AI Weekly Rundown (November 04 to November 10)
Major AI announcements from OpenAI, Samsung, NVIDIA, Amazon, GitHub, 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:
🏋️♀️ Elon Musk’s xAI launches Grok, its ChatGPT rival🙊 OpenAI's huge ChatGPT updates leak ahead of DevDay
🤖 RunwayML introduces the first AI physical device for video
⭐️ OpenAI kicking off big AI announcements (DevDay highlights)
🚀 xAI launches PromptIDE to accelerate prompt engineering🔥 Amazon is developing a model to rival OpenAI
🤖 MySpace co-founder DeWolfe unveils latest text-to-video AI
📱 Samsung to rival ChatGPT with 3 new AI models🔒 GitHub launches AI features to enhance security
💻 NVIDIA’s EOS Supercomputer now trains 175B parameter AI in 4 mins🚀 Humane officially launches the AI Pin
🤝 OpenAI to partner with organizations for new AI training data
🖼️ Adobe creates 3D models from 2D images ‘within 5 seconds’
Let’s go!
Elon Musk’s xAI launches Grok, its ChatGPT rival
Elon Musk just unveiled his ChatGPT competitor, Grok, an AI modeled after the Hitchhiker’s Guide to the Galaxy. Here’s what sets it apart:
Intended to answer almost anything and, far harder, even suggest what questions to ask!
Designed to answer questions with a bit of wit and has a rebellious streak
Real-time knowledge of the world via the 𝕏 platform. So, it will also answer spicy questions that are rejected by most other AI systems.
Grok is still a very early beta product, made with only 2 months of training, and it will improve rapidly in the coming weeks. Its powering engine, Grok-1, displayed strong results, with 63.2% on the HumanEval coding task and 73% on MMLU.
OpenAI's huge ChatGPT updates leak ahead of DevDay
Leaks ahead of OpenAI DevDay, its developer conference tomorrow, suggest big updates are coming to ChatGPT. These include a new interface for ChatGPT and completely new features.
Custom chatbots via Gizmo: The tool will specialize in creating, managing, and selecting custom chatbots. There will also be a Magic Creator or Magic Maker to help you create the chatbots.
Workspace and team plan: OpenAI may also be introducing new enterprise subscriptions and a “Team Plan”. With different pricing plans, it will give abilities to update Workspace metadata such as role, department, etc., external systems integrations (CRM, ERP), internally shareable chat templates, and more.
"context connectors" to Google Drive and Microsoft 365: These will allow you to connect apps to access their information in ChatGPT.
Some other rumors include GPT-4 API price reduction, speed improvement, token 32K general access, and more.
RunwayML introduces the first AI physical device for video
RunwayML is introducing the 1stAI Machine, the first physical device for video editing generated by AI.
It is anticipated to match the quality of videos with that of photos. “At that point, anyone will be able to create movies without the need for a camera, lights, or actors; they will simply interact with the AIs. A tool like 1stAI Machine anticipates that moment by exploring tangible interfaces that enhance creativity.”
OpenAI kicking off big AI announcements (DevDay highlights)
OpenAI held its first developer event yesterday (11/06/2023), which was action-packed. The company launched improved models, new APIs, and much more. Here is a summary of all announcements:
1. Announced a new GPT Builder: GPT Builder will allow anyone to customize and share their own AI assistants with natural language; no coding is required. That combines instructions, extra knowledge, and any combination of skills and then shares that creation with others. Plus and Enterprise users can start creating GPTs this week.
2. GPT-4 Turbo with 128K context at 3x cheaper price: GPT4 can now read a prompt as long as an entire book. It has knowledge of world events up to April 2023. GPT-4 Turbo performs better than our previous models on tasks that require carefully following instructions, such as generating specific formats (e.g., “always respond in XML”). This beta feature is useful for use cases such as replaying requests for debugging, writing more comprehensive unit tests, etc.
3. GPT Store for user-created AI bots: OpenAI’s GPT Store lets you build (and monetize) your own GPT. OpenAI plans to launch a marketplace called the GPT Store, where users can publish their GPTs and potentially earn money. The company aims to empower people with the tools to create amazing things and give them agency in programming AI with language.
4. Launches Assistants API that lets devs build ‘assistants’ into their apps: Developers can build their own “agent-like experiences.” The API enables developers to create assistants with specific instructions, access external knowledge, and utilize OpenAI's generative AI models and tools. Use cases for the Assistants API include natural language-based data analysis, coding assistance, and AI-powered vacation planning.
5. OpenAI launches text-to-image model, DALL-E 3 API: It is now available through API with in-built moderation tools. Open AI has priced the model at $0.04 per generated image.
The API includes built-in moderation to prevent misuse and offers different format and quality options. However, it is currently limited compared to the DALL-E 2 API, as it cannot create edited versions or variations of existing images.
6. A new text-to-speech API called Audio API with 6 preset voices and two generative AI model variants. Alloy, Echo, Fable, Onyx, Nova, and Shimer. The company does not offer control over the emotional effect of the generated audio.
7. Announced a new program called Copyright Shield: Promising to protect businesses using the AI company’s products from copyright claims. They said they will pay the costs incurred if you face legal claims around copyright infringement while building with tools.
xAI launches PromptIDE to accelerate prompt engineering
Right after announcing Grok, xAI launched xAI PromptIDE. It is an integrated development environment for prompt engineering and interpretability research.
At the heart of the PromptIDE is a code editor and a Python SDK. The SDK provides a new programming paradigm that allows implementing complex prompting techniques elegantly. You also gain transparent insights into the model's inner workings with rich analytics that visualize the network's outputs.
PromptIDE was originally created to accelerate development of Grok and give transparent access to Grok-1 (the model that powers Grok) to engineers and researchers in the community. It has helped xAI iterate quickly over different prompts and prompting techniques. Its features empower you to deeply understand Grok-1's outputs.
IDE is currently available to members of the Grok early access program.
Amazon is developing a model to rival OpenAI
Amazon is investing millions in training an ambitious LLM, hoping it could rival top models from OpenAI and Alphabet. The model, codenamed “Olympus”, has 2 trillion parameters, making it one of the largest models being trained. (OpenAI's GPT-4 is reported to have one trillion parameters.)
According to sources, the head scientist of artificial general intelligence (AGI) at Amazon, Prasad, brought in researchers who had been working on Alexa AI and the Amazon science team to work on training models, uniting AI efforts across the company with dedicated resources. However, there is no specific timeline for releasing the new model.
MySpace co-founder DeWolfe unveils latest text-to-video AI
Chris DeWolfe unveiled his latest social-media product, which uses AI to turn text into videos. PlaiDay creates three-second clips for free after a few prompts. Typing in “1970s male disco dancer,” for example, generates a prancing animated video.
But here is the notable feature– add your photo, and the dancer looks like you. It uses your selfies to personalize the video, which you can then share with friends and followers. The video duration will expand in the future, and the company is also working on adding an audio capability.
One example the company showed using the prompt “English Bobby, 1800s style, streets of London, close-up, life-like.” is below.
The personalized video is a little wonky since the user’s selfie doesn’t show them with a mustache.
Samsung to rival ChatGPT with 3 new AI models
Samsung has introduced its own generative AI model called Samsung Gauss at Samsung AI Forum 2023. Which consists of three tools:
Samsung Gauss Language: It’s an LLM that can understand human language and perform tasks like writing emails and translating languages.
Samsung Gauss Code: It focuses on development code and aims to help developers write code quickly. It works with its code assistant called code.i.
Samsung Gauss Image: It’s image generation and editing feature. For example, it could be used to convert a low-resolution image into a high-resolution one.
The company plans to incorporate these tools into its devices in the future. Samsung aims to release the Galaxy S24 based on its Generative AI model in 2024.
Samsung has also introduced "Galaxy AI," a comprehensive mobile AI experience that will transform the everyday mobile experience with enhanced security and privacy. One of the upcoming features is "AI Live Translate Call," which will allow real-time translation of phone calls. The translations will appear as audio and text on the device itself. Samsung's Galaxy AI is expected to be included in the Galaxy S24 lineup of smartphones, set to launch in 2024.
GitHub launches AI features to enhance security
GitHub Advanced Security has introduced AI-powered features to enhance application security testing. Code scanning now includes an autofix capability that provides AI-generated fixes for vulnerabilities in CodeQL, JavaScript, and TypeScript alerts, allowing developers to quickly understand and remediate issues.
Secret scanning leverages AI to detect leaked passwords with lower false positives, while a regular expression generator helps users create custom patterns for secret detection.
Additionally, the new security overview dashboard provides security managers and administrators with historical trend analysis for security alerts.
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.
NVIDIA’s EOS Supercomputer now trains 175B parameter AI in 4 mins
NVIDIA's supercomputer, Eos, can now train a 175 billion-parameter AI model in under 4 minutes, breaking the company's previous speed record by 3 times. And 3.7 trillion tokens in just 8 days. The benchmark also demonstrates Nvidia's ability to build powerful and scalable systems, with Eos achieving a 2.8x performance scaling and 93% efficiency.
The system utilizes over 10,000 GPUs to achieve this feat, allowing for faster training of models. Also, Nvidia's H100 GPU continues to lead in performance and versatility in the MLPerf 3.1 benchmark.
(Source)
Humane officially launches the AI Pin
After months of demos and hints about what the AI-powered future of gadgets might look like, Humane finally took the wraps off of its first device: the AI Pin. Here’s a tldr;
It is a $699 wearable in two parts– a square device and a battery pack that magnetically attaches to your clothes or other surfaces.
$24 monthly fee for a Humane subscription, which gets you a phone number and data coverage through T-Mobile’s network.
You control it with a combination of voice control, a camera, gestures, and a small built-in projector.
More in this video👇
The Pin’s primary job is to connect to AI models through software the company calls AI Mic. Humane’s press release mentions both Microsoft and OpenAI, and previous reports suggested that the Pin was primarily powered by GPT-4– Humane says that ChatGPT access is actually one of the device’s core features.
The device will ship in early 2024, and preorders begin November 16th.
OpenAI to partner with organizations for new AI training data
OpenAI is introducing OpenAI Data Partnerships, where it will work together with organizations to produce public and private datasets for training AI models.
Here’s the kind of data it is seeking:
Large-scale datasets that reflect human society and that are not already easily accessible online to the public today
Any modality, including text, images, audio, or video
Data that expresses human intention (e.g. conversations), across any language, topic, and format
It will also use its next-generation in-house AI technology to help organizations digitize and structure data.
Also, it is not seeking datasets with sensitive or personal information or information that belongs to a third party. But it can help organizations remove it if needed.
Adobe creates 3D models from 2D images ‘within 5 seconds’
A team of researchers from Adobe Research and the Australian National University have developed a groundbreaking AI model that can transform a single 2D image into a high-quality 3D model in just 5 seconds.
Detailed in their research paper LRM: Large Reconstruction Model for Single Image to 3D, it could revolutionize industries such as gaming, animation, industrial design, augmented reality (AR), and virtual reality (VR).
LRM can reconstruct high-fidelity 3D models from real-world images and images created by AI models like DALL-E and Stable Diffusion. The system produces detailed geometry and preserves complex textures like wood grains.
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. 😊