The SEC Newgate AI Weekly

It’s been another fast-paced week in the world of AI, with big developments from Microsoft, Midjourney and LinkedIn.

Here’s a rundown.

Microsoft

Microsoft announced that an early preview of its AI-powered Windows Copilot personal assistant would be available to insiders in the Windows 11 Dev channel.

The company’s promo video for Copilot shows a text-based personal assistant that can:

• Give suggestions on using your computer more efficiently
• Recommend music to you before opening Spotify to play it
• Create logos in Adobe Express before posting the results for you in Teams

The way in which the AI assistant can integrate with other applications like Adobe Express certainly looks impressive and could really improve efficiency.

Midjourney

Midjourney has made a name for itself this year with its AI-generated images, created by natural language descriptions. Until now, Midjourney would present you with four image options, based on your prompt. You could then ask for more variations of the image you liked the most, to fine tune the final output. But what if you wanted to extend one of those images?

Now Midjourney can ‘pan’, meaning you can extend the image to the left and right. You can also extend the image above and below, which is still referred to as panning, despite an up/down movement usually being referred to as a tilt.

This option has been available in Open AI’s DALL·E 2 since last year, where it is referred to as ‘Outpainting’. With Midjourney adding ‘panning’, both the main image generating AIs have image extending options.

LinkedIn

With powerful AI systems like Midjourney and DALL·E 2 able to produce such realistic images of people you may be wondering how it’s possible to tell a picture of a real person from one generated by AI.

This is a question that has also been troubling LinkedIn, who want to ensure the accounts on its platform are of real people.

As a result, LinkedIn teamed up with University of California, Berkeley on a study to detect AI generated profile images.

The resulting method can detect AI profile pictures 99.6% of the time, which should help LinkedIn to ensure that the profiles on its platform are authentic.