The reason GM's autonomous vehicle project failed, after spending $10 billion

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 ‏‏‎ ‎

• Why GM is giving up on robotaxis and the doubts it raises about autonomous vehicles

• Archer’s partnership with Anduril to build next-gen military aircraft

• Google’s breakthrough chip for quantum computing

• … and much more

Let's jump in!

1. GM is giving up on robotaxis, raises doubts about autonomous vehicles

Despite spending over $10 billion to develop a commercial robotaxi, General Motors is now abandoning the project.

This is significant news for the autonomous vehicle industry and it’s a strong indicator that the technology won’t be ready for many years.

That will impact their competitors in the robotaxi space - such as Waymo and Zoox - as they may find that investors are less willing to fund their research.

Back in 2016, GM acquired Cruise for $1 billion and since then it has invested huge sums into their research.

But Cruise’s downfall started several years later, when their CEO Kyle Vogt decided that they needed to accelerate their work and expand rapidly into other cities.

By expanding quickly, Vogt hoped that they could collect more real-world data than their competitors.

This would allow them to improve their safety systems and get a head start on the other companies.

While their technology is impressive, it cannot handle the wide range of situations that we deal with every day - like roadworks, ambulances blocking the street, or workers pouring wet concrete.

As a result, Cruise’s vehicles have caused issues across SF and angered locals. Just last year, their robotaxi crashed into a pedestrian and dragged her for 20 feet - causing serious injuries.

Cruise wasn’t entirely truthful with regulators after this incident, leading to their certifications being revoked. They were fined millions and hundreds of their robotaxis were removed from the roads.

With the company losing billions every year, it’s not surprising that GM has decided to give up on Cruise - but this could have a huge impact on the other self-driving companies.

2. Google reveals a breakthrough chip for quantum computing

The company's new quantum computing chip, named Willow, was able to finish a complex computing challenge in under five minutes.

That task would take one of the world's fastest supercomputers 10 septillion years, which is longer than the age of the universe.

Google’s research team also made progress in reducing errors, which is a significant challenge in quantum computing due to the nature of qubits - as they can exist in multiple states simultaneously.

The researchers found that by introducing more qubits into the system, they could correct errors in real-time.

This is a crucial step towards making quantum computing more reliable and practical. But there’s still a long road to go before the technology can be properly adopted.

3. Archer partners with Anduril to build next-gen military aircraft

This is a startup that I’ve covered before, who manufacture eVTOL aircraft and are planning to offer them as air-taxis.

They have recently signed an exclusive deal with Anduril and they will work on developing next-generation aircraft for the military.

Archer has also announced a new funding round of $430 million. This new funding will help to support its expansion into the defence sector.

The US Air Force is currently evaluating their eVTOL technology and how it could be used in military operations.

In the short-to-medium term, Archer will develop a new version of their eVTOL that uses Anduril’s autonomous software and weapons systems.

Specifically, their Lattice software will be used to quickly detect threats on the battlefield, track them, and help pilots to engage the target.

4. LLMs are reaching the limits of available training data, says OpenAI’s cofounder

At the annual NeurIPS conference, Ilya Sutskever cautioned that AI development is nearing the limits of available training data.

Sutskever compared training data to fossil fuels, as they’re both finite resources. The approaches that we use today are finally starting to reach their natural limits.

To progress further, we will need a combination of new techniques and useful synthetic data - which is when we use AI systems to create data for us.

As I’ve covered in the last few weeks, Ilya Sutskever isn’t the only one to realise this. OpenAI, Anthropic, and Google have all seen diminishing returns.

It’s a useful reminder that despite all the hype around the technology, it won’t continue to rapidly improve - unless there is another big breakthrough in how we train these models.

Until then, companies will continue to collect as much data as they can. They will then try to create their own synthetic data and hope that it will bring further improvements.

But there’s no guarantee that it will work, as that data could be low quality and not lead to any meaningful results.

5. Amazon forms a lab to develop AI agents

The new research lab is being established in San Francisco and will build the foundations needed for AI agents.

The lab will be led by David Luan, co-founder of AI startup Adept, and will work towards creating agents that can handle complex workflows.

Ultimately, it will be able to figure out the tasks that need to be done and then use a list of tools.

These tools can involve controlling your computer, web browser, or writing code.

If you want to learn more about AI agents and how they can be used by businesses, I’ve written a short explainer that you can read here.

Google reveals an AI agent that controls your browser

Google has unveiled Project Mariner, which is an AI agent that can control your Chrome browser.

The tool is able to navigate websites like a human would. It can move the cursor, click buttons, and even fill out forms.

To do this, the system regularly takes screenshots of the browser window and then sends them to their Gemini model.

It then decides what action should be taken.

However, this is a very expensive process. A lot of computing power is needed to process these images, with dozens of screenshots used for each task.

Google’s tool is very similar to Anthropic’s, which works in the exact same way. In Google’s example, they used it to navigate a grocery website and add items to the shopping cart.

While it’s fascinating to see this work, I do have concerns around trust. Since we use GenAI models to decide the action to take, there’s a genuine possibility that it will make a mistake and buy the wrong thing.

If we’re facing a future where these agents can control our computers and use our credit cards, that’s a worrying thought and it could lead to the technology only being used for very trivial situations.

Until hallucinations are severely reduced and near zero, that will always serve as a limit on how we adopt these tools.

Nevertheless, it’s incredible to see these autonomous agents in action - even if they are pretty slow and prone to making mistakes.

OpenAI has finally released their video generator

Following the US election, OpenAI has released the first version of their video generator - known as Sora.

Sora uses your text prompts to create these videos, which are incredibly high quality. But like many other video generators, they suffer from weird inconsistencies that limit their use.

The tool can only be used if you pay for ChatGPT Plus, as a lot of computing power is needed to generate the video clips.

To get the best results, you probably want to use very simple prompts. Sora seems to struggle when you use long, complex instructions.

That’s pretty strange behaviour, as many image generators work in the opposite way. When you provide them with more detail, they usually produce better results.

ChatGPT Plus and Pro users can try the tool, by using the link below.

🛸 Pentagon says the mystery drones over New Jersey are 'not US military,' not likely foreign

🎙️ Google's NotebookLM now lets you talk to its AI podcast hosts

📱 Nearly half of US teens are online almost constantly, Pew study finds

🌱 Microsoft launches a group to invest $9 billion in renewable energy

🎬 Meta debuts a new tool for watermarking AI videos

📹 ChatGPT now understands real-time video

🧳 United and Air Canada staff will use Apple AirTags to find lost luggage

💻 Apple could develop an AI server chip with Broadcom

🗣️ ElevenLabs' voice models were used to spread misinformation

⚡ Google joins project that will invest $20 billion in renewable energy

🔍 China opens an antitrust investigation into Nvidia

🤖 Microsoft launches Phi-4, a new small language model

Refute

This London startup is working to detect and respond to disinformation attacks, especially those that target large companies.

It has been a growing sector in the last few years, since these organised campaigns are becoming easier to carry out.

Plus, companies can suffer huge reputational and financial damage in just a few hours.

But Refute isn’t the only company working in this space. I met with the co-founder of LetsData last year in San Francisco, who are building similar tools.

Refute is trying to differentiate itself by providing both detection and response tools, rather than solely focusing on detection.

Tom Garnett, the startup’s CEO, has previously worked in national security at Detica and BAE Systems - with experience of investigating criminal activity and terror campaigns.

The company has raised £2.3 million in a pre-seed round, which will help them to scale further and improve their current toolset.

This Week’s Art

We’ve covered quite a bit this week, including:

• Why GM is giving up on robotaxis and the doubts it raises about autonomous vehicles

• Google’s breakthrough chip for quantum computing

• Archer’s partnership with Anduril to build next-gen military aircraft

• Why LLMs are reaching the limits of available training data and what that means for future advances

• Amazon’s new lab to develop AI agents

• Google’s reveal of an AI agent that can control your browser

• OpenAI’s release of their video generator

• And how Refute are helping companies to fight back against misinformation campaigns

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.