My second short story release in English is ready: Tales of Illustrious Computer Scientists: Iola Varga, nun and computer scientist.

Don't ask me why sometimes I need to reach such a point.

A repository with two of the very early Redis versions and some email digging to reconstruct the very first days history:.

RT @cooperx86: @antirez Looking forward to watching this tomorrow. I've been playing with them recently and they work so well! Really natur….

If you are interested to use the new Redis Vector Sets, I just uploaded a YouTube video with the first introduction. More will follow, to progressive details up to the implementation level:.

Such a great news! A very warm welcome, Marc.

@ollama Maybe @mixedbreadai knows more about it.

The feature is the 931th (index 930) of the returned vectors. Such vectors were obtained using the @ollama API.

Preparing the YouTube video series on Redis Vector Sets I just discovered a fancy mxbai-large embedding model "issue". It over-expresses a specific component in every output I tested. This may be some very generic/ubiquitous feature the model needed to model based on training.

RT @badlogicgames: Just found this by @antirez on HN. This perfectly captures the issue. Full autonomy is a bad idea given the current stat….

RT @jasonnochlin: More XOR magic: you can extend antirez’s approach to pinpoint the bad link in the special case of exactly one bad link. h….

About the recent XOR article in Hacker News.

Profiling is always an eyes-opening activity.

Solved by using dsymutil redis-server, but before you have to make sure Redis is not compiled with -flto, that is enabled by default with -O3. I guess last time I profiled Vector Sets they were still built as a module and not linked to Redis itself.

WTF of the day: Instruments on MacOS no longer showing the symbols of Redis. Every day some tangent problem that should not be there or that software itself should be able to trace.

What you would expect from somebody with so many open buttons on the chest?.

With prompt caching I don't mean KV caches, but caching the embedding of questions and the reply, so that no tokens are generated AT ALL when a very similar question is seen again. question -> embedding cache. question2 -> search -> match -> serving old replies.

Prompt caching is ruining LLMs. Now Gemini does it even in the middle of the chat if the question looks generic and is cached, even if it is quite obviously related to the previous context.

Why I dislike LoRaWAN as much as I love LoRa itself, this is anyway a big milestone for RadioLib and for work environments where LoRaWAN must be used. Kudos.

LLMs peek performance: having to review shitty code written by somebody else, something that was always horrific to do for the (almost) active effort in such code to hide how things work / complicate stuff. Poo for the mind. Now LLMs can do it for us \o/.

About the recente Apple paper: many of us believe CoT is a form of sampling/search in the representations the model possess: this allows to "ground" the reply with a large context of related ideas/info. No surprise that the Tower of Hanoi is a problem where sampling doesn't help.