antirez (@antirez) header
antirez (@antirez) profile picture

Reproducible bugs are candies. I like programming too much for not liking automatic programming. In elenco AGCOM.

Sicily, Italy
Joined May 2007
Photo by antirez (@antirez)Photo by antirez (@antirez)Photo by antirez (@antirez)Photo by antirez (@antirez)Photo by antirez (@antirez)Photo by antirez (@antirez)Photo by antirez (@antirez)Photo by antirez (@antirez)Photo by antirez (@antirez)Photo by antirez (@antirez)
My second short story release in English is ready: Tales of Illustrious Computer Scientists: Iola Varga, nun and computer scientist. invece.org/iola.html
24
13
1
267
373,170
Why right now LLMs benchmarking is broken? Because we need fundamentally benchmarks by *trusted* third parties that are not VC backed but AI-labs jointly backed that perform mainly benchmarks where models *compete* with each other on tasks.
23
9
1
210
9,617
So we live in the future and even if YouTube Studio stats are *terrible*, we could create our own, right? Not so easy: today's corporate software is not just bad, is also closed: no API for you.
10
1
1
138
9,670
Accept the reality that Google AI answers are for many searches an improvement for the user. Would not be maybe in the web of 20 years ago. Now it was a shit party of useless sites, advertising of all the kind, and other garbage.
30
4
4
484
18,954
Hi friends, in the latest weeks I worked with GPT 6.2 Astra. With it I built Redis features and most of YouRec (testflight.apple.com/join/N4…). My feeling is that it is the *best* model for system programming in C and for complex data structures. (continue)
86
79
26
1,498
98,749
I found its abilities to reason about complicated bugs superior to Fable 5.2 (which is also an awesome model). For YouRec it did a wonderful job, using Fable 5.2 as a consultant for the UI / UX elements.
4
3
2
222
16,384
Huang returning as CEO would also normalize people having their fucking life that may not algin with puritans of all kinds, and the fact that, first, you do the investigation, to see if there is real wrongdoing, and only later ask for heads.
40
61
6
1,145
49,369
Strong opinions since 1977
113
5,377
If you, like me, are in the effort of inference software, be aware that you are just a facilitator: the whole battle is played in the field of open models and affordable GPUs and VRAM, which is entirely out of our small fishes control.
39
25
7
548
28,845
YouRec is live in the AppStore (MIT licensed source code in my GitHub): apps.apple.com/app/id6830752…
27
22
2
304
27,662
This application is making my recording / editing of videos a zero issues experience. Trivial things like remembering the order of my preferred mic lists. I just in the car, connect, open, and it selects the right one. A good experience is made of the right details.
2
26
5,510
I'm surely using @FrameworkPuter Desktop computers to support LLM models, but I'm also using them to develop Redis and *what wonderful machines* they are! Real beasts with a lot of memory, a fan that is low pitch and never create issues. It is the computer I own that I love most.
15
4
231
19,910
So about this NVIDIA scandal. Jensen Huang is of mainland Chinese origins and driven a GPU revolution in the US. GPUs made in Taiwan with NL stuff. The dream for NN based AI is European-American (with Hinton being the star in the sky). Can't you see this is a human-wide effort?
34
17
2
510
38,913
I have been youtubing for quite some time now (2 years next month). I use max 5% of my time for YouTube, on an average week, so I built tools along the way. YouRec is one of such tools, the camera+editing app I needed for the iPhone. What to join the beta? testflight.apple.com/join/N4…
23
11
1
236
16,283
The fact AI is now in a new stage is clear from the fact that the biggest produced of AI slop in the past, that is: a small prompt producing a lot of code (CC @noperator) now it can actually produce a good result if the operation has clear boundaries: "Translate this in Rust".
27
4
217
22,855
The next step in the future of all that will be: "Write me the best possible database" -> When this will produce something great, even if it is is an "open" task, we will be at the next level, however it is a continuum, so this kind of POV can be used to track AI progresses.
9
1
47
4,346
Obvious for people that tested stuff seriously but worth repeating ad libitum.
I don’t know why this needs to be re-iterated: if you are lucky and having a M5 Max / M5 Ultra, don’t use Qwen 3.8 27B, you will be much happier with a low-bit DeepSeek 4.1 Flash 1017. A good 2-bit quant for that model is just so much better than 27B dense.
15
6
1
360
35,493
I'm a bit worried that the recent private-life stuff of an individual that is mortal and can commit errors will have the effect, due to news release timing, of weakening his very needed position on (Chinese or not) open weight models.
31
23
3
478
28,973
I had no accesso to Mythos / Fable 5.2 before tonight, but I wanted to check if a system recently checked by expert humans and updated had still issues. 20 minutes later there were two severe bugs uncovered. I'll not tell which project it is, it does not matter. Mythos is good.
25
18
689
42,638
I'm extremely grateful to @HRF for the grant generously provided to me for DwarfStar (see: hrf.org/latest/hrfs-ai-for-i…), this grant will allow me to provide support for the open models as new affordable inference systems and new models will be available. Thank you.
20
32
512
18,286
I already see clear signs that who is refusing AI assistance for programming it is starting to lag behind in a significant way, where this means: they are living in a bubble where losing work will happen faster compared to the other folks.
63
42
13
869
44,640
After hte latest changes, DwarfStar can run DS4.1Flash in SSD streaming in a quite fast way. If it wasn't for the overthinking of course, that totally *did not* effect 4.0 Flash. It's a 4.1 flaw partially fixed in the latest update.
23
15
6
339
31,572
*affect of course
9
4,499
The kimi-k3 branch of DwarfStar can now do K3.1 PRO inference directly from safetensors weights with SSD streaming. Slow, but good enough for QA tasks.
23
2
1
164
15,769