NVIDIA researchers did it again!
They proved back-propagation isn't the only way to build an AI.
For 40 years, every deep learning model in existence has been trained the exact same way.
Back-propagation.
It is the absolute bedrock of modern artificial intelligence. But it comes with a crippling bottleneck: massive memory costs and a rigid, sequential chain of calculations that makes multi-GPU scaling a nightmare.
Now, a paper co-authored by NVIDIA researchers has shattered that dogma.
They introduced a method called EGGROLL (Evolution Guided General Optimization via Low-rank Learning).
Instead of using gradients to look backward through every layer, it uses evolutionary strategies paired with low-rank matrix structures.
Think about what that means.
It bypasses the backward pass entirely.
The results completely rewrite what is possible in machine learning:
• A 100fold increase in training speed for billion-parameter models at scale.
• Achieves up to 91% of the throughput of pure batch inference.
• Enables the stable training of models operating purely in int8 datatypes.
• Competes directly with state-of-the-art reinforcement learning on reasoning tasks.
For years, people assumed that scaling foundation models meant doubling down on massive, gradient-based compute clusters.
This paper proves there is an entirely parallel path.
We’ve spent four decades treating back-propagation as the only law of physics in neural networks.
NVIDIA just rewrote the rules.
Researchers proved every major LLM is secretly obsessed with Japan.
And they finally figured out why.
For years, we’ve been told that AI is entirely Western-centric, that it just reflects Silicon Valley and American values.
A landmark paper by Cardiff and Basque researchers tested 31,680 cultural prompts across 24 languages on frontier models like ChatGPT, Claude, and Gemini.
The results shattered that assumption.
In six out of eight frontier models, Japan was the single most frequently referenced country when asked open-ended cultural questions.
Ask about traditional dances, festivals, or everyday practices in an open context, and the AI defaults to Japan.
Over and over again.
Here is the twist nobody expected.
This bias doesn't come from raw pre-training internet data.
The researchers tracked where the obsession forms. It emerges after pre-training, during the supervised fine-tuning and alignment phase when humans teach the AI how to behave.
Why Japan?
Because decades of global soft power, rich cultural export, and clean, universally admired digital archives make Japanese culture uniquely "safe" for AI safety filters to lean on.
When labs train models to be harmless and universally pleasing, the AI defaults to the cultural equivalent of comfort food.
It avoids controversy by talking about anime, sushi, and tradition.
要約可能なものの無価値性について。きょわいのは、サイトの価値に留まらず、少なくとも市場においてという意味での人間の(以下検閲により削除)/The Great Blogging Collapse: What Happened to 100 Successful Blogs? [Study]
https://t.co/AyrMlGbc6D