Yes; heavy tool use. It seems to have mostly solved it via code (PIL, cv2), but using multimodal intuition to debug. E.g. one attempt within the CoT generates a path that simply traverses the outside of the maze, but it recognizes on its own this is wrong so it refines the code.
MACHINE LEARNING
-
Image preprocessing and BFS maze-solving workflow
By
–
The CoT is meandering but it seems to use PIL/cv2 to get the maze into form where it can solve with BFS, notices the resulting solution is broken (simply traverses the outside of the maze) and refines the code until the output looks sane
-
Meta FAIR Releases Advanced Vision Encoder for Machine Intelligence
By
–
🚀 Meta FAIR is releasing several new research artifacts on our road to advanced machine intelligence (AMI). These latest advancements are transforming our understanding of perception.
— AI at Meta (@AIatMeta) 17 avril 2025
1️⃣ Meta Perception Encoder: A large-scale vision encoder that excels across several image &… pic.twitter.com/cAf79LkBvnMeta FAIR is releasing several new research artifacts on our road to advanced machine intelligence (AMI). These latest advancements are transforming our understanding of perception. Meta Perception Encoder: A large-scale vision encoder that excels across several image &
-
Cost-per-solution vs cost-per-token for AI models
By
–
I think what Aiden’s saying is that cost per solution makes sense but cost per token does not. If a model A gets the right answer as reliably as model B but at lower cost, the number of tokens it generated to achieve that doesn’t matter.
-
OpenAI O3 and O4-mini Achieve Top Scores on Humanity’s Last Exam
By
–
OpenAI O3 et O4-mini explosent les scores dans le Humanity's Last Exam. Pour ceux qui ne connaissent pas : le HLE, c’est un benchmark de 3 000 questions couvrant les mathématiques, les sciences humaines et les sciences naturelles. Honnêtement, c’est la première vraie différence
-
Prediction: GPT-5 development expected within a few months
By
–
I think they will develop GPT-5 in few months.
-
Scaling Vision Pre-Training to 4K Resolution
By
–
Scaling Vision Pre-Training to 4K Resolution
Paper: https://
arxiv.org/abs/2503.19903
Page: https://
nvlabs.github.io/PS3/
Code (Coming Soon): https://
github.com/NVLabs/PS3 -
Berkeley NVIDIA Launch PS3 Vision Encoder for 4K Pretraining
By
–
Berkeley and NVIDIA dropped a major breakthrough: PS3, the first vision encoder to enable efficient visual pretraining at 4K ultra-high resolution—with no extra compute overhead. On top of that, they introduced VILA-HD, a multimodal large model built on PS3.
-
ICCV Review Completed: Computer Vision Research Insights
By
–
iccv review done..
My eyes need to have a rest.. -
Models interpret same prompts in intriguing experiment
By
–
It's intriguing to see how different models can interpret the same prompts. Great experiment! Can't wait to check out those videos.