zx - Network

I should note while I can do LLM stuff for you, I am very good at deploying things that are many orders magnitude cheaper and with better...

I should note while I can do LLM stuff for you, I am very good at deploying things that are many orders magnitude cheaper and with better performance on your business metrics!

Hey!I am looking for work again. So if you are Germany based and looking for an awesome machine learning, data scientist, statistician,...

Hey!

I am looking for work again. So if you are Germany based and looking for an awesome machine learning, data scientist, statistician, software human I want to talk to you!

#fedihired #getfedihired

My lodging plans for #xoxofest fell through. Does anybody know of someone with couch space for a few days?

My lodging plans for #xoxofest fell through. Does anybody know of someone with couch space for a few days?

The curse of being able to read a paper and know at a glance which plotting software they used to make their figures.Then drawing spurious...

The curse of being able to read a paper and know at a glance which plotting software they used to make their figures.

Then drawing spurious inferences about the authors because of it.

Not enough people are doing prior predictive checks in their Bayesian modeling workflow and it shows!

Not enough people are doing prior predictive checks in their Bayesian modeling workflow and it shows!

I wish pytorch just copied the jax vmap functionality instead of went with the "heavily-inspired" line

I wish pytorch just copied the jax vmap functionality instead of went with the "heavily-inspired" line

Anybody ever wish their training dashboard was like a video game where you could interactively, during training, increase/decrease their...

Anybody ever wish their training dashboard was like a video game where you could interactively, during training, increase/decrease their learning rate, batch size, or some other hyperparameter?

Machine learning models are often have a hard dependency on the specific library versions used for training it. A migration plan that...

Machine learning models are often have a hard dependency on the specific library versions used for training it. A migration plan that doesn't include migrating these models to newer library versions is a recipe for disaster.

But there are no decent tools to support this, and for many libraries it might be impossible to migrate a model without re-training from scratch.

Does anyone have a good solution for migrating ML models trained with outdated software? I'm surprised there aren't more ways to...

Does anyone have a good solution for migrating ML models trained with outdated software? I'm surprised there aren't more ways to migrate models with the infra used to create it

There are so many opportunities for massive improvements. The Cola paper gives a hint of some of them and that's just in terms of using...

There are so many opportunities for massive improvements. The Cola paper gives a hint of some of them and that's just in terms of using less memory and having more numerically stable algorithms

https://arxiv.org/abs/2309.03060

I really wish more APIs embraced treating positive semidefinite matrices as their own data type where internally the Cholesky decomposition...

I really wish more APIs embraced treating positive semidefinite matrices as their own data type where internally the Cholesky decomposition is stored.

There are whole classes of bugs and errors that could be eliminated at the type level this way

We need GPL-like licenses for datasets. That way when you build on a model that consumes a freely available dataset you also have to make...

We need GPL-like licenses for datasets. That way when you build on a model that consumes a freely available dataset you also have to make the resulting model freely available.

I'm at #PyConDe / #PyDataBerlin today and tomorrow. If you or your company are interested in Bayesian modelling, find me!

I'm at #PyConDe / #PyDataBerlin today and tomorrow. If you or your company are interested in Bayesian modelling, find me!

I can't believe how many bugs were fixed by multiplying by two in a single line of my code

I can't believe how many bugs were fixed by multiplying by two in a single line of my code

One reason these autoregressive models have difficulty making jokes is in many ways jokes are written punchline first and then a premise is...

One reason these autoregressive models have difficulty making jokes is in many ways jokes are written punchline first and then a premise is created in reaction to it.

Hey @tomjennings Are there any old textfiles I can read about how content moderation happened on FidoNet? Did sysops ever block anything...

Hey @tomjennings

Are there any old textfiles I can read about how content moderation happened on FidoNet? Did sysops ever block anything from coming in from other sysops?

My partner and I share unreasonably strong opinions on the Likert scale, and it's honestly not something I thought we'd bond over...

My partner and I share unreasonably strong opinions on the Likert scale, and it's honestly not something I thought we'd bond over when we first started dating.

@mcc I went to your talk yesterday, and I think I agree. The security for a social network that uses content-addressed data has to be at the...

@mcc I went to your talk yesterday, and I think I agree. The security for a social network that uses content-addressed data has to be at the level of accessing the network itself.

Oh @emilymbender a new Mastodon frontend has appeared. It's still fairly new, but does look actively developed.elk.zoneBelow is an...

Oh @emilymbender a new Mastodon frontend has appeared. It's still fairly new, but does look actively developed.

elk.zone

Below is an example of the UI using a celebrity's post

https://elk.zone/@MarkRuffalo@mastodon.social/109638532487238395

I never stop finding it fascinating that when people are asked to predict what AI / ML will do to transform the world. They are much more...

I never stop finding it fascinating that when people are asked to predict what AI / ML will do to transform the world. They are much more pessimistic about the potential within their own industry vs industries they know much less about.