Fake images have never looked so real, thanks to this German startup

Welcome to this edition of Loop!

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

‏‏‎ ‎ HIGHLIGHTS ‏‏‎ ‎

• Apple’s work on a new home device with a robotic arm

• How Iran tried to influence the US election with ChatGPT

• Why Google’s new Gemini Live assistant is a big deal and will challenge OpenAI’s advanced voice mode

• … and much more

Let's jump in!

1. Apple has hundreds working on a robot arm for home-use

We start with Apple, who are developing an expensive tabletop home device that combines an iPad-like display with a robotic arm.

The project, codenamed J595, was approved by Apple's executive team in 2022 and now has several hundred people working on it.

This has been rumoured for a while, but it’s only recently that more details have been leaked.

It seems that Apple wants it to be used for controlling your smart home, video calls, and as an entertainment hub.

With Apple Intelligence being released later this year, the home environment will give them a new touch point into peoples’ lives and encourage them to use those AI features.

This is part of the wider strategy to keep customers engaged with the Apple ecosystem and reduce the risk that they will be overthrown by new challengers - especially given the rapid rise of AI companies, such as OpenAI and Anthropic.

2. IKEA will expand its drone program to manage warehouses

The Swedish furniture giant is embracing drones and using them to manage inventory across its massive warehouses.

The company has deployed 100 drones across Europe, which will provide around-the-clock updates to inventory and can reach physical heights that are beyond most humans and robots.

The drones are being produced by Verity, which is a Swiss startup that has raised $43 million in the last few months.

It’s an interesting idea and has clearly exceeded IKEA’s expectations, given that they’re rolling it out to over 16 locations.

But I think there’s a lot of potential for more hazardous locations, such as chemical plants or contaminated buildings.

If we can automate multiple drones to work within warehouses and scan inventory, then we should be able to do the same for dangerous situations and search areas.

3. Researchers have developed a hair-thin battery to power tiny robots

MIT researchers have developed a tiny battery that’s designed to power miniature robots.

The battery’s so small that it measures a mere 0.1mm long and 0.002mm thick - roughly the size of a human hair.

It’s able to generate up to 1 volt, which is enough to power sensors, circuits, and to move actuators in tiny robots.

Next, the researchers want to create a fully enclosed version of the battery and then generate a lot more voltage.

While it’s a tiny amount of power in a tiny battery, we’ve seen for years that battery technology has been slower to progress than compute - so it’s encouraging to see this advancement.

4. Iran tried to influence the US elections with ChatGPT

OpenAI has banned several ChatGPT accounts that were connected to an Iranian campaign, known as Storm-2035.

Not the most discreet name for influencing the 2025 change of President.

Iran’s campaigners were using ChatGPT to generate articles and social media comments that focused on the US election, wars in the Middle East, and other divisive topics.

OpenAI says that the Iranians didn’t get much audience engagement, with most of their posts receiving very little interactions.

But that misses the point. While only a few hundred people might have seen one post, the true number is a lot more when dozens of fake accounts are used.

Unfortunately, we should expect a lot more of this going forward - and it’s only going to become harder to spot AI messages as time goes on.

5. Filmmakers say AI will completely change their art-form

The panel at SIGGRAPH were concerned that there could be a flood of low-quality films, as the big studios pursue even bigger profits.

Although, we could also see a new genre created - with it becoming more interactive and combining some of the elements we see in video games, movies, and real-life.

I don’t think that Generative AI, and other technologies, will impact the industry in the short term.

But it’s difficult to be certain about the medium-to-longer term, especially when you consider how fast this technology is evolving.

Yes, we don’t have a lot of control over the videos that are generated - but that will come eventually.

Just take a look at what OpenAI has achieved with DALL-E and its image editor. You can simply select an area and type a prompt, with the rest of the image remaining the same.

That allows you to fix issues in the image and it’s only a matter of time until the same can be done with video.

Google takes on OpenAI with Gemini Live

Google has unveiled a new voice chat feature, called Gemini Live. It’s pretty similar to GPT-4o’s advanced voice mode and Gemini is a lot more expressive in how it responds back.

You’re able to talk to Gemini and have a back-and-forth conversation, interrupt it mid-sentence, or tell it to change topic.

To get access you’ll need to subscribe to the Google One plan, which costs about $20 a month.

Seeing these new capabilities is really exciting.

If someone said that in 2024, you’ll be able to have a real conversation with a chatbot and tell it to mimic different emotions, you wouldn’t believe them.

And we now have two companies offering this: OpenAI and Google.

Some people have suggested it could be used as a “study-buddy” to help kids with homework. But that’s not a perfect use-case, as it’s still a LLM underneath and it will make things up.

That’s not great when you’re doing your math homework.

What I’m excited about is the real-time language translation. Google Translate has been a great resource for years, but not very accurate.

Since these LLMs understand language so well, and only make minor mistakes, I see it being more useful for when you’re in a different country and asking for directions.

Technology doesn’t have to have big, earth-shattering use-cases. Sometimes, the less-obvious ones have a greater impact on our lives.

California’s new AI bill is ill-informed, say Pelosi and Silicon Valley figures

The bill aims to prevent large AI models being misused, especially if they could cause "critical harms" against humanity.

It was passed by California’s Appropriations Committee, but has been weakened after Nancy Pelosi and Silicon Valley figures opposed it.

The bill will still apply to AI models that cost over $100 million to train, with developers made liable for implementing safety controls.

But the Frontier Model Division (FMD), which would have become a new agency in California and enforced the legislation, has been ditched.

The requirement that AI labs should submit safety certifications, “under penalty of perjury”, has also been removed.

Now, only public statements about their safety policies are required.

For those companies who are fine-tuning a model, you don’t need to worry as you won’t be impacted - as long as you spend less than $10 million on fine-tuning.

Overall, it seems like state legislators made the right choice. There are some new protections for the public, but Silicon Valley companies won’t have their hands tied by red-tape.

I visited the Valley last September and spoke to dozens of start-ups that are pushing the industry forward.

The tech scene there can’t be matched anywhere else and I’m glad to see that will continue to be the case.

❌ Study suggests that the best AI models hallucinate more than claimed

🛰️ Lockheed Martin will buy satellite maker Terran Orbital in a $450 million deal

❄️ Waymo to double down on winter testing for its robotaxis

⚠️ MIT researchers have released a repository for AI risks

🚕 WeRide, a Chinese robotaxi startup, gets approval to carry passengers in California

📚 South Korean parents are sceptical about a new AI textbook program (paywall)

📢 Waymo robotaxis keep honking at 3am and stop neighbours from sleeping

🔓 National Public Data admits it leaked Social Security numbers in a massive data breach

🤔 Teachers worry that GenAI will hurt students' critical thinking skills (paywall)

🤖 LangChain now allows you to easily create synthetic data

Black Forest Labs

This German startup has gained a lot of attention recently, after they launched a new AI image generator that had minimal safeguards.

The FLUX 1 model works pretty well and can create images that are close to Midjourney’s level.

The image above was made using a combination of FLUX and other tools. It’s genuinely surprising how realistic it looks.

They’ve just partnered with Elon Musk’s Grok, which is using the image generator to power their own tool.

Black Forest Labs has raised $31 million funding and are hoping to challenge OpenAI’s DALL-E and Midjourney. They plan to release a video generator in the coming months.

Completely rejecting safeguards is an unusual tactic, given that a lot of European startups focus more on AI safety than their American counterparts.

That’s partly due to cultural differences and the weaker VC culture in Europe, alongside the risk of new EU regulations.

So, it’s strange to see a European company lean into the lack of safeguards. Of course, they’re a startup and need to be bold if they want to disrupt the status-quo.

But they’ll need to walk a fine-line. If EU officials think that it’s being used to spread disinformation and deepfakes that target young girls, they could face heavy sanctions.

There’s no doubt that Black Forest Labs have gotten off to a flying-start. They’ll just need to be careful and not fly too close to the Sun.

This Week’s Art

Quite a lot has been covered this week, including:

• Apple’s work on a new home device with a robotic arm

• IKEA’s expansion of a drone program for their warehouses

• MIT research that pushes the boundaries of tiny batteries

• Iran’s attempt to influence the US elections with ChatGPT

• Why filmmakers are convinced AI will completely change their art-form

• Google’s unveiling of Gemini Live to challenge OpenAI

• The debate around California’s new AI bill

• And how Black Forest Labs have went from a steal-startup to partnering with Musk’s Grok

I’ll be on holiday next week enjoying the sun in Greece, so you’ll get the next post the following Monday.

Have a good week!

Liam

Share with Others

If you found something interesting in this week’s edition, feel free to share this newsletter with your colleagues.

About the Author

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