antirez

7 ms·

121.
by antirez 6mo ago
Great. Potentially can go much faster rewriting it in terms of Metal shaders.
122.
by antirez 6mo ago
Similarly most leds are photo diodes, electric motors can be used to generate current, Peltier cells can be used to generate current, and so so forth. Many of such physical processes are invertible.
123.
by antirez 6mo ago
Yep, and it was simple in Redis to augment them with the "span" in order to support ranking, that is, given al element, tell me its position in the ordered collection.
124.
by antirez 6mo ago
I was joking. I notoriously write bad English but don't like using LLMs for writing. It removes personality.
125.
by antirez 6mo ago
"ad", with a single "d". So it's a Claude ad inside a Hetzner ad inside a decent grammar ad.
126.
by antirez 6mo ago
"West" when we talk about urban spaces, walk-accessible cities and public transportation is, IMHO, the wrong category. Europe and USA are very far apart.
127.
by antirez 6mo ago
I moved two servers, one from Linode and the other from DO to Hetzner a few months ago, with similar savings. The best part was that the two servers had tens of different sites running, implemented in different languages, with obsolete libr
128.
by antirez 6mo ago
I'm happy to see this short story posted here, it is one that I deeply loved when I was 14 or alike, and read it again multiple times. But I wonder: how did it survive in those sites without being shut down by the Asimov writings copyr
129.
by antirez 6mo ago
Are you sure you selected GPT 5.4-xhigh as model, in Codex? Because this makes a huge difference, and with this setting in my experience Codex outperforms Opus for almost every coding/reasoning task. Opus is still better often times wh
130.
by antirez 6mo ago
Let's see if even mid/big companies with tons of resources, with AI and the right tooling will continue to write webview-apps or, even worse, use some kind of multi target wrapper.
131.
by antirez 6mo ago
You don't need to trust anyone. GPT 5.4 xhigh is available and you can test it for $20, to verify it is actually able to find complex bugs in old codebases. Do the work instead of denying AI can do certain things. It's a matter of
132.
by antirez 6mo ago
In the Anthropic Mythos model cards they explicitly remarked that they didn't want Mythos to be specifically good at security. They trained it to be good at coding, and as a side effect the model is (obviously) good at security. This
133.
by antirez 6mo ago
Why this is the wrong analogy: finding hash collisions, while exponentially harder with N, is guaranteed to find, with enough work, some S so that H(S) satisfies N, so an asymmetry of resources used will have the side with more work event
134.
by antirez 6mo ago
Congrats: completely broken methodology, with a big conflict of interest. Giving specific bug hints, with an isolated function that is suspected to have bugs, is not the same task, NOR (crucially) is a task you can decompose the bigger task
135.
by antirez 6mo ago
Don't focus on what you prefer: it does not matter. Focus on what tool the LLM requires to do its work in the best way. MCP adds friction, imagine doing yourself the work using the average MCP server. However, skills alone are not su
136.
by antirez 6mo ago
In the SCSI controller work I mentioned, a very big part of the work was indeed reasoning about assembly code and how IRQs and completion of DMAs worked and so forth. Opus, even if TOOLS.md had the disassembler and it was asked to use it ma
137.
by antirez 6mo ago
Very good move. In my experience, for system programming at least, GPT 5.4 xhigh is vastly superior to Claude Opus 4.6 max effort. I ran many brutal tests, including reconstructing for QEMU the SCSI controller (not longer accessible) of a
138.
by antirez 6mo ago
Hi! Loved your recent post about the new era of computer security, thanks.
139.
by antirez 6mo ago
Firstly I have a long past in computer security, so: yes, I used to write exploits. Second, the vulnerability verification does not need being able to exploit, but triggering an ASAN assert. With memory corruption that's very simple of
140.
by antirez 6mo ago
That's not what is happening right now. The bugs are often filtered later by LLMs themselves: if the second pipeline can't reproduce the crash / violation / exploit in any way, often the false positives are evicted befor
141.
by antirez 6mo ago
Another potentially usable trick is the following: based on the observation that longer token budget improves model performances, one could generate solutions using a lot of thinking budget, then ask the LLM to turn the trace into a more co
142.
by antirez 6mo ago
This is very similar to what I stated here: https://x.com/antirez/status/2038241755674407005 That is, basically, you just rotate and use the 4 bit centroids given that the distribution is known, so you don't
143.
by antirez 6mo ago
Featuring the ELO score as the main benchmark in chart is very misleading. The big dense Gemma 4 model does not seem to reach Qwen 3.5 27B dense model in most benchmarks. This is obviously what matters. The small 2B / 4B models are i
144.
by antirez 6mo ago
The latest implementation of Picol has a Tcl-alike [expr] implemented in 40 lines of code that uses Pratt-style parsing: https://github.com/antirez/picol/blob/main/picol.c#L490
145.
by antirez 6mo ago
> If a harness is needed, it can make its own. If tools are needed, it can chose to bring out these tools. If I understand correctly the model can carry only very limited memory among tests, so it looks like it's not really possible
146.
by antirez 7mo ago
Exactly. I was reading all the other comments and wondering why many looked like they were talking of something else.
147.
by antirez 7mo ago
Basically this is true for most startups in the world BUT Cursor, so here you are kinda inverting the logic of the matter. Cursor is at a size that, if they wanted to use K2.5, they could clearly state that it was K2.5 or get a license to a
148.
by antirez 7mo ago
In programming, the only rule to follow is that there are no rules: only taste and design efforts. There are too many different conditions and tradeoffs: sometimes what is going to be the bottleneck is actually very clear and one could deci
149.
by antirez 7mo ago
So you love interacting with web sites sending requests with curl? And if you need the price of an AWS service, you love to guess the service name (querying some other endpoint), then ask some tool the price for it, get JSON back, and so fo
150.
by antirez 7mo ago
As yourself: what kind of tool I would love to have, to accomplish the work I'm asking the LLM agent to do? Often times, what is practical for humans to use, it is for LLMs. And the reply is almost never the kind of things MCP exports.

More ›