Interesting inference engine. The design is clever with an hybrid CPU-GPU approach to limit the memory demand on the GPU and the amount of data transfers. The results are very interesting, especially surprising if the apparently very limited impact on the accuracy.

A bit of an older article I'm bumping into again. It lays out fairly well the current limits and issues with Free Software as it is defined today. I'm unconvinced it can be solved via licenses but the debate needs to happen... I feel that somehow it's too much ignored.

Interesting finding. This shows a potential issue in how identities are verified by providers.

This is a good set of advices for beginners. I especially like the ones about best practices, trying different things and why it makes sense to be conservative tech wise.

Nice state of the art view on how dynamic dispatch is implemented in several languages. Does a good way showing the trade-offs involved.

Good approach for tackling it indeed. The crux of the issue is really measuring the tech debt since it's still a fuzzy concept and we have no good metrics for it.

Here we are... We're really close to crossing into this territory where any fiction can disguise itself for reality. The problem is that we'll literally be drowning in such content. The social impacts can't be underestimated.

Interesting technique to speed up the generation of large language models.

Wonder how to implement such real-time simulations? This is a good summary of all the math involved. Also comes with code snippets and demos.

Interesting story about using unit tests by someone who thought it was a waste of time... until, they helped uncover a bug which was widespread. Also it was in an embedded context which comes with its own challenges.

This is indeed a more complex topic than it sounds. When someone complains about "technical debt" always inquire what it really means to them, what this is about, what are the symptoms.

That's a very good question. What will be left once all the hype is gone? Not all bubbles leaving something behind... we can hope this one will.

When SEO and generated content meet... this isn't pretty. The amount of good content on the web reduced in the past decade, it looks like we're happily crossing another threshold in mediocrity.

Or why using a profiler is not as easy as it sounds. This requires quite some experience and the ability to tap in other information not present in the profile.

Back to the history of VCS, anyone still remember and used SCCS? Well, I did use it...

This is unsurprisingly highly depend on the actual code, not only on the hardware.

Interesting new attack on the SSH protocol. This is hard to achieve outside of the LAN though.

As Threads being connected to the Fediverse might turn into a reality, this article becomes all the more important. The question of this connection being even desirable is an important one.

Looks like Meta is moving forward with more ActivityPub compatibility for Threads. This raises real questions about what they genuinely want to implement and what they'll abandon along the way.

The actual dangers of generative AI. Once the web is flooded with generated content, what will happen to knowledge representation and verifiability?