aka-rider

11 ms·

31.
by aka-rider 2mo ago
Human-written. I was trying to be short and straight to the point. LLM-powered rewrites and huge refactors are better done using 1 additional step "convert the code to <something> that represents it best". The simplest examp
32.
by aka-rider 2mo ago
I consider a wig. I'm still on a fence with Rust at this point. see comments above
33.
by aka-rider 2mo ago
I realized that I haven't answered the question. These $400 also include the tests. Fable ported "human fuzzing session" (the best bug hunter) from Go to Rust and used it to validate everything else. I used hierarchical state
34.
by aka-rider 2mo ago
All very good questions. Agents are actively destroy QA gates in many ways, usually by cheating ("the test is buggy, not my changes" — changes the test), or just rot QA slowly by writing buggy overcomplicated tests What works for
35.
I used Fable to rewrite 65kLoC of Go in Rust. It cost $400(iurii.net)
78 points by aka-rider 2mo ago|71 comments
36.
by aka-rider 2mo ago
Following the recent "Rewriting bun in Rust" I thought to run an experiment which turned out to be success.
37.
by aka-rider 2mo ago
Smaller models are overconfident and have a hard time to self-correct. If it’s stuck, usually that’s it. Bigger models “understand” better, both the prompt and the contents. If you will try to read a paper together with a smaller model, the
38.
by aka-rider 2mo ago
All models find vulnerabilities. What is special about this generation of SOTA models, including Mythos/Fable (the same model), GPT-5.6, Kimi-K3, and now GLM-5.3 — they can chain vulnerabilities and produce working exploits. Look at th
39.
by aka-rider 2mo ago
Not OP, but driving knowledge bases is the poster child use case for me https://github.com/aka-rider/llm-wiki I started with Karpathy's LLM wiki, and did everything he said not to do - downgraded the model to mere
40.
by aka-rider 3mo ago
What will be the price? GPUs costs are x3, memory and SSD x5. When OpenAI will be looking to sell and everyone knows it, the price would be x0,7 for depreciated used hardware or go bankrupt yourself.
41.
by aka-rider 3mo ago
The best metaphor I heard about LLMs so far - it's a search engine. The bigger the model the bigger the search space. Small models tend to have a "tunnel vision" or fall into "rabbit holes" - they have less visible
42.
by aka-rider 3mo ago
There is no replacement for large models, indeed. And this is not the point I'm trying to make. There are numerous applications for self-hosted models. As a simplest example, when you ask "explain what this code does" advanta
43.
by aka-rider 3mo ago
Model+harness combination means a lot. That's why all major labs are making their own. All models have quircks harnesses know about "you are reading the same file 3rd time you are in the loop, step back" I tried all frontier
44.
by aka-rider 3mo ago
On MacOS you have a seatbelt sandbox built-in. On Linux - docker with SELinux or similar utility over namespaces. You need to model attack vector first. `rm -rf` - restricted write `curl malware.sh | sh` restrict execution from writeable di
45.
by aka-rider 3mo ago
I run Qwen3.6 on RTX4090, and it does amazing job for the most parts. For coding task, one needs to break the session among multiple calls I made https://github.com/aka-rider/orqestra but it's possible to do the s
46.
A Better Programmer: Part 2 – The Conduct of No Regrets(medium.com)
3 points by aka-rider 7y ago|0 comments
47.
A Better Programmer: Part 1 – Software Design(medium.com)
1 points by aka-rider 7y ago|0 comments