1-20b doesn't benefit almost any normal person situation. Whereas an equivalent 4b dense is usable almost anywhere. Qwen3 Coder doesn't even have any dense option.
RESEARCH
-
MoE Models Memory Inefficiency Dense Models Practicality
By
–
It's terrible for most regular people. We are memory poor — MoE models make this even worse. It may be cute, but it's totally impractical. Dense models are far more usable in practice.
-
Tunable AI systems as genuine science fiction premise beyond anthropomorphism
By
–
Also, as soon as you say "tunable" you've stepped outside the strictly-anthropomorphic-AI fics I considered boring. "Tunable" would be an actual SF premise. Even "clonable", really, though somehow the lesser story-writers didn't even consider that.
-

Diabetes Platelet Hyperactivity Mechanism SEC61B Calcium Balance
By
–
Discovery of a mechanism for the increased risk of arterial blood clots in people with diabetes: platelet hyperactivity mediated by SEC61B, calcium balance inside these cells @jclinicalinvest https://
jci.org/articles/view/
184597
… -
Humans and Machines Sharing Best Practices for Optimization
By
–
In the world of humachinekind, humans and machines are continuously sharing their best practices, enabling them to optimize their existence for any conditions.
-
OpenAI Podcast Now Available on Spotify Apple YouTube
By
–
Listen to the OpenAI Podcast on—
Spotify https://
open.spotify.com/show/0zojMEDiz
KMh3aTxnGLENP
…
Apple https://
podcasts.apple.com/us/podcast/ope
nai-podcast/id1820330260
…
YouTube -
OpenAI Podcast Episode 5: Journey to AGI Breakthroughs
By
–
What’s next on the path to AGI?
— OpenAI (@OpenAI) 15 août 2025
In Episode 5 of the OpenAI Podcast, Chief Scientist @merettm and Technical Fellow @sidorszymon join host @AndrewMayne to explore the journey to AGI so far — from today’s breakthroughs to the possibilities ahead. pic.twitter.com/OxLrAxiI7eWhat’s next on the path to AGI? In Episode 5 of the OpenAI Podcast, Chief Scientist @merettm and Technical Fellow @sidorszymon join host @AndrewMayne to explore the journey to AGI so far — from today’s breakthroughs to the possibilities ahead.
-

Recovering Model Weights Exactly Post-LoRA: New Discovery
By
–
just learned that under certain conditions you actually CAN recover model weights *exactly*, post-LORA pretty crazy
-
New Vulnerability Enables Pre-Fine-Tuning Weight Recovery Attack
By
–
— dr. jack morris (@jxmnop) 15 août 2025
We uncover a new vulnerability- Pre-Fine-Tuning Weight Recovery
With a few LoRA fine-tuned models we recover the pre-fine-tuning weightsof SoTA models, undoing Stable Diffusion personalization training and Mistral alignment
Project: http://
vision.huji.ac.il/spectral_detun
ing
… -
Working on Qwen: Extraction technique and model development
By
–
that's a good point; probably should have said 'we extracted one' not 'it'? working on qwen tonight