🌊 OpenAI in stormy waters, as CEO Sam Altman is fired and joins Microsoft

Hello,

Welcome to this edition of Loop! We aim to keep you informed about technology advances, without making you feel overwhelmed.

To kick off your week, we’ve rounded-up the most important technology and AI updates that you should know about.

In this edition, we’ll explore:

- Sam Altman being fired by OpenAI
- Stability AI’s issues with copyrighted content
- Meta’s new image & video generation model
- … and much more

Let's jump in!

Top Stories

1. OpenAI’s board has fired their CEO Sam Altman [Link]

In a shock move, the company’s board have decided to oust Sam Altman, which has infuriated Microsoft and other investors. Altman has been forced to step down from his position on the board and as CEO. It comes less than 2 weeks on from OpenAI’s Dev Day, where they announced custom GPTs and their first steps in creating AI agent experiences.

The board claimed that Altman was not “consistently candid in his communications” with them and that they have therefore lost confidence in him. Soon after this, co-founder Greg Brockman announced that he has quit the company, and several other senior researchers have since followed suit.

It’s unusual for public statements like these to use such strong language, essentially accusing their former CEO of lying. We’ll likely find out more in the coming weeks about what caused such a strong rift between the other 4 board members and Altman, but it’s not the first time this has happened within OpenAI.

Anthropic AI was formed just a few years ago, after senior researchers left OpenAI due to disagreements with Sam Altman’s direction. They were pushing for the safer development of AI systems, which may also be behind the board’s recent move.

2. Altman to lead Microsoft’s Advanced AI Research team [Link]

Following the story above, it has just been announced that Sam Altman will lead the new Advanced AI Research team at Microsoft. It’s a major coup for Microsoft, who would have faced fierce competition from the likes of Meta, Google, Apple in hiring him for their own research labs.

Greg Brockman, who left OpenAI as well, will also be joining him at the company - along with “colleagues”, according to Satya Nadella. This sounds like an invitation from Microsoft to any disillusioned OpenAI colleagues who also might want to leave.

3. DeepMind unveil their new generative AI model for music [Link]

The Lyria model is to be used within two new tools. Dream Track will allow you to create music for YouTube Shorts (their TikTok competitor), while Music AI is a set of tools that are aimed at accelerating the creative process. For example, you could might hum a short tune and Music AI will convert it into a short snippet.

Dream Track is currently available as part of a limited release, with Music AI due to be launched later this year. It’s worth noting that DeepMind are not the only players in this area, as Meta made their MusicGen model available earlier in the year. You can try it out on HuggingFace.

4. Stability AI’s Head of Audio quits over their willingness to use copyrighted content [Link]

Ed Newton-Rex has resigned, since the company is willing to use copyrighted material to train their new AI products, but without the owner’s permission. Stability AI isn’t the only large AI company doing this, as many of these companies claim that using the material in this way falls under “fair use” and they don’t need the copyright holder’s approval. Newton-Rex also questions whether it is right for billion-dollar companies to use this content and then create products that could replace artists.

5. Meta announce Emu, their improved image and video generation model [Link]

The foundation model is being used to both generate new videos and to edit existing images. For the text-to-video generation, Meta has been able to make Emu much more efficient than the previous Make-A-Video model - which was announced just a year ago. With Emu, just 2 diffusion models are needed, compared to 5 with the previous generation.

But it’s the ability to more precisely edit existing images that’s most exciting. AI image generators can be a bit frustrating at times, since they create completely new versions of the image each time. You can’t just ask the model to only change a specific area in the image. Emu Edit aims to address this, since it will try to only modify the relevant pixels and will preserve the rest of the image - which gives creatives much more control over the process.

6. DeepMind’s new model pushes weather forecasting a step forward [Link]

It’s been a busy week for AI model announcements, but GraphCast is a reminder of the great work that is happening outside of the GenAI space. It’s able to outperform the traditional High Resolution Forecast (HRES) system we have today and offers 10-day forecasts that are more accurate - and all this can be calculated in under a minute. This means we can forecast extreme weather events much better than before, such as cyclone tracks, atmospheric rivers, and extreme temperatures.

While this new advancement mightn’t capture imaginations like some of the other models that have been announced this week, it will be of huge help for governments as they now have to prepare for more unpredictable weather patterns and flooding. It won’t feature in many headlines, but we’ve added a great new tool to our arsenal as we grapple with changes to the climate.

Closer Look

Need to create a new UI design?
You can now draw it yourself

A fascinating tool is TLDraw, which allows you to draw very simple website designs and will then convert them into a functioning prototype. This is all possible thanks to GPT-4 Vision, which is able to analyse the design and generate the code. You can even add text boxes like “users should be able to click this” and it will be changed into a working button. Amazing!

This will allow companies to do rapid prototyping and test lots of different ideas, while incurring very little costs. Tools like this also reduce the learning curve needed for design tools, such as Figma or Adobe’s suite, and allow for more people to get started.

UI designers may be a bit worried after seeing this, but it’s important to remember that good design is more than just how the screen looks. Designers have skills that go beyond the actual generation stage and have built up years worth of knowledge on what it takes to create great user experiences. The same is true of developers, who know how to effectively deploy software and reduce the risk of a delivery project failing. These are points we must remind ourselves, as tools like these become more prevalent.

If you’d like to try it out, you can visit the TLDraw website. However, you will need to include your OpenAI API key. For added security and to avoid huge bills if your key does get leaked, I recommend setting a spending limit on your OpenAI account.

Byte-Sized Extras

• Meta disbands its Responsible AI team [Link]

• Amazon to sell cars online, starting with Hyundai [Link]

• Toku’s AI platform aims to predict heart conditions by simply scanning your eye [Link]

• Amazon is laying off hundreds of people in their Alexa division [Link]

Startup Spotlight

Krea AI

Image generation is a fun, but slightly frustrating process. You have an idea of what you want to create, but don’t know what exact prompt will allow you to create it. Based on the work that Krea AI have been doing, that probably won’t be a problem for long.

Using their tools, you can draw shapes to indicate what you want and the AI will suggest lots of different options. You can then pick from these, or if you’re still not satisfied you can add more detail to show how’d like it to look.

It looks as though this is possible through the use of Latent Consistency Models (LCMs). They allow for rapid image generation, since the number of steps in the process are brought down from 25 to about 5. Similar to TLDraw and how it will impact designers, tools like Krea’s will turbo-charge how artists and creatives can do their work.

The idea of being able to guide the image generator and get the exact result you need, is pretty exciting and opens up a lot of new opportunities. In a few years, we’ll probably wonder why we needed to write long prompts in the first place.

If you want to request access to the platform, you can check out their website.

Analysis

It’s impossible to avoid the crisis that’s happening right now at OpenAI. Just a few weeks ago, Sam Altman was on stage for their developer conference and was often compared to Steve Jobs on Twitter. In the last few days, he was voted out by the 4 other board members and essentially accused of lying to them - echoing a similar situation that Jobs faced in 1985.

There’s no doubt this will have a big impact on staff morale at the company, especially following days of speculation that Altman could return. But that is now out of the question. The OpenAI board has since appointed the co-founder of Twitch, Emmett Shear, as their new CEO.

Some senior staff have already quit in protest and more could follow. While the company has faced a similar struggle before, this situation is much more perilous. The board has moved swiftly to remove Sam Altman, but the lack of communication as to why - perhaps due to the potential for legal action against them - risks a mass resignation event from key staff.

Clearly, the board have been so alarmed by something that Altman has done, or information that he has kept from them, that they felt they had to remove him immediately. The fact that they didn’t even alert major investors, such as Microsoft, further highlights the severity of the situation. It’ll be interesting to hear what their reasoning was for the decision.

The last few days have inflicted huge damage on OpenAI’s reputation. Before this, it looked as though the company was on an unstoppable charge - with GPT-3, DALL-E, ChatGPT, GPT-4, their Microsoft partnership, and convincing over 100 million people to use their services every week. But following recent events, that juggernaut looks a lot less stable than we previously thought.

This Week’s Art

Prompt: Envision a scene reminiscent of Salvador Dali's 'The Persistence of Memory', reimagined on Mars. The landscape features the characteristic reddish, rocky terrain of Mars, creating an otherworldly setting. Melting clocks, a hallmark of Dali's famous painting, are draped over Martian rocks and the uneven ground, blending the iconic surrealist imagery with the alien Martian environment. The Martian sky is a vivid orange with swirling, ghostly clouds, evoking a sense of the unknown and distant. In the sky, Earth appears as a small, distant dot, emphasizing the isolation of Mars. The overall composition should be dreamlike and surreal, combining elements of Dali's style with the unique Martian landscape.
Platform: DALL-E 3

End Note

What a week. We’ve seen advancements in generate AI for both image and video, a huge step forward in weather forecasting from DeepMind, concerns over copyrighted content being used to train new AI models, Meta disbanding their Responsible AI team, and - of course - Sam Altman being fired as OpenAI CEO.

Oh and we’ve gained a lot of new subscribers recently. To all of you, welcome on board and I hope you’re enjoying the insights.

Have a good week!

Liam

Share with Others

If you found something interesting, feel free to share this newsletter with your colleagues.

About the Author

Liam McCormick is a Senior Software Engineer and works within Kainos' Innovation team.
He identifies business value in emerging technologies, implements them, and then shares these insights with others.