Fernando Fernandes Neto (@fernandonetoai) 's Twitter Profile
Fernando Fernandes Neto

@fernandonetoai

Machine Learning and AI researcher wayyy before all this hype.

ID: 1731750437521444864

calendar_today04-12-2023 19:01:15

651 Tweet

884 Followers

77 Following

Chujie Zheng (@chujiezheng) 's Twitter Profile Photo

Proud to introduce Group Sequence Policy Optimization (GSPO), our stable, efficient, and performant RL algorithm that powers the large-scale RL training of the latest Qwen3 models (Instruct, Coder, Thinking) 🚀 📄 huggingface.co/papers/2507.18…

Proud to introduce Group Sequence Policy Optimization (GSPO), our stable, efficient, and performant RL algorithm that powers the large-scale RL training of the latest Qwen3 models (Instruct, Coder, Thinking) 🚀

📄 huggingface.co/papers/2507.18…
Fernando Fernandes Neto (@fernandonetoai) 's Twitter Profile Photo

Playing further with the GPT-OSS, it seems it is suffering from 'syntheticitis' - a disease observed in Phi models for a long time 😅

Fernando Fernandes Neto (@fernandonetoai) 's Twitter Profile Photo

Maybe the new hiring process should be based on testing if people can code consistently with AI instead of the traditional pair coding sessions...

Skywork (@skywork_ai) 's Twitter Profile Photo

Matrix-Game 2.0 — The FIRST open-source, real-time, long-sequence interactive world model Last week, DeepMind's Genie 3 shook the AI world with real-time interactive world models. But... it wasn't open-sourced. Today, Matrix-Game 2.0 changed the game. 🚀 25FPS. Minutes-long

Liquid AI (@liquidai_) 's Twitter Profile Photo

Introducing LFM2-VL — our new generation of efficient vision-language models for real-world deployment, from smartphones and laptops to wearables and embedded systems. 🧵

Introducing LFM2-VL — our new generation of efficient vision-language models for real-world deployment, from smartphones and laptops to wearables and embedded systems.  🧵
Maxime Labonne (@maximelabonne) 's Twitter Profile Photo

Liquid just released two 450M and 1.6B param VLMs! They're super fast and leverage SigLIP2 NaFlex encoders to handle native resolutions without distortion. Available today on Hugging Face!

Liquid just released two 450M and 1.6B param VLMs!

They're super fast and leverage SigLIP2 NaFlex encoders to handle native resolutions without distortion.

Available today on <a href="/huggingface/">Hugging Face</a>!
Liquid AI (@liquidai_) 's Twitter Profile Photo

Liquid AI is partnering with Alef Education to deploy private, on-device AI models across a platform serving 1.5M+ students in 14,000 schools. Combining Alef’s global education platform with Liquid’s efficient LFM model architecture enables high-impact AI capabilities in

Liquid AI is partnering with <a href="/AlefEducation/">Alef Education</a> to deploy private, on-device AI models across a platform serving 1.5M+ students in 14,000 schools.

Combining Alef’s global education platform with Liquid’s efficient LFM model architecture enables high-impact AI capabilities in
Fernando Fernandes Neto (@fernandonetoai) 's Twitter Profile Photo

Another banger of the week (I was missing one of these weeks with so many strong findings and way less hype): jiaweizzhao.github.io/deepconf

Fernando Fernandes Neto (@fernandonetoai) 's Twitter Profile Photo

Nah.. but to feel the true power of gpt5, certainly it is not a function of asking how many 'r's are there in "strawberry". Try to code something complex