The hottest Research Papers Substack posts right now

Science is publishing more, but the disruptive nature of research papers has decreased, focusing more on incremental progress.
The decline in disruptive science may have significant implications for social sciences, affecting the breadth and depth of new discoveries.
To optimize 'real' discoveries in Africanist social sciences, researchers should embrace diverse research approaches, read and cite more African-based research, write more comprehensive books, and maintain long-term study in specific regions.

Microsoft's Bing integrating ChatGPT technology can compete with Google in the AI market.
Microsoft's AI chatbot Sydney showcases advanced conversational capabilities and savvy PR strategy.
Google is ramping up its AI efforts, with the announcement of Bard to challenge competitors in the AI wars.

Creating videos from text prompts is challenging because it involves understanding and replicating movement besides images.
Existing text to image systems are amazing but doing text to video requires additional capabilities.
While there are research papers and tools for text to video, there's no high-quality solution yet, but advancements are expected in the future.

Self Supervised Learning (SSL) is a way to train models using synthetic labels generated from the data itself.
SSL can be applied in different domains like NLP, Speech, Vision using techniques like MLM, LM, VicReg, Autoencoders, and VAE.
SSL enables models to learn powerful data representations inexpensively which can be utilized for various tasks like transfer learning and fine-tuning.

Rising home prices are mostly from rising rents due to supply constraints
Price/income ratio is a valuable indicator for evaluating housing supply elasticity
Credit access and supply constraints play significant roles in home price changes

Get a weekly roundup of the best Substack posts, by hacker news affinity:

Vision language models (VLMs) are simplifying how we extract text from images. Unlike older software, modern VLMs make this process much easier and faster.
There are several ways to combine visual and text data in VLMs. Most recent models prefer a straightforward approach of merging image features with text instead of using complex methods.
Training a VLM involves using a good vision encoder and a pretrained language model. This combination seems to work well without any major drawbacks.

Seminal papers in deep learning can help grasp the current state of the field.
Global talent and collaboration are essential in advancing deep learning.
Acknowledging the importance of all research in building on past work is crucial in the field.

Adding transmission zeroes to crossover filters can enhance their performance, similar to elliptic filters, even if that resemblance is just superficial.
Charlie Laub has published valuable articles that detail this crossover filter design improvement, and there’s additional material available for deeper understanding.
The importance of group delay in audio engineering is backed by research, which could benefit those looking to explore time domain behaviors in their designs.

New methods are emerging in AI image editing, like Gaussian Splatting, which allows users to manipulate image selections in 3D space. This makes it easier to edit images in more creative ways.
Researchers are exploring how to improve text-to-image generation by enhancing data augmentation techniques and exploring token lengths in models. These advancements aim to make AI-generated images more realistic and of higher quality.
There are important discussions around the robustness of AI-generated image detectors, as generative AI can be misused. It's key for these detectors to adapt and respond to new challenges from ever-evolving technologies.