Is Microsoft Competing with OpenAI?

Today’s top AI Highlights:

1. Microsoft is building its first-ever large language model in-house, with 500B parameters

2. AI software engineer Devin is getting new features for more control, collaboration, and better automation

3. US govt. invests $285 million in digital twin technology to advance semicon manufacturing

4. AI wearables and devices are much worse than 1.0 versions: Rabbit R1 reviews

5. Build your Personal AI Assistant for iPhone using Apple Shortcuts, Cloudflare and Llama 3

& so much more!

Read time: 3 mins

Latest Developments 🌍

Microsoft’s AI Model to Compete with OpenAI, Google and Anthropic 🚴‍♂️

Microsoft is reportedly building its first-ever large AI model, and no it’s not with OpenAI for the GPT series. The company is developing MAI-1, an LLM with 500B parameters. The project is being overseen by Mustafa Suleyman who Microsoft had poached from Inflection AI last month. It’s certain by now that Microsoft is decreasing its dependency on OpenAI’s models for powering almost all its AI products, be it Microsoft Copilot or in GitHub.

Key Highlights:

1. Though Microsoft is building a model competing with OpenAI, the CTO says that the companies’ partnership is important for both companies to build the best AI.

2. MAI-1 will be the first-ever large AI model that Microsoft will develop in-house. Before this, they had released a series of small models called Phi, where the latest and the largest Phi-3 model has 3.8 billion parameters.

3. The company has been investing heavily to develop its infrastructure for training large-scale AI models. It is building a $100 billion supercomputer called Stargate, and has also partnered with the Indonesian government recently to develop AI and cloud computing infrastructure in Indonesia for $1.7 billion.

Devin’s New Features for More Control and Collaboration 🧑‍💻

Just a couple of months back, the first autonomous AI software engineer Devin was released by Cognitive Labs which created quite a stir in the AI community, and it has already received some major updates. An X user Andrew Gao who has access to Devin posted a thread on these new features, however, Cognitive Labs has not officially announced them. The new features let you take control and guide Dein when it’s stuck or is going wrong, and also boost its automation.

Key Highlights:

1. Devin’s UI now has an interactive mode that lets you help it navigate the web. This is particularly useful if Devin gets stuck on something like a CAPTCHA or needs assistance with specific website interactions.

2. You can now directly edit the code that Devin generates through a web-based VSCode interface, providing greater control, collaboration, and flexibility.

3. Devin can now log into websites for your account. You need not share your passwords. This enhances security while still allowing Devin to perform actions within your accounts.

4. Devin’s coding ability and website navigation appears to have improved, probably due to better prompting techniques or a stronger underlying model.

5. You can save Devin’s state using machine snapshots, ensuring continuity even if servers shut down.

Digital Twins for Chip Design and Production 👯

The US government is allocating $285 million to promote research and development in “digital twin” technology for the semiconductor industry. This funding is a part of the CHIPS and Science Act of 2022 which was enacted to boost domestic chip manufacturing and reduce reliance on foreign suppliers.

Key Highlights:

1. Digital twins are virtual copies of physical chips, which are used for testing and optimization before actual production. AI can also be leveraged to speed up the testing and manufacturing of the chips.

2. The $285 million will support research activities, workforce training, facility development, and demonstration projects related to digital twin technology.

3. A “regionally diverse” network will be created to connect companies and institutions involved in semiconductor development and manufacturing.

4. Despite this Act, US manufacturers face challenges as this industry demands huge capital requirements. Intel’s foundry alone reported an operating loss of $7 billion for 2023, and is set to receive a hefty grant of $8.5 billion in direct funding under the Act.

Rabbit R1 is like the AI Pin - Inaccurate & Unecessary! 🤦‍♀️

It seems that the current AI devices like Humane’s AI Pin are just hurried half-ready releases that promise a futuristic experience only in demos but are not living up to expectations. These devices are released as 1.0 versions but feel like the 0.5 versions. A few days back, early reviews on Humane AI Pin were out which were just disappointing! This week it was Rabbit R1 which is head-to-head with the AI Pin in both promises and real-world use.

Here’s what the reviewers had to say:

Positives:

1. Design and Interface: R1 stands out with its bright orange color, scroll wheel, and button for activating the voice assistant. The design looks quirky, the learning curve is much flatter, and the screen is more convenient than the AI Pin’s laser projector.

2. Cost: R1 costs $200 and does not require a recurring subscription fee, a more accessible alternative to the AI Pin, which costs $700 with a $24 monthly subscription.

Drawbacks:

1. Limited Integration: Far from the “Tech Mode” in the demo, R1 currently supports only 4 apps: Spotify, Uber, DoorDash, and Midjourney. Integration even with these apps is unreliable, with frequent errors even for basic tasks like playing music or ordering food.

2. Underdeveloped Features: Features like the ability to learn and use any app, are still under development. Even other basic features like setting a timer aren’t available, so the current version feels unfinished.

3. Battery Life: R1’s battery life is poor, lasting only a few hours.

4. Inaccurate: R1, like the AI Pin, Meta Ray Ban Smart Glasses and other AI devices, hallucinates with high confidence. It misidentifies objects and gives inaccurate answers, however, its responses are faster than the AI Pin.

The objective of these devices is to eliminate the need to navigate the chunks of apps on our smartphones and stay in the moment by just telling the AI assistant to do routine tasks. But considering these versions, smartphones will stay!

(Sources: Techcrunch, The Verge, Marques Brownlee)

Tools of the Trade ⚒️

1. Personal AI: Talk to your personal AI assistant on your iPhone. Leveraging Apple Shortcuts, Cloudflare Workers, and the Llama 3 language model, build your own Personal AI Assistant to interact with AI directly from your phone. With voice commands and text input, you can ask questions, get information, and even utilize custom functions like weather updates.

1. Milio: Excel in interviews with this AI interview companion. With just a microphone, Milio listens, provides real-time answers on your screen, and adapts to include new interview question types or trends, ensuring you’re always prepared.

2. Haechi AI: Create and upscale any type of content, be it image, audio, code, or text, 10x faster, using the best AI models like GPT, Gemini, Claude 3, and SDXL, all from a single subscription. With top-tier models and powerful Nvidia hardware, seamlessly switch between AI models, enjoy fast processing, and explore your creativity.

3. DevOn: Do what the AI software engineer Devin does, but with MultiOn’s AI agents. DevOn uses three special AI agents for programming, researching, and note-taking, all orchestrated and supervised by GPT-4 Vision.

Hot Takes 🔥

1. Forget forecasting the performance curve of AI, we will learn more soon about whether exponential growth continues. But it seems weird that so many people are fixating on the cost of using a GPT-4 class model today. Llama 3 400b is coming, and should drive that down pretty fast. ~Ethan Mollick

2. When someone builds an AGI, OpenAI will plug it into ChatGPT and claim ChatGPT is AGI. ~Pedro Domingos

Meme of the Day 🤡

We are (and will be the only) to build AGI 😈

That’s all for today! See you tomorrow with more such AI-filled content.

Real-time AI Updates 🚨

⚡️ Follow me on Twitter @Saboo_Shubham for lightning-fast AI updates and never miss what’s trending!

PS: I curate this AI newsletter every day for FREE, your support is what keeps me going. If you find value in what you read, share it with your friends by clicking the share button below!