π§ AI Engineer | π» Software Architect | π Ethical Acceleration #ethacc | Diving deep into the matrix of AI and programming languages. @perarneng.bsky.
Westcoast of Sweden
Joined March 2010
- Tweets10.9K
- Following2.8K
- Followers1.2K
- Likes32.5K
I'm building an inference engine for the DGX Spark that runs one model only for now, Qwen3.6-35B-A3B (NVFP4), and takes every shortcut that buys speed.
Rust + CUDA, written with AI but spec-driven, with strict architecture rules and lots of tests.
Against the vLLM recipe for the same model, on 9 benchmarks (math, code, tool calls, long context): same quality within noise, 26% faster.
A hobby project. Perhaps open source it if im happy with the quality and some other model supported.
#AI #DGXSpark @NVIDIAAI
π€ Made with AI
Its can be expensive to run locally, but its freedom #ethacc
We become more and more one with our AI:s. We thrive with our AI assistants. Companies like @AnthropicAI is trying to limit our personal freedom. Where does it end? They are not doing anything illegal but its anti human. If a human can legally hack their own TV they should be able to do so with an AI assistant. If a human can legally disassemble an app they should be able to do so with AI. Do not support anti human companies. Use Open Weight Models instead if you can. #ai
If you bought the TV it's your device. I really wish @AnthropicAI will fail because they are anti humanity. They try to disguise it as having higher morals but it's actually against human freedom. It's their service so they have their right to limit it but that doesn't mean that its right. They claim to be about humans but they work against human nature. Humans want freedom to explore and discover things. Hacking your own TV is one example that should be encouraged. Sure it might void the warranty but that is the owners responsibility. It's such a strange world we live in where we need to turn to China to get freedom to be humans and not have a company to set immoral limits on us. Open Weight Models are the future and only way forward. It's about freedom from corporate greed.
#php2shell proves that we need to stop using C/C++. Because humans can't write safe enough C/C++ code. It's not a language war. It's a consequence of AI being to good at finding CVE:s. #CyberSecurity
Governments needs to install php honeypots everywhere to catch the hackers. #php2shell #CyberSecurity
4th floor in our office building. No sound or feelings of vibrations anywhere else but the bottle picked it up. They are doing som building close by though so it's of natural causes. Still cool. #vibrations
How long wil it take for @sundarpichai and @satyanadella to realize that AI is coming after their companies in two ways:
1. Software and services can easily be recreated now with AI
2. Cloud hosting with open source products is so much easier now with AI
3. Most importantly. Mass unemployment will destroy the purchasing power of the end consumers
It's like they and everyone elese is racing towards the edge. They don't have any choice but they could collectively stand together and speak openly about how the economy is going to crash.
I wish people could understand this.
I agree, I build 4 projects at the time at home and have really fun doing it. I really enjoy it. I reverse engineer things to automate my home. I build new software for hobby purposes. At work I build AI agents that can automate processes that will make it possible for 2 to do the work of 10. I believe it's possible to do so for all white collar worker. The consequences are that this will cause mass unemployment which will trigger people to be extremely negative towards AI and regulations will follow since we live in a democracy. This means that progress will halt and the fun will stop. Similar thing happened with nuclear energy but now it will affect people more directly and make them angry. Either they or a family member or relative will be affected. Most software engineers know the should be worried unless they are in denial. Most white collar workers will become worried. The people worried will become negative towards AI and that will affect politician in the end.
I love AI and don't want it to be regulated. What I don't like is how it is being introduced without a plan to manage the consequences of its implementation.
youtu.be/nV3tBWnZ4wg?si=GLzLβ¦
Initial attempt at recreating George, BARTβs announcement voice, with AI helping me build a speech synthesizer in Rust. Original first, recreation second. It was much harder than I thought. Please try to take on the challenge and see if you can do it. π #BARTAIchallenge #AI
Initial attempt at recreating George, BARTβs announcement voice, with AI helping me build a speech synthesizer in Rust. Original first, recreation second. It was much harder than I thought. Please try to take on the challenge and see if you can do it. π #BARTAIchallenge #AI
Btw rules of the challenge: Use AI to generate code that generates voice. Do not train models or use voice cloning. #BARTAIchallenge
The only thing you have to start with is this reference audio from this page: bart.gov/news/articles/2009/β¦ #BARTAIchallenge