yoquan
Re:The recurring dream of replacing developers
on Sunday January 18, 2026 @09:04AM (#46667879)
Yes. On an analogy level you can also examine whether you are in a dream or in reality by looking exactly to such details :)
Open data 3D point cloud of Tokyo
Don't Sleep on Single-Agent Systems
What Succeeding at AI Safety Will Involve
Achieving Energetic Superiority Through System-Level Quantum Circuit Simulation
Re:Start Presentations on the Second Slide
on Saturday June 15, 2024 @04:28AM (#40688329)
Just to share a perspective. Doing business in Japan often requires a certain level of trust. Showing that you (the presenter) have something to count on is thus important.
There is a new tunnel under Tokyo
Re:Mojo Presented at LLVM 2023
on Sunday October 29, 2023 @11:32AM (#38059523)
The slides describing the "whys" of Mojo and its early vindication did a good job of clearing suspicion around its marketing hype, imo.
Mojo Presented at LLVM 2023
Llama2 in Mojo is 15-20% faster than llama2.c
Scientists use quantum device to slow chemical process by factor of 100B
Pgvector 0.4.0 Performance (Vs Qdrant)
OpenLLaMA 7B model trained on RedPajama dataset
Re:A brief history of LLaMA models
on Sunday April 30, 2023 @05:51AM (#35761081)
Because it's fun :-) And effort to bring up a could-be-commercial version is on going.
Information is the frontier for the study of life
Dr Chad Rigetti to Resign
China unlikely to become a dominant superpower
Bank of Japan Can Stomach Weak Yen
Brain study suggests the workday needs to be radically restructured
Mind Wanders Because Your Brain Whispers
Re:Stablecoin Mechanics 2: We found the Tether-Celsius loans
on Monday October 17, 2022 @11:29PM (#33242638)
For a deeper look into stablecoin mechanism and performance of them during May 2022 event, I would recommend [1] which I think is much more interesting for curious readers.
[1] https://www.fdic.gov/analysis/cfr/bank-research-conference/a...
Easy-to-use image segmentation library with pre-trained model zoo
Re:Understanding HTML with Large Language Models
on Wednesday October 12, 2022 @05:49AM (#33174772)
The model used by this research is T5 which is open sourced already. So I think once the dataset is released, we'll see the open version of pre-trained model very soon.
Bert-Large: Prune Once for DistilBERT Inference Performance
Re:OCRmyPDF: Add an OCR text layer to scanned PDF file
on Friday July 08, 2022 @10:53PM (#32031045)
DocTR: https://github.com/mindee/doctr
It also has TensorFlow.js version to run in-browser: https://blog.tensorflow.org/2022/06/ocr-in-browser-using-ten...
Obviously it will relate to complexity in deep learning but I can't resist thinking about some AI models involving quantum computation stuffs :-)
Reddit AMA from a researcher leaving Google AI after 3 years
Re:DeepNet: Scaling Transformers to 1k Layers
on Thursday March 03, 2022 @07:43AM (#30540163)
Actually no. Each layer requires output from previous one, which means sequentially computation. While wider layers can utilize GPU parallel computation better. This is kind of trade-off between less memory (less parameters) vs longer time.