NVIDIA has a competitor

Reading time: 9 minutes

Newsletter | News | Tools

Today we will discuss -

• 📸Meta releases new AI image-creation model

• 🍟How AMD's latest AI chip could be a game-changer

• 🚀OpenAI introduces updates to their API

All this and more - Let's dive in!

🍚What’s cooking in the newsroom

Meta releases new AI model that can create more realistic images

🍋News - Meta is making its new AI model, I-JEPA, open source so that researchers can use it to accelerate the development of AI research. Meta believes that by making I-JEPA available to a wider audience, more people will be able to contribute to its development and improvement, which will lead to faster innovation in the field of AI.

🤔But what’s the function of the new model?

• I-JEPA can complete unfinished images more accurately than existing models. It was trained on a massive dataset of images and text, and it can learn to identify the important features of an image without having to be told what to look for.

• It uses a technique called "background knowledge" to fill in missing pieces of images rather than looking only at nearby pixels like other generative AI models.

• For example, if you have an image of a car with the top half missing, I-JEPA can use its knowledge of cars to fill in the missing top half. It knows that cars have roofs, windshields, and headlights, so it can fill in these missing parts based on its knowledge of cars. This is in contrast to other generative AI models, which would only be able to fill in the missing top half of the car by looking at the nearby pixels.

AMD challenges Nvidia with new MI300X AI chip

🥣News - On Tuesday, AMD announced its new MI300X AI chip, which could challenge Nvidia's dominance in the AI chip market.

The MI300X is designed for use in a variety of AI applications, including natural language processing, computer vision, and machine learning. AMD says that the MI300X will be available to customers towards the end of 2023.

😈How AMD’s new chip could give competition to NVIDIA?

• The MI300X can use up to 192GB of memory, Nvidia’s rival H100 only supports 120GB. Thus, MI300X is more suitable for training and running large language models.

• The MI300X is based on AMD's latest generation of GPU architecture, which is designed to be more efficient and powerful than previous generations. This could give the MI300X a performance advantage over NVIDIA's current generation of AI chips.

🤡Did AMD get a customer? Well, Amazon Web Services (AWS) is considering using AMD's new chips in its cloud services, according to Dave Brown, vice president of elastic compute cloud at Amazon. However, AWS has not made any public announcement yet.

OpenAI introduces key updates to help developers build better AI applications

📻News - OpenAI on 13 June announced four major updates to make its platform more accessible and useful for developers.

These updates include:

1. More steerable API models: This means that developers will have more control over the behavior of the API models, which will allow them to build more customized and sophisticated applications. An API is a way for two pieces of software to talk to each other.

2. Improved function calling capabilities: When developers talk about "improved function calling capabilities," they're talking about the ability to call custom functions from within an API. This means that developers can use the API to do things that the API itself can't do. Function calling capabilities give developers more flexibility in building their applications.

   🙈Here's an analogy - Imagine you're playing a video game. You can only do the things that the game allows you to do. But if you could write your code, you could add new features to the game. This would make the game more fun and more challenging.

3. Longer context: OpenAI is also releasing a new version of their GPT-3.5-turbo model with a much larger context window. It means that the model can keep track of more information from the conversation, which can help it stay on topic.

4. Lower prices: OpenAI has made its text-generating models cheaper to use. GPT-3.5-turbo will now cost $0.0015 per 1,000 input tokens and $0.002 per 1,000 output tokens. This means that you can generate about 700 pages for every dollar you spend.

🤖What else is happening?

• Google delays EU launch of its AI chatbot after privacy regulator raises concerns

• Hugging Face and AMD partner on accelerating state-of-the-art models for CPU and GPU platforms

• Itoka wants to license AI-generated music via the blockchain

• Adobe announces ‘Generative Recolor’ AI feature for Adobe Illustrator

• Oracle develops new generative AI service with Cohere

🎵AI song recommendation

• On the Edge is a rock song that was composed by AIVA, an AI music composer. The song has a catchy melody and powerful vocals, and it has been praised by critics for its originality and creativity.

• Sensitive by SKYGGE and C Duncan is a folk song that was created using the AI music platform Amper Music.

• Not Mine by Miquela is a pop song that was composed using the AI voice of Miquela, a digital human created by the company Brud. The song has been praised for its catchy melody and its futuristic sound.

🤯 Discover mind-blowing AI tools

1. Voicemod - AI-powered voice-changing software that can be used to modify voices in real-time

2. GPTZero - A tool for accurately detecting AI plagiarism

3. Synthesia - Enables businesses to create videos from plain text in minutes

4. Phind - An AI-powered search engine that provides users with instant answers to technical questions

5. Craiyon - An AI model that can draw images from any text prompt

6. Blackbox - Enables users to copy code from videos, images, PDFs, and other formats

7. Pitch - Combines productivity, design, and presentation software for fast-moving teams

8. vidIQ - Provides users with an intelligent mix of technological expertise to get more views on YouTube

9. Rytr - Helps you create high-quality content, in just a few seconds

10. BlueWillow - An AI tool designed to help users create logos, graphics, photo-realistic scenes

😱Did you know?

• In 2017, a Google AI called Eugene Goostman fooled 33% of human judges into thinking it was a real 13-year-old boy. This made Eugene Goostman the first AI to pass the Turing test, a test of a machine's ability to exhibit intelligent behavior equivalent to, or indistinguishable from, that of a human.

• In 2018, an AI called OpenAI Five defeated a team of professional Dota 2 players. This was the first time that an AI had defeated a team of professional human players in a competitive video game.

• During the 1970s and 1980s, progress in AI research faced significant challenges and experienced what became known as the "AI winter." The initial enthusiasm for AI did not translate into practical applications, leading to decreased funding and interest in the field.