AI Weekly Rundown (September 29 to October 6)
Major AI announcements from Apple, Google, OpenAI, Meta, and more.
Hello, Engineering Leaders and AI Enthusiasts!
Another eventful week in the AI realm. Lots of big news from huge enterprises.
In today’s edition:
🍎 Apple’s Pursuit of ChatGPT-like AI🧨 Humane’s first AI device creating buzz
🤯 LLM Lie Detector catching AI lies
🤖 Stability AI launches LLM for portable smart devices
🚀 Rewind launches AI Pendant, your always-on AI assistant
💥 StreamingLLM for infinite text input with unwavering precision
⚔️ Zoom's Inexpensive AI Rivals Microsoft & Google📈 Google DeepMind Scaling Up Robotics Learning
🎓 OpenAI Unlocking New Careers in AI/ML
📱 Google announces next-level AI features for its Pixel 8 series
📜 DeepMind's Promptbreeder automates prompt engineering
🎉 Canva bolsters its AI toolkit by partnering with Runway
🧠 Meta AI decoding brain speech with 73% accuracy🚗 Wayve's GAIA-1 9B enhancing autonomous vehicle training
🔌 OpenAI considers in-house AI chips to reduce nvidia dependency
Let’s go!
Apple’s pursuit of ChatGPT-like AI and more
Apple’s CEO Tim Cook has confirmed that the company is working on ChatGPT like AI chatbot and expects to hire more AI staff in the UK. Cook mentioned that AI is already integrated into Apple products, such as the Apple Watch's Fall Detection and Crash Detection features.
Also, Apple is planning to upgrade its search engine in the App Store and potentially develop a Google competitor "Pegasus". It’s being integrated into iOS and macOS, with the possibility of using gen AI tools to enhance it further. Apple's Spotlight search feature already allows users to search for web results, app details, and documents.
Humane’s first AI device creating buzz
Humane Inc. has unveiled its first AI device ‘Humane Ai Pin’, at Coperni's Paris fashion show. The device was worn by supermodel Naomi Campbell. It is a screenless, standalone wearable with sensors for natural and intuitive interactions.
It does not need to be paired with a smartphone and features AI-powered optical recognition and a laser-projected display. The full capabilities of the Humane Ai Pin will be revealed on November 9.
LLM Lie Detector catching AI lies
This paper discusses how large language models can "lie" by outputting false statements even when they know the truth. The authors propose a simple lie detector that does not require access to the LLM's internal workings or knowledge of the truth. The detector works by asking unrelated follow-up questions after a suspected lie and using the LLM's yes/no answers to train a logistic regression classifier.
The lie detector is highly accurate and can generalize to different LLM architectures, fine-tuned LLMs, sycophantic lies, and real-life scenarios. This suggests that LLMs have consistent lie-related behavioral patterns that can be detected.
Stability AI launches LLM for portable smart devices
Stability AI has launched an experimental version of Stable LM 3B, its latest in the suite of high-performance generative AI solutions. At 3 billion parameters (vs. the 7 to 70 billion parameters typically used by the industry), Stable LM 3B is a compact language model designed to operate on portable digital devices like handhelds and laptops.
Its key is its smaller size and efficiency. But despite its size, it is highly performant– it outperforms the previous state-of-the-art 3B parameter language models and even some of the best open-source language models at the 7B parameter scale.
Rewind launches AI Pendant, your always-on AI assistant
Rewind Pendant is a wearable necklace that captures what you say and hear in the real world and then transcribes, encrypts, and stores it entirely locally on your phone. It is like a wearable AI assistant who you can then ask any question using AI.
No more forgetting what your spouse asked you to pick up at the grocery store, eh?😉
StreamingLLM for infinite text input with unwavering precision
Deploying LLMs in streaming applications is urgently needed but comes with challenges due to efficiency limitations and reduced performance with longer texts. Window attention provides a partial solution, but its performance plummets when initial tokens are excluded.
Recognizing the role of these tokens as “attention sinks", new research by Meta AI (and others) has introduced StreamingLLM– a simple and efficient framework that enables LLMs to handle unlimited texts without fine-tuning. By adding attention sinks with recent tokens, it can efficiently model texts of up to 4M tokens. It further shows that pre-training models with a dedicated sink token can improve the streaming performance.
Here’s an illustration of StreamingLLM vs. existing methods. It firstly decouples the LLM’s pre-training window size and its actual text generation length, paving the way for the streaming deployment of LLMs.
Zoom's Inexpensive AI Rivals Microsoft & Google
Zoom has announced the launch of Zoom Docs, a collaboration-focused "modular workspace" with built-in AI collaboration features. The platform integrates Zoom's AI Companion, which can generate new content or populate documents from other sources.
Users can ask the AI Companion to summarize meetings, chats, and information, and the platform supports inter-document linking and embedding. Zoom AI Companion is included in the price of paid subscription plans.
Google DeepMind Scaling Up Robotics Learning
Researchers of Google Deepmind have created a dataset called Open X-Embodiment, which combines data from 22 different types of robots. They also developed a robotics transformer model called RT-1-X, trained on this dataset, to transfer skills across various robot types.
Testing showed that the RT-1-X model performed significantly better than models trained on data from individual robot types. Additionally, training a visual language action model on data from multiple embodiments tripled its performance. The Open X-Embodiment dataset and RT-1-X model are now available to the research community, with the aim of advancing cross-embodiment research and transforming the way robots are trained.
OpenAI Unlocking New Careers in AI/ML
OpenAI has launched “OpenAI Residency”, a six-month program that helps exceptional researchers and engineers from different fields gain the necessary skills and knowledge to transition into the AI and ML space.
It is ideal for researchers specializing in fields like mathematics, physics, or neuroscience, as well as talented software engineers looking to work in AI research. Residents work on real AI problems with OpenAI's Research teams and receive a full salary during the program.
📢 Invite friends and get rewards 🤑🎁
Enjoying AI updates? Refer friends and get perks and special access to The AI Edge.
Get 400+ AI Tools and 500+ Prompts for 1 referral.
Get A Free Shoutout! for 3 referrals.
Get The Ultimate Gen AI Handbook for 5 referrals.
When you use the referral link above or the “Share” button on any post, you'll get credit for any new subscribers. Simply send the link in a text, email or share it on social media with friends.
Google announces next-level AI features for its Pixel 8 series
At its Made by Google event, the tech giant announced its latest Pixel 8 and Pixel 8 Pro phones built with AI at the center for a more helpful and personal experience. Here are some impressive AI features included:
Best Take uses the photos you take to get the photo you thought you took. An on-device algorithm creates a blended image from a series of photos to get everyone’s best look.
Magic Editor in Google Photos uses generative AI to let you reposition and resize subjects or use presets to make the background pop– all with just a few taps.
Audio Magic Eraser lets you easily reduce distracting sounds in your video, like howling winds or noisy crowds. This first-of-its-kind computational audio capability uses advanced ML models to sort sounds into distinct layers.
Zoom Enhance uses generative AI to take corrective steps to improve photo quality and gaps between pixels when you crop an image.
Call Screen will be updated with clever new features that will allow users to better determine the calls they want to skip and those they want to take and more using AI.
Upgraded Gboard will generate higher-quality reply suggestions with better overall conversational awareness powered by an LLM running on the Pixel 8 Pro.
In addition, Pixel 8 Pro will be the first hardware to run Google’s generative AI models on-device. Its custom-built Tensor G3 chip can run “distilled” versions of Google’s text- and image-generating models to power a range of applications, like image editing.
DeepMind's Promptbreeder automates prompt engineering
Google DeepMind researchers have introduced Promptbreeder, a self-referential self-improvement method that employs LLMs like GPT-3 to iteratively improve text prompts. But, it also improves the way it is improving prompts.
Here’s how it works (a simple overview): Promptbreeder initializes a population of prompt variations for a task and tests them to see which performs best. The winners are "mutated" (modified in some way) and inserted back into the population. Rinse and repeat. But it makes mutations smarter over time. It uses AI to generate "mutation prompts" (instructions for how to mutate and improve a prompt).
The results: Prompts that are specialized and highly optimized for specific applications. On math, logic, and language tasks, Promptbreeder outperforms other SoTA prompting techniques.
Canva bolsters its AI toolkit by partnering with Runway
Canva is celebrating its 10th anniversary with the Magic Studio update, one of its biggest product launches ever but this time with AI.
It includes a new generative video feature called Magic Media through a partnership with Runway ML. It will generate up to 18 seconds of video based on the user's input in text or still image. In the case of a still image input, the image will be used as the basis of the video and motion and camera movement applied.
Meta AI decoding brain speech ~ 73% accuracy
Meta Researchers have developed a model that can decode speech from non-invasive brain recordings with a high level of accuracy. The model was trained using contrastive learning and could identify speech segments from magneto-encephalography signals. The model's performance allows for the decoding of words and phrases that were not included in the training set.
The study highlights the importance of contrastive objectives, pretrained representations of speech, and a common convolutional architecture for achieving accurate speech decoding from brain activity. This research offers a promising approach to decoding language from brain recordings without the need for invasive procedures.
Wayve's GAIA-1 9B enhancing autonomous vehicle training
British startup Wayve announces the release of GAIA1, A 9B parameter world model trained on 4,700 hours of driving data. This model is for autonomous driving that uses text, image, video, and action data to create synthetic videos of various traffic situations for training purposes.
It is 480 times larger than the previous version and offers incredible results. And designed to understand and decode key driving concepts, providing fine-grained control of vehicle behavior and scene characteristics to improve autonomous driving systems.
OpenAI considers in-house AI chips to reduce Nvidia dependency
OpenAI is considering developing its own AI chips and has even evaluated a potential acquisition target, sources say. While no final decision has been made, recent discussions within the company have centered on addressing the scarcity of high-cost AI chips integral to OpenAI's operations.
That's all for now!
If you are new to ‘The AI Edge’ newsletter. Subscribe to receive the ‘Ultimate AI tools and ChatGPT Prompt guide’ specifically designed for Engineering Leaders and AI Enthusiasts.
Thanks for reading, and see you on Monday. 😊