marcus
@mk2112.bsky.social
neurons, tokens, understanding.
github.com/mk2112
github.com/mk2112
Pinned
marcus
@mk2112.bsky.social
· Jan 11
GitHub - MK2112/nn-zero-to-hero-notes: Jupyter Notebook notes on Andrej Karpathy's tutorial series, "Neural Networks: Zero to Hero."
Jupyter Notebook notes on Andrej Karpathy's tutorial series, "Neural Networks: Zero to Hero." - MK2112/nn-zero-to-hero-notes
github.com
The best ways to learn are by teaching or by breaking something important and urgently having to fix it.
September 27, 2025 at 9:30 PM
The best ways to learn are by teaching or by breaking something important and urgently having to fix it.
Reposted by marcus
Vibe of the day (every day)
January 17, 2025 at 10:00 PM
Vibe of the day (every day)
Reposted by marcus
"After 47 years living here, I decided to move to Orlando to be closer to Disney World."
January 11, 2025 at 9:36 PM
"After 47 years living here, I decided to move to Orlando to be closer to Disney World."
Generalizing arxiv.org/abs/2402.044... further would be so cool... or maybe applying it to robotics...
Amortized Planning with Large-Scale Transformers: A Case Study on Chess
This paper uses chess, a landmark planning problem in AI, to assess transformers' performance on a planning task where memorization is futile $\unicode{x2013}$ even at a large scale. To this end, we r...
arxiv.org
January 11, 2025 at 9:17 PM
Generalizing arxiv.org/abs/2402.044... further would be so cool... or maybe applying it to robotics...
psst, www.deep-ml.com
Deep-MLDeep-ML
Practice Machine Learning with Deep-ML
www.deep-ml.com
January 11, 2025 at 8:58 PM
psst, www.deep-ml.com
Underrated. github.com/swyxio/ai-no...
GitHub - swyxio/ai-notes: notes for software engineers getting up to speed on new AI developments. Serves as datastore for https://latent.space writing, and product brainstorming, but has cleaned up c...
notes for software engineers getting up to speed on new AI developments. Serves as datastore for https://latent.space writing, and product brainstorming, but has cleaned up canonical references und...
github.com
January 11, 2025 at 8:30 PM
Underrated. github.com/swyxio/ai-no...
AI as a service, AaaS.
January 11, 2025 at 8:29 PM
AI as a service, AaaS.
Shipping AI without calling it AI is underrated.
January 11, 2025 at 8:19 PM
Shipping AI without calling it AI is underrated.
Reposted by marcus
softmax is fundamentally unable to sustain robust reasoning behaviours across all possible inputs, as its output coefficients are necessarily dispersing provided sufficient input elements.
Google Deepmind's softmax is not enough (for sharp out-of-distribution)
arxiv.org/abs/2410.01104
Google Deepmind's softmax is not enough (for sharp out-of-distribution)
arxiv.org/abs/2410.01104
October 26, 2024 at 7:34 AM
softmax is fundamentally unable to sustain robust reasoning behaviours across all possible inputs, as its output coefficients are necessarily dispersing provided sufficient input elements.
Google Deepmind's softmax is not enough (for sharp out-of-distribution)
arxiv.org/abs/2410.01104
Google Deepmind's softmax is not enough (for sharp out-of-distribution)
arxiv.org/abs/2410.01104
We destroyed nuance with social media only to have to rebuild it with AI.
October 25, 2024 at 10:28 PM
We destroyed nuance with social media only to have to rebuild it with AI.
The answer to life, the universe and everything is CUDA OUT OF MEMORY
October 24, 2024 at 9:58 PM
The answer to life, the universe and everything is CUDA OUT OF MEMORY
2025 will bring great integration of generally understanding transformers with agentic player behavior.
October 24, 2024 at 9:45 PM
2025 will bring great integration of generally understanding transformers with agentic player behavior.
Your daily reminder that PayPal limits passwords to 20 characters. Not in line with widely recommended security practices, as stated by #PayPal themselves, btw. paypalobjects.com/en_US/vhelp/...
Password Guidelines
paypalobjects.com
October 17, 2024 at 8:08 PM
Your daily reminder that PayPal limits passwords to 20 characters. Not in line with widely recommended security practices, as stated by #PayPal themselves, btw. paypalobjects.com/en_US/vhelp/...
Reposted by marcus
The Nobel Prize in chemistry was awarded to David Baker at the University of Washington and Demis Hassabis and John M. Jumper of Google DeepMind.
Nobel Prize in chemistry awarded to three scientists for work on proteins
The prize was awarded Wednesday for research that cracked the code of proteins.
wapo.st
October 9, 2024 at 12:14 PM
The Nobel Prize in chemistry was awarded to David Baker at the University of Washington and Demis Hassabis and John M. Jumper of Google DeepMind.
Those who will stick to learning instead of GPT-pasting will own the future.
October 17, 2024 at 6:44 PM
Those who will stick to learning instead of GPT-pasting will own the future.
github.com/MK2112/mobileYOLOv3
Full explanations, code, and visualizations to help you learn about YOLO with a back-to-back example.
Full explanations, code, and visualizations to help you learn about YOLO with a back-to-back example.
GitHub - MK2112/mobileYOLOv3: YOLOv3 on a MobileNetV3_Small architecture; trained, explained, pruned and quantized for text detection.
YOLOv3 on a MobileNetV3_Small architecture; trained, explained, pruned and quantized for text detection. - MK2112/mobileYOLOv3
github.com
October 17, 2024 at 6:31 PM
github.com/MK2112/mobileYOLOv3
Full explanations, code, and visualizations to help you learn about YOLO with a back-to-back example.
Full explanations, code, and visualizations to help you learn about YOLO with a back-to-back example.
The intersection of #ML and efficient computing booms along quietly. Building reliable small-scale models with large-scale capabilities is key for more sustainable deployment.
October 17, 2024 at 6:27 PM
The intersection of #ML and efficient computing booms along quietly. Building reliable small-scale models with large-scale capabilities is key for more sustainable deployment.
Have you seen my gradients?
December 27, 2023 at 6:58 PM
Have you seen my gradients?