lukax

5 ms·

1.
by lukax 9d ago
OpenCode checks model name and registers the appropriate tools. const usePatch = model.modelID.includes("gpt-") && !model.modelID.includes("oss") && !model.modelID.includes("gpt-4") Pi us
2.
by lukax 9d ago
What matters more is that you use the tools that the target model was fine-tuned on. E.g. for editing files with Claude models you should use Edit(file_path, old_string, new_string, replace_all) but with GPT models you should use apply_patc
3.
by lukax 9d ago
https://news.ycombinator.com/submitted?id=vertigoruntime This looks like a new account from a PR agency focusing on AI labs.
4.
by lukax 1mo ago
This is only a problem with OpenAI Chat Completions API. With OpenAI Responses API and Anthropic Messages API a tool output can be text, image or file.
5.
DeepSeek v4 Price Increase(xcancel.com)
11 points by lukax 1mo ago|4 comments
6.
by lukax 1mo ago
DeepSeek-V4-Flash Input (Cache Miss) Previous: $0.14 New - Off-Peak: $0.22 (approx. 1.57x) New - Peak: $0.44 (approx. 3.14x) Input (Cache Hit) Previous: $0.0028 New - Off-Peak: $0.007 (2.5x) New - Peak: $0.014 (5x) Output Previous: $0.28 Ne
7.
by lukax 2mo ago
Yes. They are. https://news.ycombinator.com/item?id=46389934
8.
by lukax 2mo ago
They somehow forgot to mention that Hugging Face tried to use frontier models to analyze the attack but all models rejected. They had to use GLM 5.2 deployed locally. https://huggingface.co/blog/security-incident-july-2
9.
by lukax 2mo ago
Also for WASM? It's just not really there compared to Rust's wasm bindgen.
10.
by lukax 2mo ago
I can highly recommend Rust for this. The same Rust engine powering web app (WASM + React), iOS (SwiftUI), Android (Kotlin Compose) and desktop (Tauri). https://github.com/koofr/vault
11.
by lukax 3mo ago
Could it be that users of Pi are more senior and know better how to prompt and that's why the pass rate is higher?
12.
by lukax 3mo ago
Yes, this is a known issue. A significant amount of Edit tool calls fails in Pi witg newer models. https://lucumr.pocoo.org/2026/7/4/better-models-worse-tools/
13.
by lukax 3mo ago
Well, Microsoft just started offering Kimi K2.7 through Copilot hosted on Azure. https://github.blog/changelog/2026-07-01-kimi-k2-7-is-now-av... Cursor Composer 2 and 2.5 are also fine tunes of Kimi K2.5 It looks like
14.
by lukax 3mo ago
It's refreshing to see a tech article that isn't about AI. It feels like 5 years ago.
15.
by lukax 3mo ago
NUMA can cause really crappy performance. We deployed a Go based LLM gateway in Kubernetes deployed on a server with hundreds of CPU cores. We didn't explicitly set GOMAXPROCS so Go runtime scheduled goroutines over different CPUs and
16.
by lukax 3mo ago
Well, that's how data for training LLMs is scraped.
17.
by lukax 4mo ago
Zero data retention policies.
18.
by lukax 4mo ago
There is also Retrotick. https://retrotick.com/ It simulates x86 (win32 and win16) and implements Windows APIs in javascript and renders window frames with DOM and contents with canvas (e.g. GDI translates to browser canvas
19.
by lukax 4mo ago
Huawei Ascend AI Accellerators. DeepSeek V4 model architecture was optimized for Chinese hardware.
20.
by lukax 4mo ago
That's just not true. Let's say you have a form validation library with a public api that supports custom validators Validate(name string, value string) bool. Then you decide that your validator now needs to make an HTTP request.
21.
by lukax 4mo ago
I wonder what would happen if they used Kimi 2.5 directly instead of Cursor Composer 2.5. Composer is a fine tune of Kimi. Probably they didn't want to test "Chinese" models.
22.
by lukax 5mo ago
It looks like migration to Azure is not going very well https://news.ycombinator.com/item?id=45517173
23.
by lukax 5mo ago
They are migrating from their own datacenters to Azure
24.
by lukax 5mo ago
See the implementation of Python's math.isclose https://github.com/python/cpython/blob/d61fcf834d197f0113a6a...
25.
by lukax 5mo ago
You generally want both relative and absolute tolerances. Relative handles scale, absolute handles values near zero (raw EPSILON isn’t a universal threshold per IEEE 754). The usual pattern is abs(a - b) <= max(rel_tol * max(abs(a), abs(
26.
by lukax 6mo ago
Do you really need to roll your own NIO HTTP server? You could just use Jetty with virtual threads (still uses NIO under the hood though) and enjoy the synchronous code style (same as Go)
27.
by lukax 6mo ago
Rust wasm ecosystem also needs a lot of crates to do anything useful, a lot of them unmaintained.
28.
Vite Vulnerable to Arbitrary File Read via Vite Dev Server WebSocket(github.com)
2 points by lukax 6mo ago|1 comments
29.
by lukax 6mo ago
Combine that with CVE-2025-24010 and any website was able to read any file on developers' computers. https://github.com/advisories/GHSA-vg6x-rcgg-rjx6
30.
by lukax 7mo ago
Looks nice but still a bit sad that Flutter is used instead of something native given that they don't need the app to be cross-platform. Well, even Microsoft uses React Native for a lot of Windows-only apps.

More ›