DAI#43 – AI Apples, search power, and Meta’s EU ‘pretty please’

Welcome to our roundup of this week’s biologically generated AI news.

This week AI took over Apple products and gave us iPhone 15 envy.

A Chinese text-to-video generator could beat Sora.

And Google’s AI search plows through our electricity and water resources.

Let’s dig in.

Searching for resources

Google’s Search Engine Experience (SGE) had a bumpy start, but despite some weird suggestions, its AI Overviews feature is actually quite useful. And extremely resource-hungry.

Guess how much more electricity and water an AI Overviews search uses than a traditional search query. Guess again.

After reading this you’ll have a different perspective on what it takes to help you find that restaurant online or generate a cute cat pic.

Apple Intelligence

Apple was slow out of the blocks in the AI race but this week it made up for lost time. At its WWDC conference, the company announced that new Apple products would get a slew of AI features including an upgraded Siri.

Apple announced that it has partnered with OpenAI to integrate access to ChatGPT from within its apps, including Siri. Would you be ok with OpenAI having access to your emails, calls, and messages?

Elon Musk made it pretty clear what he thought would happen to your data.

pic.twitter.com/7OgZAAdPf6

— Elon Musk (@elonmusk) June 10, 2024

Musk co-founded OpenAI but isn’t happy with the direction the company has taken. In February, he filed a lawsuit saying OpenAI’s pursuit of profit and keeping its models closed was a breach of contract.

This week Musk withdrew the lawsuit against OpenAI with growing speculation over the reasons behind the move. Success envy? Or one lawsuit too many as insider trading allegations mount?

No lights, no camera, Chinese action

If your job involves operating a camera or building a movie set then you’re probably not excited by the rapid developments in the text-to-video space.

We’re still waiting for OpenAI to release Sora but a Chinese company just released a text-to-video generator called Kling and it looks really good. It generates clips twice as long as Sora does at 1080p and 30fps.

The beta release of Kling while OpenAI still holds off on Sora’s release offers an interesting snapshot of the East v West approach to AI ‘safety’.

While we wait for Sora, Luma AI just made its Dream Machine publicly available. It’s free so of course the server is completely slammed. It only generates 5-second clips but the demos look good though. Expect plenty of animated memes.

Introducing Dream Machine – a next generation video model for creating high quality, realistic shots from text instructions and images using AI. It’s available to everyone today! Try for free here https://t.co/rBVWU50kTc #LumaDreamMachine pic.twitter.com/Ypmacd8E9z

— Luma AI (@LumaLabsAI) June 12, 2024

EU AI

Meta is planning on rolling out its Meta AI functionality in Europe, but the company says it needs a little help. It wants to train its AI using social media posts from European users but there’s a big pushback from privacy advocates.

Would you let Meta use your Facebook and Instagram content to train its AI? Silly question. If you’re not in the EU, then Meta has been doing it for a while now without asking your permission.

If all the younger tech-savvy EU social media users opt out, leaving only training data from older folks, you’ve got to wonder how Llama will turn out.

Another AI company that will need to navigate GDPR rules and the EU AI Act is the French startup Mistral AI.

It just secured $640M in series B funding and its AI models are challenging big players like OpenAI and Google.

As AI applications evolve we need new benchmarks to help us make meaningful comparisons.

Having an AI act as an agent to handle your mundane tasks across platforms is the next step in AI’s evolution. Google DeepMind developed NATURAL PLAN, a benchmark that measures LLMs on natural language planning.

Which LLM is best at planning a trip or scheduling a meeting with friends? It turns out that they’re all pretty bad at handling this. And GPT-4o is worse than GPT-3.5 in some cases.

AI events

This week saw two interesting AI events take place:

The AI Summit London 2024 saw speakers from OpenAI and NASA discuss the latest developments in AI and real-world applications of the transformative tech.

The AI Accelerator Institute hosted the Generative AI Summit New York 2024 bringing together a vibrant community pushing the boundaries of what’s possible with generative AI.

If you’re in Canada next week and want to see how AI is transforming the public sector then check out the CDAO Canada Public Sector event.

In other news…

Here are some other clickworthy AI stories we enjoyed this week:

• Anthropic explains the thought process behind developing Claude’s character.
• Ashton Kutcher says soon you’ll be able to use tools like Sora to generate full movies.
• The EU launched an AI-powered digital twin of Earth to improve the accuracy of climate predictions.
• Forbes accused AI search engine Perplexity of blatantly ripping off news content without attribution.
• Apple gives technical insight into its new foundation models. Just don’t expect useful industry benchmark figures.
• Google will test an AI anti-theft feature on its phones in Brazil.

And that’s a wrap.

I’m looking at my iPhone and wondering if an upgrade is justified to try out Apple’s new AI features. Do you think Elon Musk is right to be concerned about OpenAI getting up close and personal with user data?

If you live in Europe we’d love to know if you’re pro or against Meta training its AI with your social media posts. If you live outside the EU, would you opt out if you could?

Maybe the ideas of being connected to truly useful AI-powered devices and maintaining privacy are mutually exclusive. On-device AI seems the only way to go.

Let us know what you think, chat with us on X, and keep sending us those AI news links.

The post DAI#43 – AI Apples, search power, and Meta’s EU ‘pretty please’ appeared first on DailyAI.