Ad
Skip to content

Matthias Bastian

Matthias is the co-founder and publisher of THE DECODER, exploring how AI is fundamentally changing the relationship between humans and computers.
Read full article about: OpenAI's Assistants API gets major update with improved file search, vector storage, and more

OpenAI has announced major updates to its Assistants API. The API now includes an improved search tool called file_search, which can ingest up to 10,000 files per Assistant - a 500x increase over the previous version. In addition to file_search, OpenAI introduces vector_store objects that automate file parsing, chunking, and embedding for seamless searches. The update also includes granular control over token usage, support for common model configuration parameters such as temperature and peak P-value, and the ability to use fine-tuned models, though initially limited to gpt-3.5-turbo-0125. The API now supports streaming, and streaming and polling helper functions have been added to the OpenAI Node and Python SDKs. Developers should consult the migration guide to learn how to upgrade their tool usage to the latest version of the Assistants API.

Read full article about: AMD announces new Ryzen CPUs with better AI performance

AMD's latest Ryzen CPUs are designed to deliver better AI performance. The company announced two new families of chips: the Ryzen 8000 Pro series for desktops and the Ryzen Pro 8040 series for mobile devices. Both families feature a Neural Processing Unit (NPU) that AMD says will offer better on-device AI performance for business users, as companies are reluctant to use the cloud for AI due to privacy concerns. The chips will be available in Lenovo and HP PCs starting in the second quarter. AMD promises higher performance and efficiency compared to Intel's latest Meteor Lake chips. The launch is another step in the race for AI PCs. AMD and Intel are increasingly focusing on AI performance rather than traditional benchmarks to demonstrate a chip's capabilities.

AMD's new desktop CPUs are designed to offer more AI performance. | Image: AMD
Comment Source: AMD
Read full article about: Amazon Music launches Maestro, an AI playlist generator

Amazon Music has launched Maestro, an AI playlist generator available in beta to a small number of US customers on iOS and Android. Users enter a short prompt, which can include emojis or activities. Maestro then creates what Amazon says is a unique playlist tailored to the user. Amazon has implemented systems to block offensive language and inappropriate prompts. User feedback will be used to improve Maestro. Amazon Music Unlimited subscribers can listen to and save playlists instantly, while Prime members and ad-supported customers can listen to a 30-second preview of playlists. Access will be expanded to other customers over time. Spotify recently launched AI playlists for its own platform.

Read full article about: Baidu's Ernie chatbot has been on a growth spurt, but is still far behind ChatGPT

Baidu's AI chatbot, Ernie Bot, has attracted more than 200 million users, double the number in December. Ernie Bot's application API is accessed 200 million times a day, the company said, with 85,000 enterprise customers. Ernie Bot was the first Chinese chatbot to go public last September. Globally, however, it still lags far behind ChatGPT, which had 1.86 billion hits in March but has not been growing on the web for months.

Read full article about: ALOHA Unleashed: Google Deepmind shows progress in autonomous, versatile robots

Google Deepmind introduces ALOHA Unleashed, a new generation of AI-powered robots with impressive dexterity. In a series of videos, the robots show how they can hang shirts, insert precise gears, and even tie shoelaces. The robots can generalize to untrained objects such as sweaters. The videos were shot in one go with no editing, and the robots are completely autonomous. With ALOHA Unleashed, Google Deepmind says it wants to push the boundaries of dexterity for low-cost robots. The technology could be used in industry and everyday life in the future.

Video: Google Deepmind