyone is just having them written by machine?
November 14, 2025 at 7:10 PM
There was never any point to having reference letters. That's why we've all started using AI to do this nonesense task.
References should only be used for short-listed candidates for important positions/awards, and ideally, be done via a call to get the most honest opinion possible.
References should only be used for short-listed candidates for important positions/awards, and ideally, be done via a call to get the most honest opinion possible.
Reposted by Mitch Harris
Free trade, open borders, Mexi-Korean Halal trucks on every corner.
November 14, 2025 at 11:44 AM
Free trade, open borders, Mexi-Korean Halal trucks on every corner.
Reposted by Mitch Harris
People sometimes ask why we want to rate DAGs. Couldn't we just rate statistical models instead?
We like to explain it with capybaras and bicycles.
We like to explain it with capybaras and bicycles.
October 17, 2025 at 12:52 PM
People sometimes ask why we want to rate DAGs. Couldn't we just rate statistical models instead?
We like to explain it with capybaras and bicycles.
We like to explain it with capybaras and bicycles.
Reposted by Mitch Harris
postultimate - the day after the absolute deadline, which you missed
November 12, 2025 at 3:02 PM
postultimate - the day after the absolute deadline, which you missed
Reposted by Mitch Harris
training data is literally the most important ‘ingredient’ that can reveal so much about model behaviour
but again, opening up your training data is like putting your dirty laundry out and can bring an entire company down
but again, opening up your training data is like putting your dirty laundry out and can bring an entire company down
November 12, 2025 at 10:49 AM
training data is literally the most important ‘ingredient’ that can reveal so much about model behaviour
but again, opening up your training data is like putting your dirty laundry out and can bring an entire company down
but again, opening up your training data is like putting your dirty laundry out and can bring an entire company down
Reposted by Mitch Harris
ultimate = last in a series
penultimate = second-to-last in a series
antepenultimate = third-to-last in a series
preantepenultimate = fourth-to-last in a series
propreantepenultimate = fifth-to-last in a series
penultimate = second-to-last in a series
antepenultimate = third-to-last in a series
preantepenultimate = fourth-to-last in a series
propreantepenultimate = fifth-to-last in a series
November 12, 2025 at 3:00 PM
ultimate = last in a series
penultimate = second-to-last in a series
antepenultimate = third-to-last in a series
preantepenultimate = fourth-to-last in a series
propreantepenultimate = fifth-to-last in a series
penultimate = second-to-last in a series
antepenultimate = third-to-last in a series
preantepenultimate = fourth-to-last in a series
propreantepenultimate = fifth-to-last in a series
Reposted by Mitch Harris
AI Bait & Switch:
bait: we’re gonna make an AI that can solve any problem experts could solve. it’s gonna transform the whole world.
switch: what we have actually made is fun and amazing, but rarely reliable and often makes mistakes – but ordinary people makes mistakes, too. So … AGI solved!
bait: we’re gonna make an AI that can solve any problem experts could solve. it’s gonna transform the whole world.
switch: what we have actually made is fun and amazing, but rarely reliable and often makes mistakes – but ordinary people makes mistakes, too. So … AGI solved!
June 9, 2025 at 7:56 PM
AI Bait & Switch:
bait: we’re gonna make an AI that can solve any problem experts could solve. it’s gonna transform the whole world.
switch: what we have actually made is fun and amazing, but rarely reliable and often makes mistakes – but ordinary people makes mistakes, too. So … AGI solved!
bait: we’re gonna make an AI that can solve any problem experts could solve. it’s gonna transform the whole world.
switch: what we have actually made is fun and amazing, but rarely reliable and often makes mistakes – but ordinary people makes mistakes, too. So … AGI solved!
Reposted by Mitch Harris
"And how do you determine what is factual?"
"If it matches what I currently believe to be true, then it is factual. If it does not match what I currently believe to be true, then it is biased and false."
"Are you a child."
"If it matches what I currently believe to be true, then it is factual. If it does not match what I currently believe to be true, then it is biased and false."
"Are you a child."
October 30, 2025 at 3:21 PM
"And how do you determine what is factual?"
"If it matches what I currently believe to be true, then it is factual. If it does not match what I currently believe to be true, then it is biased and false."
"Are you a child."
"If it matches what I currently believe to be true, then it is factual. If it does not match what I currently believe to be true, then it is biased and false."
"Are you a child."
Reposted by Mitch Harris
C'est la première carte que vous avez vraiment regardé !
Ne mentez pas :)
#Astérix #Gaulois #VillageGaulois
Ne mentez pas :)
#Astérix #Gaulois #VillageGaulois
October 29, 2025 at 8:41 AM
C'est la première carte que vous avez vraiment regardé !
Ne mentez pas :)
#Astérix #Gaulois #VillageGaulois
Ne mentez pas :)
#Astérix #Gaulois #VillageGaulois
Reposted by Mitch Harris
Prediction models that are used to guide medical decisions are usually regulated under medical device regulation. This means, putting a calculator out there to promote the use your new prediction model is likely to break some rules.
November 1, 2025 at 1:46 PM
Prediction models that are used to guide medical decisions are usually regulated under medical device regulation. This means, putting a calculator out there to promote the use your new prediction model is likely to break some rules.
Reposted by Mitch Harris
The lessons, such as they are:
1) Don’t trust anything you see online
2) Hard to imagine that AI video will not have a major impact on short form video platforms
3) Its pretty fun to be able to summon whatever idea you have out of latent space
1) Don’t trust anything you see online
2) Hard to imagine that AI video will not have a major impact on short form video platforms
3) Its pretty fun to be able to summon whatever idea you have out of latent space
October 31, 2025 at 1:41 AM
The lessons, such as they are:
1) Don’t trust anything you see online
2) Hard to imagine that AI video will not have a major impact on short form video platforms
3) Its pretty fun to be able to summon whatever idea you have out of latent space
1) Don’t trust anything you see online
2) Hard to imagine that AI video will not have a major impact on short form video platforms
3) Its pretty fun to be able to summon whatever idea you have out of latent space
Reposted by Mitch Harris
The thing about using AI art that I don't get is that it is not that expensive to hire an artist. I needed a logo for my podcast and it only cost $50, and the artist did it exactly how I wanted. It looks way better than anything AI could produce!
October 31, 2025 at 4:46 AM
The thing about using AI art that I don't get is that it is not that expensive to hire an artist. I needed a logo for my podcast and it only cost $50, and the artist did it exactly how I wanted. It looks way better than anything AI could produce!
Reposted by Mitch Harris
“The bad review will come from your list of suggested reviewers”
October 31, 2025 at 7:39 AM
“The bad review will come from your list of suggested reviewers”
Reposted by Mitch Harris
A Brief History of Halloween Isn’t Fun Anymore
🧵
🧵
October 26, 2025 at 6:58 PM
A Brief History of Halloween Isn’t Fun Anymore
🧵
🧵
This is why it's hard to explain anything at all.
It's like going up to someone who doesn't know about conservation of energy, and telling them you have a wheel that never stops spinning, and expecting them to be blown away.
COMIC ◆ www.smbc-comics.com/comic/intuit...
PATREON ◆ www.patreon.com/ZachWeinersm...
STORE ◆ smbc-store.myshopify.com
COMIC ◆ www.smbc-comics.com/comic/intuit...
PATREON ◆ www.patreon.com/ZachWeinersm...
STORE ◆ smbc-store.myshopify.com
October 27, 2025 at 7:30 AM
This is why it's hard to explain anything at all.
Reposted by Mitch Harris
It's like going up to someone who doesn't know about conservation of energy, and telling them you have a wheel that never stops spinning, and expecting them to be blown away.
COMIC ◆ www.smbc-comics.com/comic/intuit...
PATREON ◆ www.patreon.com/ZachWeinersm...
STORE ◆ smbc-store.myshopify.com
COMIC ◆ www.smbc-comics.com/comic/intuit...
PATREON ◆ www.patreon.com/ZachWeinersm...
STORE ◆ smbc-store.myshopify.com
October 26, 2025 at 10:30 PM
It's like going up to someone who doesn't know about conservation of energy, and telling them you have a wheel that never stops spinning, and expecting them to be blown away.
COMIC ◆ www.smbc-comics.com/comic/intuit...
PATREON ◆ www.patreon.com/ZachWeinersm...
STORE ◆ smbc-store.myshopify.com
COMIC ◆ www.smbc-comics.com/comic/intuit...
PATREON ◆ www.patreon.com/ZachWeinersm...
STORE ◆ smbc-store.myshopify.com
Reposted by Mitch Harris
Technically, this is a book of spells.
October 24, 2025 at 3:08 PM
Technically, this is a book of spells.
Reposted by Mitch Harris
And yet here we are.
October 22, 2025 at 1:31 AM
And yet here we are.
Reposted by Mitch Harris
Under Over Over - West Hampstead
So it's taken a while, but I've finally found a location where you can clearly see London Underground trains traveling over London Overground trains.
The vantage point is at West Hampstead Overground station at the end of the platform.
So it's taken a while, but I've finally found a location where you can clearly see London Underground trains traveling over London Overground trains.
The vantage point is at West Hampstead Overground station at the end of the platform.
October 20, 2025 at 6:54 AM
Under Over Over - West Hampstead
So it's taken a while, but I've finally found a location where you can clearly see London Underground trains traveling over London Overground trains.
The vantage point is at West Hampstead Overground station at the end of the platform.
So it's taken a while, but I've finally found a location where you can clearly see London Underground trains traveling over London Overground trains.
The vantage point is at West Hampstead Overground station at the end of the platform.
Reposted by Mitch Harris
It's quite ironic how people who built the best prediction models are such bad predictors themselves.
They throw all their knowledge about how to make good predictions overboard and just claim things like: AI will replace radiologists in a few years or when they expect AGI.
They throw all their knowledge about how to make good predictions overboard and just claim things like: AI will replace radiologists in a few years or when they expect AGI.
December 17, 2024 at 7:55 AM
It's quite ironic how people who built the best prediction models are such bad predictors themselves.
They throw all their knowledge about how to make good predictions overboard and just claim things like: AI will replace radiologists in a few years or when they expect AGI.
They throw all their knowledge about how to make good predictions overboard and just claim things like: AI will replace radiologists in a few years or when they expect AGI.
1) awesome to propose a "comprehensive and testable definition" that is clear and coherent.
2) I don't disagree with its contents but
a) it probably won't match in general people's intuitive definitions
b) missing -a lot-.
c) still too vague
x.com/DanHendrycks...
www.agidefinition.ai
2) I don't disagree with its contents but
a) it probably won't match in general people's intuitive definitions
b) missing -a lot-.
c) still too vague
x.com/DanHendrycks...
www.agidefinition.ai
Dan Hendrycks on X: "The term “AGI” is currently a vague, moving goalpost. To ground the discussion, we propose a comprehensive, testable definition of AGI. Using it, we can quantify progress: GPT-4 (2023) was 27% of the way to AGI. GPT-5 (2025) is 58%. Here’s how we define and measure it: 🧵 https://t.co/CVfwRF1FpU" / X
The term “AGI” is currently a vague, moving goalpost. To ground the discussion, we propose a comprehensive, testable definition of AGI. Using it, we can quantify progress: GPT-4 (2023) was 27% of the way to AGI. GPT-5 (2025) is 58%. Here’s how we define and measure it: 🧵 https://t.co/CVfwRF1FpU
x.com
October 18, 2025 at 7:04 PM
1) awesome to propose a "comprehensive and testable definition" that is clear and coherent.
2) I don't disagree with its contents but
a) it probably won't match in general people's intuitive definitions
b) missing -a lot-.
c) still too vague
x.com/DanHendrycks...
www.agidefinition.ai
2) I don't disagree with its contents but
a) it probably won't match in general people's intuitive definitions
b) missing -a lot-.
c) still too vague
x.com/DanHendrycks...
www.agidefinition.ai
Reposted by Mitch Harris
I very much appreciate this: "Technologies like LLMs have utility, but the absurd way they've been over-hyped, the fact they're being forced on everyone, and the insistence on ignoring the many valid critiques about them make it very difficult to focus on legitimate uses where they might add value."
Okay, for the folks who asked: here's the majority AI view, writing up the reasonable, thoughtful view on AI that the vast majority of people in tech hold, that gets overshadowed by the bluster and hype of the tycoons trying to shill their nonsense. anildash.com/2025/10/17/t... Please share!
The Majority AI View - Anil Dash
A blog about making culture. Since 1999.
anildash.com
October 18, 2025 at 6:06 PM
I very much appreciate this: "Technologies like LLMs have utility, but the absurd way they've been over-hyped, the fact they're being forced on everyone, and the insistence on ignoring the many valid critiques about them make it very difficult to focus on legitimate uses where they might add value."