ENIGMA is a novel multi-subject decoding model that reconstructs seen images from electroencephalography (EEG) recordings. This model achieves state-of-the-art (SOTA) performance while requiring less than 1% of the parameters needed by previous EEG-to-Image approaches.

The goal of ENIGMA is to enable practical brain-computer interfaces (BCIs), addressing limitations such as the need for quick deployment on new subjects, effectiveness on affordable scanning hardware, and small enough size for local execution. It can effectively fine-tune on as little as 15 minutes of data, significantly improving performance across both research-grade and consumer-grade EEG hardware.

The approach integrates a subject-unified spatio-temporal backbone with multi-subject latent alignment layers and an MLP projector to map raw EEG signals to a rich visual latent space. It has been evaluated using extensive image reconstruction metrics and behavioral evaluations by human raters, marking a substantial step towards the development of practical BCI applications.

👉 Read the full article on arXiv

Amazon is reportedly considering the launch of a marketplace where media companies can sell their content directly to AI companies. This move is likely driven by the increasing demand for vast amounts of data necessary for training generative AI models.

Currently, many AI companies gather data through web scraping or licensing agreements. However, this new marketplace could offer a new revenue stream for content owners, allowing media organizations to directly provide their articles, images, and other content to AI companies in exchange for compensation.

The platform is expected to include mechanisms to ensure content quality and licensing transparency. Amazon's entry into this space could reshape data acquisition for AI development and foster a new ecosystem between content creators and AI companies.

👉 Read the full article on TechCrunch

A contradiction exists between the nutrition advice promoted by Realfood.gov, a government website supported by Robert F. Kennedy Jr., and Elon Musk's AI chatbot Grok. Realfood.gov declares, "We are ending the war on protein," emphasizing the prioritization of high-quality protein from both animal and plant sources.

However, when prompted about protein intake, Grok initially suggested a lower recommendation of 0.8 grams per kilogram of body weight per day to a reporter. This recommendation was only adjusted upwards after the reporter specified regular exercise, implying that for many Americans who do not exercise regularly, Grok might suggest a lower protein intake than the administration's stated goals.

The U.S. Dietary Guidelines, released under Health Secretary Kennedy Jr., advocate for increased protein intake (1.2 to 1.6 grams per kilogram of body weight per day) and reduced consumption of ultra-processed foods. The conflicting advice from Grok highlights a potential source of confusion for the public regarding optimal protein consumption.

👉 Read the full article on Wired