A week in Generative AI: Midjourney, Perplexity & Robots
News for the week ending 25th August 2024
This week we’ve finally got the launch of Midjourney on the web, allowing everyone to try it for free without having to navigate Discord servers. There’s also an impressive new video model from Luma and news that Perplexity will start offering advertising on their platform later this year. There’s also the announcement of an incredibly cheap humanoid robot from Unitree and Boston Dynamic’s Atlas doing push ups!
In ethics news, Meta and Spotify’s CEOs criticise the EU’s AI regulations for holding back progress in the region, OpenAI come out in opposition of California’s new AI laws that disappoints former employees, and DeepMind workers protest against Google’s defence contracts.
There’s also a good long read from the Verge exploring how we’re not ready for the Google Pixel 9’s Magic photo Editor - our basic assumptions about photos capturing reality are about to go up in smoke.
Midjourney's AI-image generator website is now officially open to everyone
Midjourney has arguably been the leading text-to-image model for a long time, but has always been unit in Discord, which isn’t for everyone, and not very intuitive to use. It’s great to see them finally releasing their capabilities on a straightforward website, with lots of the rich editiing features also avaialble. It’s free for anyone to try out, so I’d highly encourage anyone who hasn’t given MidJourney a try to now give it a go!
Perplexity AI plans to start running ads in fourth quarter as AI-assisted search gains popularity
There’s been plenty of discussion about what the future of search looks like with Generative AI and Perplexity have been at the forefront of figuring that out. Advertising has always been a huge part of search, so its no surprise to see Perplexity start to explore advertising as they seek to monetise their technology and bolster their revenue streams.
Perplexity will be offering advertisers a CPM (cost per thousand impressions) model, which is a different model to the search advertising indsuty’s traditional CPC (cost per click) model. CPMs will be >$50, which is a high premium, with advertisers able to sponsor “related questions“ below Perplexity-generated answers and also buy display ads to the right of them. Key to Perplexity’s high prices are claims that their users are high net worth individuals and advertisers will get a higher share of voice on the patform.
It will be interesting to see how this plays out. There’s definitely volume there with 230m queries a month, but I’m not convinced that advertising revenues are the answer to funding Generative AI search platforms, especially as Sam Altman has clearly stated they’re not interested in an advertising model for any of their products.
Luma launches Dream Machine v1.5
There’s been a lot of progress made by various text-to-video models this year, so it’s great to see one of the pioneers release the next version of their model. Dream Machine 1.5 offer higher-quality text-to-video, smarter understanding of your prompts, custom text rendering, and improved image-to-video generation.
One of the things I really like about Luma is they’re honest about the limitations their models still face. On their website they list things like morphing (where objects change), movement (where movement is inconsistent), and text (generating text in videos) as challenges their model still faces. Regardless, the demo video above is very impressive!
Unitree Launches Production Version of G1 Humanoid Robot
One of the issues I’ve seen with the progress being made with robotics is that whilst there have been lots of impressive demo videos there’s not been much news about when these robots will ship, how much they will cost, or any videos of them operating in real world scenarios.
Unitree is now looking to change that with the announcement that they are launching the production version of their G1 humanoid robot for just $16,000 🤯. That’s 90% cheaper than their previous robot, the H1, which cost $150,000.
The G1 is only 4 feet tall, but can perform complex actions such as spinning, navigating stairs, bouncing, and maintaining balance when pushed or stumbled. However, the G1 is not yet ready for consumer tasks like cooking breakfast or tidying up a home, it’s being positioned as more of an affordable platform for research and development. Still, there’s a huge amount of technology packed into such a small and relatively affordable package!
Boston Dynamics’ new electric Atlas can do pushups
I love me some Boston Dynamics demo videos and this one is a bit of fun! On the face of it, nothing too impressive with a robot doing a press up, but when you think about it, I’m not aware of any robot previously being able to do this. It shows good progress in strength, dexterity, and co-ordination and I’m still a bit freaked out by how this robot moves 🤓.
AI Ethics News
EU Commission opens consultation on General-Purpose AI Code of Practice
Demand for AI is driving data center water consumption sky high
OpenAI's opposition to California's AI law 'makes no sense,' says state Senator
‘Disappointed but not surprised’: Former employees speak on OpenAI’s opposition to SB 1047
DeepMind workers sign letter in protest of Google’s defense contracts
Procreate takes a stand against generative AI, vows to never incorporate the tech into its products
Long Reads
The Verge - No one’s ready for this
Benedict Evans - Competing in search
“The future is already here, it’s just not evenly distributed.“
William Gibson