Anthropic's new AI can control your computer

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 ‏‏‎ ‎

• The startups that are securing megadeals with VCs

• Why the Pentagon is being ordered to quickly adopt AI tools

• Google’s research on an AI that can use your PC

• … and much more

Let's jump in!

1. Startups are securing megadeals with VCs, but AI isn’t the #1 sector

According to Aumni's Venture Beacon report, down rounds were still at near record highs through the first half of 2024, particularly for late-stage deals.

However, new data from Crunchbase reveals a boom in megadeals, with nearly 240 funding rounds of $100 million or more for US-based startups so far this year, surpassing the total for all of 2023.

Interestingly, biotech and healthcare startups accounted for the majority of these megadeals, with 87 deals, while AI came in second with 26 deals.

Some notable health tech megadeals include the AI drug discovery company Xaira Therapeutics, which launched with a $1 billion round, and Superluminal Medicines' $120 million Series A.

Cybersecurity is another sector attracting mega-rounds, with startups like Kiteworks, Cyera, and Wiz raising substantial funding.

2. Apple Intelligence will make its big debut this week (sort of)

The company plans to gradually release their AI features, rather than all-at-once, as performance improvements are still needed.

Although, they will release the first batch of features - including writing tools, such as Text Rewrite, Proofread, Summarise Text, and Smart Reply.

Siri is also being revamped, with a new UI and the option to type questions instead.

It’s strange that Apple never thought to enable this feature before. No one wants to talk out-loud to Siri when you’re in public.

You will also be able to use “Clean Up” for your photos, which uses AI to remove objects from your photos.

But some of the most eye-catching features, such as Visual Intelligence, image generation, and ChatGPT integration won’t be included. They’re scheduled for release later in the year.

While it makes sense for Apple to delay these features and ensure that they work properly, it will irritate users - especially those who have bought the most recent iPhone.

It’s not a good look that Apple Intelligence isn’t available at launch, but delaying their AI tools for several months is even worse - it reinforces the perception that Apple has been caught off-guard by OpenAI and are behind the curve.

3. Perplexity targets a $8 billion valuation, but faces multiple setbacks

If you’ve never heard of Perplexity, they’re a startup that is developing an AI search engine.

Ultimately, they aim to erode Google’s dominance in the sector and have prioritised users that are high-earners.

While the company has had a somewhat difficult year, it has seen huge levels of growth. They are now serving over 100 million search queries every week.

But the company faces several issues. Huge media companies have accused Perplexity of unauthorised web scraping and plagiarism.

The New York Times recently sent a cease-and-desist letter and News Corp has even decided to sue the startup.

Media companies are very worried about Perplexity and what it means for their business model.

Since the startup is summarising their content, they are not earning any ad revenue from Perplexity users - which threatens how they can fund investigative journalism.

Other websites are also at risk, including many of your typical blogs or favourite websites, as they will struggle to fund future maintenance.

4. Google is developing an AI agent that controls your computer

The tech giant is reportedly developing a large action model, codenamed "Project Jarvis," which could be released as early as December.

The tool will be able to carry out tasks for you - such as research, purchasing items, or even booking flights.

According to insiders, it will use a new version of Google’s Gemini model and will be specially designed for the Chrome browser.

The news has come at a good time. Anthropic has just released a similar tool, which I cover below in a lot more detail.

5. White House orders the Pentagon to move faster and adopt AI

President Biden has issued a new national security memorandum, which is aimed at accelerating AI adoption within Pentagon and intelligence operations - while also addressing security concerns.

Jake Sullivan, the US National Security Adviser, has emphasised that the US has a strong lead over other nations - but swift adoption is needed to maintain this advantage.

A new framework will ensure that the private sector works more closely with government agencies, rather than the patchwork approach they currently have.

They will also establish a new working group, which will tackle procurement challenges and recommend AI technologies that should be adopted by the national security agencies.

However, this raises new security risks for the US - since adversaries will be desperate to get their hands on these tools.

As a result, all agencies have been ordered to urgently review their security systems.

Anthropic's new AI model can control your computer

For some time now, the AI startup has been on a roll. They released Claude 3 Opus in March, which is an incredible model and surpasses ChatGPT.

Then they released 3.5 Sonnet, which is a less powerful model but is very good at writing code.

Last week, Anthropic announced that their latest version of Claude can now control your computer - bringing us a step closer to AI Agents that can automatically complete any task.

The tool is able to type, move the mouse, and click on buttons. It uses screenshots to identify what is being shown on screen and then figures out what the next action should be.

In many ways, that’s quite worrying - especially for office workers who use computers on a daily basis.

While this tool is somewhat expensive to use and doesn’t perform great, it could eventually automate a lot of the work that people do.

Governments around the world are not prepared for this scenario. But they really should be.

If we circle back and focus on today’s capabilities, Anthropic’s new tool suffers from several issues.

Firstly, the security risk is pretty huge. It’s possible that Claude will make a mistake and delete files on your computer.

Or it could copy the wrong text, like someone’s salary, and send it in an email. Yikes.

For now, you should use a secure environment to run the tool and carefully monitor what it is doing. This could be a virtual machine, with no access to confidential data.

It’s only a matter of time until Google, OpenAI, and Microsoft announce their own Agent tools, similar to Claude.

Google releases their AI watermark technology

SynthID is used by Google to identify AI-generated content. It’s one of the best watermarks available, especially for images, as it is stored within the pixels.

For text, SynthID determines the likelihood that tokens will be generated. This is then used to create a unique score, which serves as a watermark.

Google claims that the tool doesn’t compromise the quality, accuracy, or speed of text generation and remains effective even when the text is modified.

However, it’s not as effective when used to text short snippets of text. That’s unfortunate, as it’s unlikely that SynthID can be used to identify AI-generated tweets - the most likely form of misinformation.

Unfortunately, there isn’t an industry standard for watermarks. Many tech companies have instead decided to develop their own, which makes it difficult for developers to integrate them.

With recent AI legislation being passed in California and the EU, companies will soon need to adopt these watermarking tools and flag content as AI-generated.

🛸 FAA issue clearance for air-taxis in US airspace

✍️ Thousands of creatives sign petition against AI data scraping

👤 Meta will use facial recognition to spot scams that feature celebrities

🚗 Qualcomm and Google partner to create software for cars

💸 Cruise lost $435 million this quarter, despite pausing its robotaxis

💰 Waymo raises $5.6 billion in funding

🚁 Joby Aviation launches $200 million public offering

🔓 Apple will pay security researchers $1 million to hack its private AI cloud

🚙 Wayve’s autonomous vehicles have launched in San Francisco

📧 Notion has unveiled their own email client

CrewAI

Continuing with the theme of AI agents, CrewAI is a popular framework that allows you to develop your own AI “team”.

Essentially, you can specify what AI bots you need for a task. Then you give them instructions on how they should act and the bots will work together on the task.

It’s pretty impressive to see. Currently, the company is based between San Francisco and Brazil.

While it’s a small startup with just 16 employees, the company only launched in January of this year.

It already has 150 customers using the tool, with over 100,000 agent executions every day.

CrewAI has raised $18 million from investors and is valued at around $100 million. They were one of the first agent frameworks to be released, but Microsoft is one of their biggest competitors.

AutoGen was released by Microsoft in 2023 and has been publicly supported by their CEO, Satya Nadella.

CrewAI is hoping that it can beat the tech giant and emerge as a critical tool for AI development, similar to LangChain.

This Week’s Art

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

• The startups that are securing megadeals with VCs

• Apple’s new AI features for iOS

• Perplexity’s setbacks with lawsuits from media companies

• Google’s work on an AI that can use your PC

• Why the Pentagon is being ordered to quickly adopt AI tools

• Anthropic’s update for Claude, which can now control your computer

• Google’s watermark technology that can identify AI-generated text

• And how CrewAI are working on a framework for AI agents

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.