Guillaume Ausset
@ausset.me
I used to have a PhD in ML doing survival analysis. I still do but I used to.
Looking for a job in ML for (European) defense.
Looking for a job in ML for (European) defense.
The insane length people will go to to keep using « Python » ( scare quote because sometime the result isn’t even compatible with the whole python ecosystem ) is truly baffling
April 12, 2025 at 7:33 PM
The insane length people will go to to keep using « Python » ( scare quote because sometime the result isn’t even compatible with the whole python ecosystem ) is truly baffling
Oh so all the other types of bias are handwaved by « well it’s how the data/world is, deal with it » but here clearly the world/data is wrong and it must be « corrected »? Funny how that works
April 10, 2025 at 4:13 PM
Oh so all the other types of bias are handwaved by « well it’s how the data/world is, deal with it » but here clearly the world/data is wrong and it must be « corrected »? Funny how that works
They can just say « it was a prank bro » to their new supplier to break the contracts. Apparently it’s legal.
April 10, 2025 at 7:19 AM
They can just say « it was a prank bro » to their new supplier to break the contracts. Apparently it’s legal.
Not only that but it’s more importantly ( and damningly… ) just not very good
April 9, 2025 at 7:41 AM
Not only that but it’s more importantly ( and damningly… ) just not very good
I’m going to send the UN my report entitled « The World as we know it will end if we don’t give Guillaume Ausset One trillion dollars ». It evens include (AI generated slop) illustration to drive the point home.
April 4, 2025 at 6:16 PM
I’m going to send the UN my report entitled « The World as we know it will end if we don’t give Guillaume Ausset One trillion dollars ». It evens include (AI generated slop) illustration to drive the point home.
It's cheap if you are american.
It's not if you are not.
It's not if you are not.
February 26, 2025 at 10:36 AM
It's cheap if you are american.
It's not if you are not.
It's not if you are not.
Turns out they were right, gender affirming care is a problem. Just mmm not like they thought.
February 26, 2025 at 8:53 AM
Turns out they were right, gender affirming care is a problem. Just mmm not like they thought.
Most / all these countries went back to democracy thanks to the support or overt action of stronger democracies.
What stronger-than-the-US democracy exists that can do that for the US?
What stronger-than-the-US democracy exists that can do that for the US?
February 23, 2025 at 2:37 PM
Most / all these countries went back to democracy thanks to the support or overt action of stronger democracies.
What stronger-than-the-US democracy exists that can do that for the US?
What stronger-than-the-US democracy exists that can do that for the US?
No they aren’t. They want to *appear* appalled. You don’t applaud someone for saying « hey this is too much » at a Nazi conference. Instead you ask them why the fuck they were a speaker at a Nazi conference in the first place
February 23, 2025 at 2:18 PM
No they aren’t. They want to *appear* appalled. You don’t applaud someone for saying « hey this is too much » at a Nazi conference. Instead you ask them why the fuck they were a speaker at a Nazi conference in the first place
No the important thing to remember is that we had to *fight* them with all our might for them to lose otherwise they would have won.
February 21, 2025 at 9:41 AM
No the important thing to remember is that we had to *fight* them with all our might for them to lose otherwise they would have won.
Things are much easier when you just violate the law, constitution and very basis of democracy, it turns out.
February 15, 2025 at 8:43 PM
Things are much easier when you just violate the law, constitution and very basis of democracy, it turns out.
That patch is awesome 🥺
February 14, 2025 at 6:06 PM
That patch is awesome 🥺
In my experiments it was actually positive to have a "weak" and potentially biased estimator for G ( like something very very smooth, or clipped away from 0) to prevent numerical explosion.
February 7, 2025 at 1:52 PM
In my experiments it was actually positive to have a "weak" and potentially biased estimator for G ( like something very very smooth, or clipped away from 0) to prevent numerical explosion.
At the time it seemed (to me) that the problem was when the estimator of G gave probabilities too close to 0 to some samples ( and therefore basically infinite weights ) and it would get lost in a loop of increasingly weird estimated losses.
February 7, 2025 at 1:50 PM
At the time it seemed (to me) that the problem was when the estimator of G gave probabilities too close to 0 to some samples ( and therefore basically infinite weights ) and it would get lost in a loop of increasingly weird estimated losses.
Interesting!
We proposed something very similar a while ago ( jmlr.org/papers/volum... ).
I toyed at the time with the idea of jointly estimating the weight and the survival estimator in an iterative fashion but it wasn't stable.
Did you have any divergence problems ?
We proposed something very similar a while ago ( jmlr.org/papers/volum... ).
I toyed at the time with the idea of jointly estimating the weight and the survival estimator in an iterative fashion but it wasn't stable.
Did you have any divergence problems ?
February 7, 2025 at 1:49 PM
Interesting!
We proposed something very similar a while ago ( jmlr.org/papers/volum... ).
I toyed at the time with the idea of jointly estimating the weight and the survival estimator in an iterative fashion but it wasn't stable.
Did you have any divergence problems ?
We proposed something very similar a while ago ( jmlr.org/papers/volum... ).
I toyed at the time with the idea of jointly estimating the weight and the survival estimator in an iterative fashion but it wasn't stable.
Did you have any divergence problems ?
The "when" is doing a lot of work here. Even an "if" is stretch.
Plus hey, they can just pardon themselves of any crime.
Plus hey, they can just pardon themselves of any crime.
February 7, 2025 at 8:56 AM
The "when" is doing a lot of work here. Even an "if" is stretch.
Plus hey, they can just pardon themselves of any crime.
Plus hey, they can just pardon themselves of any crime.
You can use PHA as a filament. it’s 100% *actually biodegradable*. Ofc you shouldn’t really make functional parts with it ( well, it’s unwise ) but still. PLA is renewable and reusable too ( and *technically* recyclable, even if only in industrial operations )
February 3, 2025 at 6:49 AM
You can use PHA as a filament. it’s 100% *actually biodegradable*. Ofc you shouldn’t really make functional parts with it ( well, it’s unwise ) but still. PLA is renewable and reusable too ( and *technically* recyclable, even if only in industrial operations )
And yes I’m aware you can manufacture some simple barrels like for the FGC9, but that’s not what people do except in Myanmar
February 3, 2025 at 6:45 AM
And yes I’m aware you can manufacture some simple barrels like for the FGC9, but that’s not what people do except in Myanmar
Ghost guns are basically not a problem in other countries as they have sensible laws and decided that the regulated part should be the pressure bearing parts ( slide, bolt, barrel ), aka the hard to make parts to make
February 3, 2025 at 6:44 AM
Ghost guns are basically not a problem in other countries as they have sensible laws and decided that the regulated part should be the pressure bearing parts ( slide, bolt, barrel ), aka the hard to make parts to make
Meh. The thing people forget to mention when talking about « ghost guns » is that it’s only possible thanks to the US stupid laws. People print the kinda useless plastic frame ( it’s regulated) and then buy freely the rest of the whole ass metal firearm ( it’s not regulated ).
February 3, 2025 at 6:42 AM
Meh. The thing people forget to mention when talking about « ghost guns » is that it’s only possible thanks to the US stupid laws. People print the kinda useless plastic frame ( it’s regulated) and then buy freely the rest of the whole ass metal firearm ( it’s not regulated ).
And of course they are an offshoot from LessWrong.
Not particularly surprised for some reason…
Not particularly surprised for some reason…
February 3, 2025 at 6:30 AM
And of course they are an offshoot from LessWrong.
Not particularly surprised for some reason…
Not particularly surprised for some reason…
The easiest and most sensible solution is obviously to move to Czech Republic
January 26, 2025 at 3:08 PM
The easiest and most sensible solution is obviously to move to Czech Republic