Very interesting work on the use of sparse neural networks for faster inference. This is a good way to run originally larger models with less CPU and memory penalty than using the original model while retaining result quality.

Interestingly it's likely we'll see more of those shortages in the future, probably in cycles and not only for cars. In the end this is mainly because the access to the raw materials to make semiconductors is becoming harder. As more parts of our society make things "smarter" there will be more competition from different industries to get access to those semiconductors.

Interesting discussion... could people go on strike toward providing data to big tech to demand change? At that point it seems to me more like an interesting thought experiment than something really doable... Probably worth monitoring where the conversation goes.

Fascinating findings around artificial multimodal neurons. Especially interesting is how they encode information but also how they open the door to new attacks to the corresponding neural networks.

Funny way to recall some of the odd complexity classes. :-)

So talking about the devil being in the details... That confirms profiling is still very much in the picture (unsurprisingly) but more disturbing:

• this has implications which creates new venues for easier fingerprinting (so extra risks!)
• this is in fact not that hard to cross-reference cohort and user identity (oops, wasn't it main initial motive to prevent this?)
• thus, this will help with monitoring behavior changes over time

And of course that's to be added to the fact that targeted advertising is very much not going away. FLoC or third party cookies... pick your poison I guess?

Interesting comparison even though the conclusion is slightly unsurprising: Pandas is slower but more convenient, Rust is fast, consumes less memory but more work is involved. At least this gives a few indications on what type of APIs could be added to Rust to ease some things. It also indicates that Pandas can be great to develop the pipeline with then a switch to Rust when this needs to be optimized for higher volumes of data.

Fascinating account on mental models and then statistical power

It starts with how a flawed mental model (coming from Facebook's founder) about identity and social role became imposed on others.

Then it continues on the mental model we tend to apply to vaccines. That shows again how bad we are at intuitively grasping statistics and their application. They do require an effort even when you're trained at them.

A neat little catalogue of monitoring tools on Linux. Learned a couple of them I didn't know of.

A nice list of easy mistakes one can make in their Nginx configuration opening the door to security issues.

Interesting take on optimisation and complexity... Indeed it's probably best to not use the latter to achieve the former otherwise that complexity can never be removed anymore.

EN: Great interview from one of the designers very involved with Framasoft. She's doing an excellent job at framing what UX design really is and highlighting some of the problems in the Free Software community.

FR: Excellente interview d'une designer très investie dans Framasoft. Elle fait un excellent boulot à expliquer ce qu'est réellement le UX design et à pointer certains des problèmes avec la communauté du Logiciel Libre.

EN: Interesting example (Unsplash as case study) on how simple design changes can greatly impact the ecological footprint of a service.

FR: Exemple intéressant (Unsplash est utilisé comme cas d'étude) où de simples modifications de conception peuvent grandement impacter l'empreinte écologique d'un service.

Obviously didn't read it all but this is a very large knowledge repository of practices from many companies one can get inspired by to work on Site Reliability Engineering. It is especially comprehensive since it's not only about technical tips but also deals with hiring, team building and culture (which is almost as important if not more).

Still a young project but that looks like an interesting approach. At least it would make for much smaller bundles than Electron while still enabling the same type of uses. I still have slight concerns about the complexity involved because of all the layers or how much memory this can consume (akin to Electron). Let's wait and see how it evolves.

Very nice set of tip and tricks for git. Quite a few are probably known to the more advanced users but still there might be one or two for everyone's taste.

Interesting analysis showing how the M1 SoC from Apple is really a good way for them to close and control their desktop ecosystem even more than before. In my opinion, if you care even one bit about your personal freedom you should likely stay away from this ecosystem.

Looks like an interesting book (didn't read it yet, would love an epub version out of the box) about AI and machine learning.

This new version supports JSON. Looks like learning XSLT wasn't completely a waste then. :-)

Very interesting FOSDEM talk which I see as part of the debate about "Open Source as just licenses"/"Post Open Source". This shows very well how the OSD is very narrow and that their should be a better alignment between what is labelled FOSS and the actual customs. This is something which licenses alone can't capture.