antirez

24 ms·

391.
by antirez 1y ago
Sure if you have a specific need you can specialize some NN with the right architecture, collecting the data, doing the training several times, testing the performances, ... Or: you can download an already built LLM and write a prompt.
392.
by antirez 1y ago
Sure, it's not a 1:1 map, but often who does an excellent job, does it along all the line: form, content, and the best papers are even shorter normally... They don't have to justify with many pages the lack of real content. About
393.
by antirez 1y ago
Yep I guess that's true, I often times see that the best papers are written better and make broader cultural references. However, recently, with all the non mother tongue English speakers around, especially from China, I often see grea
394.
by antirez 1y ago
I believe that it is not that style helps the content to be more right, not in the way PG believes (like in the example about writing shorter sentences), it is that a richer style (so, not shorter, but neither baroque: a style with more pos
395.
by antirez 1y ago
Very small: can run on the edge to allow something like a Raspberry Pi to make basic decisions for your appliance even if disconnected from the internet. Example: those are some time series parameters and instructions, decide if watering th
396.
by antirez 1y ago
I think that LLMs already changed the way we code, mostly, but I believe that agentic coding (vibe coding) is right now able to produce only bad results, and that the better approach is to use LLMs only to augment the programmer work (howev
397.
by antirez 1y ago
You may want to check my channel perhaps. There are videos of coding with LLMs doing real world things. Just search for "Salvatore Sanfilippo" on YouTube. The videos about coding+LLM are mostly in English.
398.
by antirez 1y ago
I never ever use agents for coding. Just the web interface of Gemini, Claude, ..., you are perfectly right that agentic coding just creates a layer of indetermination and chaos.
399.
by antirez 1y ago
I agree that instrumenting the model is useful in many contexts, but I don't believe it is something so unique to value Cursor such valuation, or all the attention RAG, memory, MCP get. If people say LLMs are going to be commodities (w
400.
by antirez 1y ago
My 2 centes: 1. Learn basic NNs at a simple level, build from scratch (no frameworks) a feed forward neural network with back propagation to train against MNIST or something as simple. Understand every part of it. Just use your favorite pro
401.
by antirez 1y ago
Check my YouTube channel for an experiment on how I added UTF-8 support to Kilo showcasing what was, back then, some early experiment at using LLMs to ship working code. I need to push the changes after some cleanup, but that's not eno
402.
by antirez 1y ago
Orthogonal: the lemons in the picture, from Palermo (Sicily), could not only being lemon or lemon-shaped soap, but also a sweet, our very famous "frutta martorana": https://en.wikipedia.org/wiki/Frutta_martora
403.
by antirez 1y ago
In such case, that would be very unfortunate.
404.
by antirez 1y ago
I believe that dropping an email will be enough and that the lack of attribution is not misunderstanding on how licensing works. The level of licensing ignorance in not so old developers is quite shocking. We are no longer in the 90s where
405.
by antirez 1y ago
Languages learning curve can be very funny. Phonetic regularity plays a lot in this game, for instance. French and Italian are very similar languages (more than Italian and Spanish), yet they are so phonetically apart (with Italian at the e
406.
by antirez 1y ago
Maybe you could look into establishing a proper-technical-terminology-direct-literal-translation-enforcement-protocol that uses "-" and translates German words more or less literally. The effect should be very obvious for German s
407.
by antirez 1y ago
So because they need to have a better business model, they will try to move users to weaker models compared to the best available? This "AI inside the editor" thing makes every day less sense in many dimensions: it makes you not r
408.
by antirez 1y ago
The best part of this is that, imagine tomorrow you want it running on ESP32 because out of the blue your board is now costly too much (this is just hypothetical: I'm for RP2040/50 for ALL uses but the most cheap requirements that
409.
by antirez 1y ago
Background for folks that are not into MicroPython. This release is so important because MicroPython is almost completely able to replace lower level languages like C for many embedded use cases on the RP20[45]0 and ESP32 and other MCUs at
410.
by antirez 1y ago
For a coincidence two hours ago I posted a video on my channel where it can be seen how king-man+woman also works with CLIP image embeddings. This may be obvious for people that worked with CLIP extensively, or that tried embeddings math in
411.
by antirez 1y ago
The classic: sin(x/2+t)+cos(y/2+sin(t))
412.
by antirez 1y ago
Related, at a different layer of abstraction: Kitty graphical protocol, implemented also by Ghostty terminal emulator.
413.
by antirez 1y ago
Interacting with the LLM directly is more work. What I mean is that a wrapper in the best conditions will not damage too much the quality of the LLM itself. In the chat, you continuously avoid suboptimal choices executed by the model, if
414.
by antirez 1y ago
Mandatory reminder that "agentic coding" works way worse than just using the LLM directly, steering it ad needed, filling the gaps, and so forth. The value is in the semantical capabilities of the LLM itself: wrapping it will make
415.
by antirez 1y ago
Well, 48 is 50ish more than 40s, but I understand what you mean: I believe that if I would not enter such mentality ASAP I wouldn't do many things I'm doing in recent years. To trigger it too late, is too late: a constant I see in
416.
by antirez 1y ago
I don't know about OpenSearch implementation, but recently I implemented from scratch Vector Sets for Redis using the HNSW as a data structure, and there are many other stores that use the same data structure. When HNSWs are well imple
417.
by antirez 1y ago
I'm 48 and one thing I universally notice, among my friends, is that they don't understand that GAME OVER is near and they should hurry up and do what the want to do. Instead they still feel like they are young, taking me sometime
418.
by antirez 1y ago
You can have two arguments that are semantically as distinct and important as quantity and price and be both integers, and if you swap them is a big issue anyway. And you would be forced, if you like this kind of programming, to create dist
419.
by antirez 1y ago
Also useful to turn spheres into two parts you can screw one with the other, like in this design of mine: https://makerworld.com/it/models/99223-death-star-christmas-...
420.
by antirez 1y ago
Thank you. Strange. If the memory numbers are accurate, it is so slow because likely layers are loaded from disk before inference of each layer or something like that, otherwise it could not do the inference of such model in 500MB. But if t

More ›