AI Weekly Rundown (October 21 to October 27)
Major AI announcements from Meta, NVIDIA, OpenAI, Google and more this week.
Hello, Engineering Leaders and AI Enthusiasts!
Another eventful week in the AI realm. Lots of big news from huge enterprises.
In today’s edition:
🤖 Meta’s Habitat 3.0 can train AI agents to assist humans in daily tasks🔥 NVIDIA's AI teaches robots complex skills on par with humans
🧪 OpenAI’s secret sauce behind Dall-E 3’s accuracy
💪 Qualcomm's new PC chip for AI to challenge Apple, Intel🔝 Microsoft is outdoing its biggest rival, Google, in AI
🚀 Samsung Galaxy S24 is your upcoming pocket AI machine
😱 OpenAI’s new rival Jina AI has open-source 8k context🔍 LLM hallucination problem will be over with “Woodpecker”
🌟 NVIDIA Research has announced new AI advancements
🛡 OpenAI forms 'Preparedness' team to study advanced AI risks
🔒 Google’s new ventures for safe and secure AI
🐕 Robot dog turns into a talking tour guide with ChatGPT
Let’s go!
Meta’s Habitat 3.0 can train AI agents to assist humans in daily tasks
Meta has announced three major advancements toward the development of socially intelligent AI agents that can cooperate with and assist humans in their daily lives:
Habitat 3.0: The highest-quality simulator that supports robots and humanoid avatars and allows for human-robot collaboration in home-like environments. AI agents trained with Habitat 3.0 learn to find and collaborate with human partners in everyday tasks like cleaning up a house. These AI agents are evaluated with real human partners using a simulated human-in-the-loop evaluation framework (also provided with Habitat 3.0).
Habitat Synthetic Scenes Dataset (HSSD-200): An artist-authored 3D scene dataset that more closely mirrors physical scenes. It comprises 211 high-quality 3D scenes and a diverse set of 18,656 models of physical-world objects from 466 semantic categories.
HomeRobot: An affordable home robot assistant hardware and software platform in which the robot can perform open vocabulary tasks in simulated and physical-world environments.
NVIDIA's AI teaches robots complex skills on par with humans
A new AI agent developed by NVIDIA Research that can teach robots complex skills has trained a robotic hand to perform rapid pen-spinning tricks– for the first time as well as a human can.
The above are some of nearly 30 tasks that robots have learned to expertly accomplish thanks to Eureka, which uses LLMs to automatically generate reward algorithms to train robots. Eureka is powered by the GPT-4. Eureka-generated reward programs outperform expert human-written ones on more than 80% of tasks.
OpenAI’s secret sauce of Dall-E 3’s accuracy
OpenAI published a paper on DALL-E 3, explaining why the new AI image generator follows prompts much more accurately than comparable systems.
Prior to the actual training of DALL-E 3, OpenAI trained its own AI image labeler, which was then used to relabel the image dataset for training the actual DALL-E 3 image system. During the relabeling process, OpenAI paid particular attention to detailed descriptions.
Qualcomm's new PC chip for AI to challenge Apple, Intel
Qualcomm has unveiled a new laptop processor designed to outperform rival products from Intel Corp. and Apple Inc. Snapdragon X features 12 high-performance cores capable of crunching data at 3.8 megahertz.
The chip is as much as twice as fast as a similar 12-core processor from Intel while using 68% less power. Qualcomm also claims it can operate at peak speeds 50% higher than Apple's M2 SoC
In addition to overall improved performance, the new processor boasts features explicitly designed for AI. The chipmaker contends that AI’s full potential will be realized when it extends beyond data centers and into end-user devices such as smartphones and PCs.
Microsoft is outdoing its biggest rival, Google, in AI
Samsung is going all in with AI on its next flagship. It wants to make the Galaxy S24, Galaxy S24+, and Galaxy S24 Ultra the smartest AI phones ever. The series will have features lifted straight from ChatGPT and Google Bard, such as the ability to create content and stories based on a few keywords provided by the user.
There will also be features Samsung has designed on its own, such as text-to-image Generative AI, and a lot of them will be available both online and offline. Speech-to-text functionality is one area that will see improvements.
Samsung Galaxy S24 is your upcoming pocket AI machine
SAP announced new business AI and user experience innovations in its comprehensive spend management and business network solutions to help customers control costs, mitigate risk, and increase productivity.
SAP will also embed Joule, its new generative AI copilot, throughout its cloud solutions, with availability in its spend management software planned for 2024. It has also unveiled SAP Spend Control Tower, which offers advanced AI features and the ability to see across all SAP spend solutions.
All these new AI innovations are being developed with security, privacy, compliance, ethics, and accuracy in mind.
OpenAI’s new rival Jina AI has open-source 8k context
Berlin-based AI company Jina AI has launched Jina-embeddings-v2, the world's first open-source 8K text embedding model. This model supports an impressive 8K context length, putting it on par with OpenAI's proprietary model. Jina-embeddings-v2 offers extended context potential, allowing for applications such as legal document analysis, medical research, literary analysis, financial forecasting, and conversational AI.
Benchmarking shows that it outperforms other leading base embedding models. The model is available in two sizes, a base model for heavy-duty tasks and a small model for lightweight applications. Jina AI plans to publish an academic paper, develop an embedding API platform, and expand into multilingual embeddings.
Enjoying the weekly updates?
Refer your pals to subscribe to our newsletter and get exclusive access to 400+ game-changing AI tools.
When you use the referral link above or the “Share” button on any post, you'll get the credit for any new subscribers. All you need to do is send the link via text or email or share it on social media with friends.
LLM hallucination problem will be over with “Woodpecker”
Researchers from the University of Science and Technology of China and Tencent YouTu Lab have developed a framework called "Woodpecker" to correct hallucinations in multimodal large language models (MLLMs).
Woodpecker uses a training-free method to identify and correct hallucinations in the generated text. The framework goes through five stages, including key concept extraction, question formulation, visual knowledge validation, visual claim generation, and hallucination correction.
The researchers have released the source code and an interactive demo of Woodpecker for further exploration and application. The framework has shown promising results in boosting accuracy and addressing the problem of hallucinations in AI-generated text.
NVIDIA Research has announced new AI advancements
NVIDIA Research has announced new AI advancements that will be presented at the NeurIPS conference. The projects include new techniques for transforming text-to-images, photos to 3D avatars, and specialized robots into multi-talented machines.
The research focuses on generative AI models, reinforcement learning, robotics, and applications in the natural sciences. Highlights include improving text-to-image diffusion models, advancements in AI avatars, breakthroughs in reinforcement learning and robotics, and AI-accelerated physics, climate, and healthcare research. These advancements aim to accelerate the development of virtual worlds, simulations, and autonomous machines.
OpenAI forms 'Preparedness' team to study advanced AI risks
To minimize risks from frontier AI as models continue to improve, OpenAI is building a new team called Preparedness. It tightly connect capability assessment, evaluations, and internal red teaming for frontier models, from the models OpenAI develops in the near future to those with AGI-level capabilities.
The team will help track, evaluate, forecast, and protect against catastrophic risks spanning multiple categories including:
Individualized persuasion
Cybersecurity
Chemical, biological, radiological, and nuclear (CBRN) threats
Autonomous replication and adaptation (ARA)
The Preparedness team mission also includes developing and maintaining a Risk-Informed Development Policy (RDP). In addition, OpenAI is soliciting ideas for risk studies from the community, with a $25,000 prize and a job at Preparedness on the line for top ten submissions.
Google’s new ventures for safer, more secure AI
Google has announced a bug bounty program for attack scenarios specific to generative AI through expanding its Vulnerability Rewards Program (VRP) for AI. It shared guidelines for security researches to see what’s “in scope”.
To further protect against machine learning supply chain attacks, Google is expanding its open source security work and building upon prior collaboration with the Open Source Security Foundation. It has earlier released Secure AI Framework (SAIF) that emphasized AI ecosystems must have strong security foundations.
Google is also to support a new effort by the non-profit MLCommons Association to develop standard AI safety benchmarks. The effort aims to bring together expert researchers across academia and industry to develop standard benchmarks for measuring the safety of AI systems into scores that everyone can understand.
Robot Dog turns into a talking tour guide with ChatGPT
Named Spot, the four-legged robot could run, jump, and even dance. To make Spot “talk,” Boston Dynamics used OpenAI’s ChatGPT API, along with some open-source LLMs to carefully train its responses. With ChatGPT, it can answer questions and generate responses about the company’s facilities while giving a tour.
It also outfitted the bot with a speaker, added text-to-speech capabilities, and made its mouth mimic speech “like the mouth of a puppet”.
That's all for now!
Subscribe to The AI Edge and gain exclusive access to content enjoyed by professionals from Moody’s, Vonage, Voya, WEHI, Cox, INSEAD, and other esteemed organizations.
Thanks for reading, and see you on Monday. 😊