We’re not saving the strangest AI news for last, so here it is: AI is now targeting mosquitoes and taking them out for you. Are you a mosquito magnet? Then this is good news for you.

The Bzigo Iris is an AI-powered gadget with ‘AI-powered computer vision’ that tracks mosquitoes in your home, highlights them with a laser, and alerts you via a smartphone app. While it doesn't kill mosquitoes itself, it makes it easy for you to swat them using a rechargeable electric swatter. 

Speaking of strange developments over the week, Elon Musk has come out in support of… no not Donald Trump, that’s old news. But he has extended his support to the California AI regulation bill SB 1047, VERY unexpectedly. 

We have discussed at length about the bill in our previous newsletters as well.

SB 1047 aims to regulate AI by holding developers spending over $100 million accountable for the safety of their models. The bill has passed the state Senate and is up for a State Assembly vote soon.

Musk’s support is significant because it comes at a time when most major tech companies like Google, Meta and OpenAI have raised concerns that the bill's strict regulations could slow down AI development and hurt California's competitive edge.

Musk’s support is also significant because he too has a company, xAI, which is developing and releasing AI models–like the recent Grok 2 assistant for X–which will come under the purview of the Bill (if passed into law).

Google DeepMind workers urge company to end military contracts

Around 200 workers at Google DeepMind, Google's AI division, signed a letter urging the company to stop working with military organizations, as per a report by Time magazine. 

The letter reflects concerns that DeepMind's technology is being used in warfare, which the workers believe violates Google's own ethical guidelines for AI. 

This internal conflict highlights a broader dispute within Google, as the AI division has pledged not to work on military projects, while Google Cloud continues to sell AI services to governments, including those of Israel and the US. 

Google’s $250 million deal with California newsrooms

But that’s not all the trouble that is brewing for Google. The search engine giant’s $250 million deal with California to "support" local newsrooms for over five years, has drawn much ire. 

Critics, including journalist unions, are unhappy, saying the deal mostly benefits Google and doesn't do enough to support struggling newsrooms. The deal includes funding for an AI "accelerator" to help journalists, but some worry that AI could harm the journalism industry by increasing distrust in news and potentially replacing human jobs.

Midjourney adds editing of AI-generated images

AI image generation tool Midjourney’s latest update has introduced a Web Editor that allows you to make detailed edits to generated images without starting from scratch. 

This new feature lets users edit specific parts of an image, adjust the canvas size, change perspectives, and even modify the original text prompt. 

The update also includes tools for more precise image editing and improved integration between Midjourney's web platform and the Discord community. This comes as the company is embroiled in ongoing legal issues over copyright claims.

Also, Midjourney announced that it is now available for free. Once again. Users can create up to 25 images during a trial period. Previously, image generation required navigating through Discord, which many found cumbersome. Although, signing up for the free service requires a user to have a Discord or Google account.

Hotshot AI Video Maker came out last week

It seems like every week a new multimodal video model drops, creating a buzz and getting labeled as the next 'Sora-killer.' I have to admit, I'm just as eager about these AI text-to-video marvels and love discussing them with you each week.

And this week, it's Hotshot's moment in the spotlight.

Originally a free AI photo app, Hotshot’s new video model can generate up to 10 seconds of 720p footage and was developed over the past four months. Previous models, Hotshot-XL and Hotshot Act-One, created shorter videos with lower frame rates. 

The latest Hotshot model was built using 600 million clips and many GPUs, facing challenges with hardware failures during intense training.

Anyone can generate three videos a day. Naturally, I had to try it. This is the prompt I gave the model: ‘an Asian Indian woman entering an Indian temple.’

And this is the result:

Speaking of AI text-to-video models, Luma AI has released Dream Machine 1.5, which better understands your prompts and produces higher-quality videos.

Condé Nast partners with OpenAI

Condé Nast, the publisher of magazines like Vogue and GQ, has partnered with OpenAI, the creator of ChatGPT. This deal will allow Condé Nast's content to appear on AI platforms like ChatGPT and SearchGPT, which is yet to be released to the public.

While Condé Nast sees this as a way to embrace new technology and boost revenue, it contrasts with other media companies like The New York Times, which are more cautious about copyright issues with AI.

You can read more about it in a Condé Nast blog.

Apple event next month

Apple sent out invites for a big event on September 9 at Apple Park, where the spotlight will likely be on the new iPhones, along with other gadgets like Apple Watches and AirPods. This event is a huge deal because the iPhone is Apple’s crown jewel, both in strategy and revenue.

The buzz is building around the new features, especially in Apple Intelligence software. The invite was cryptic but intriguing, simply stating, "It's Glowtime," with a neon Apple logo.

More on that in the next newsletter.

Did a friend forward this e-mail to you?

IE₊ SUPPORT INTERESTING ENGINEERING
_{Invest In Science And Engineering}

Enjoy exclusive access to the forefront of AI content, highlighting trends and news that shape the future. Join a community passionate about AI, delve into the latest AI breakthroughs, and be informed with our AI-focused weekly premium newsletters. With IE+, AI reporting goes beyond the ordinary - and it is Ad-Free.

NEWS

• Japan: AI-powered drones to monitor disaster zones and identify criminals
  Drones move faster than police cars or guards, reaching incident site quickly and allowing for prompt action and response.

• UK scientists use AI on 1.6 million brain scans to predict dementia early
  A robust collaboration in the UK has amassed over a million brain scans to develop better prevention tools using AI.

• Dung beetle’s use of Milky Way for navigation inspires new AI sensor for drones
  Taking cues from the humble dung beetle, scientists have created an AI sensor that precisely measures the orientation of the Milky Way.

• Photos: Robot dog achieves 90% accuracy in finding destructive fire ant nests
  In field tests, the CyberDog identified three times more RIFA nests than human inspectors, showcasing its superior accuracy.

MUST-READ

Tesla’s Cortex AI supercomputer revealed by Elon Musk, 350,000 Nvidia GPUs planned

Elon Musk shared a video on Monday showing the inside of Cortex, Tesla’s new AI training supercluster, which is being built at its headquarters in Austin, Texas.

In an X post, the CEO described Cortex as “the giant new AI training supercluster being built at Tesla HQ in Austin to solve real-world AI.”

AI PICTURE OF THE WEEK

A New York Times article this weekend discussed the concept of "model collapse" in generative AI systems. This occurs when AI models are repeatedly trained on their own outputs, leading to a decline in quality and diversity over time. 

The model in question is a type of generative AI that, when trained on synthetic data rather than real-world data, begins to produce increasingly flawed and homogenized results, potentially amplifying biases and reducing the accuracy and utility of AI-generated content.

The article uses the example of training an AI on handwritten digits to illustrate "model collapse." Initially, the AI generates clear, accurate numbers. However, as the model is repeatedly retrained on its own output, the quality of the digits declines. 

After several generations, the numbers become blurry, and eventually, they all converge into a single indistinct shape. This shows how AI can degrade when trained too much on its own data, losing both accuracy and diversity in its output.

This highlights the problem of AI-generated content flooding the internet, making it difficult to discern what's real and what’s not. As AI systems train on their own outputs, a feedback loop can degrade their quality and diversity.

This means AI could become less accurate and more biased over time. To combat this, the article proposes that AI companies need high-quality, human-origin data and better detection methods.

AI TOOL OF THE WEEK

Whenever I have had to give a job interview or meet someone important (like a government official), I can feel anxiety deep in my stomach. I'm waiting, churning, and coiling around. Preparing as much as possible before the meeting helps relieve my anxiety as the day draws nearer. 

I recently discovered an app that offers practice questions, AI-generated cover letters (although I think it's best to write your own), and flashcards. It might not completely eliminate your anxiety, but it can definitely help you get in a few practice runs before the big day. That’s not too bad, right?

It’s called Final Round AI and here’s how it works:

1. Sign Up
   Start by creating an account on the Final Round AI website.

2. After Signing Up

• Once you’re logged in, you’ll land on a page that asks, ‘How can we help you crush your next interview?’

• You’ll get three options: A live interview assistant powered by Copilot, which is only available to Premium users; the second option allows you to practise using mock interviews; and the last option is to build your resume and explore the questions bank.

3. Since I don’t have a Premium membership, I went for the mock interview.

• The website would ask you to fill out the company you want to join and the position you’d be applying for.

• Enter details: Input your resume and any other relevant information.

• Start mock interview: Begin your AI-powered mock interview, where you'll receive real-time feedback on your responses.

• Review and improve: After the session, review the feedback provided to refine your answers for the actual interview.

For more detailed guidance, visit the Final Round AI blog.

You can also watch the detailed tutorial here.