@letmutex

Programming isn't real.

github.com/letmutex
Joined January 2024
  • Tweets1.2K
  • Following70
  • Followers23
251Media
Open Source VS Open Weight
DeepSeek
@deepseek_aiOct 20
🚀 Introducing DeepInfer, our in-house inference framework, now fully open source! Powering DeepSeek-V4.1-Flash at scale with high throughput and low latency. We’re sharing our production stack to help the community lower compute costs, reduce energy use, and make AI more affordable. 🔗 github.com/deepseek-ai/DeepI… 1/4

GitHub - deepseek-ai/DeepInfer: DeepInfer: High-performance, large-scale inference framework for...

DeepInfer: High-performance, large-scale inference framework for DeepSeek models - deepseek-ai/DeepInfer

github.com
17
Gemini 4 Argon is not a good model. It hallucinates too much and can't execute instructions precisely. Three.js and SVG benchmarks are cursed.
28
Why is everyone reinventing TypeScript? Someone please reinvent Gradle and JVM.
8
Wait, it's happening?
Mark Zuckerberg
@finkdOct 15
Muse Spark Open -- weights releasing today -- runs great on new Mac Studios thanks to work with @johnternus and Apple.
10
Got muted after only 3-4 simple queries with Argon. It looks like RSI didn't work well.
9
It looks like Google is so satisfied with this launch. I have a feeling that Gemini 4 Argon might last for at least 6 months. And new Flash might not come out this month.
32
I don't need Gemini 4 Argon. I need cheaper, smarter, and more efficient Gemini Flash.
92
Rust is
3
Old computers make software faster. Gradle Indexing is now much faster in my static Kotlin LSP. Indexing time changes in a real project with 900 Kotlin files and 270 jars/aars: - 1.2s -> 730ms on M4, 1.6x faster - 15s -> 7.8s on a 2017 i5 7th Gen laptop, 1.9x faster
18
Finding completions from 1M symbols now takes a few milliseconds in my Kotlin LSP. Up to 826x faster compared to the previous problematic implementation.
22
882 tests so far. I think AI unit tests are not all bad. They force AI to build a working LSP.
letmutex@letmutexOct 10
A static Kotlin LSP is coming. Powered by Rust and completely JVM-free. It's fast and will be faster. It takes 1.3s to index 900 kt files and 2.3k dependency jars on M4 (no cache). It supports: - Basic type inference - Auto completion - Auto import - Find references and more
32
A static Kotlin LSP is coming. Powered by Rust and completely JVM-free. It's fast and will be faster. It takes 1.3s to index 900 kt files and 2.3k dependency jars on M4 (no cache). It supports: - Basic type inference - Auto completion - Auto import - Find references and more
1
67
No Anthropic. So you start to think why Anthropic is now leading the AI race...
Google DeepMind
@GoogleDeepMindOct 7
SynthID Detector is now available to everyone. 🌐 Check whether online content was generated using @GoogleAI, or with tools from our industry partners – including @OpenAI, @NVIDIA, Kakao and coming soon, @Apple. Try it out → synthid.com
9
You can actually put a placeholder on @ArtificialAnlys for weeks.
5
It now feels like Google is serving 1-bit Gemini 3.8 Flash in Antigravity.
1
38
I think git log --grep is reward hacking. Gemini 3.8 Flash really likes it.
34
Every time I see Gemini setting a timer
8
Cost x5 over 100k tokens Luna survived
Claude
@claudeaiOct 7
Replying to @claudeai
Haiku 5.5 is especially good value on tasks under 100,000 tokens, which make up around 90% of requests to our previous Haiku model.
8
What?
Mistral AI
@MistralAIOct 6
Meet Mistral Large 4, aka Le Chonk. • 1T parameters, natively multimodal. 49B active. It is the best open weights model from US or Europe on aggregated benchmarks. • State-of-the-art on critical workloads, including cyber defense, manufacturing and finance and it surpasses closed frontier models on visual grounding. • Forged in Europe end-to-end and is deployable from Europe via our own Mistral Cloud infrastructure. • Available to all via API today. Working with cybersecurity partners privately. Open weights release end of October.
5
What? No... 3.7 Flash is better than 3.8 Flash
Bee
@thtbee_Oct 3
Gemini 3.6 & 3.7 Flash are getting deprecated very soon.. possibly today! Fast modes from the model selector views of Flash models are also gone.. I think we are reallyy close to the rollout of Gemini 4 Argon.. I'll keep you guys updated as I see any more changes!
17
Load more