Beginning with AI news fresh and hot-out-of-the-oven.

In a big upset for companies like Meta and Google, who are investing and accelerating the development of AI, the state of California voted to pass legislation that requires these companies to test the safety of their programs and create security measures to ensure they can't be manipulated or go rogue. Not a hard ask, right?

But the opponents of the legislation, mainly big tech companies, are up in arms. They say that the rules will stifle AI innovation. However, the bill applies only to exceptionally powerful AI models with operation costs over $100 million. Additionally, the bill asks that AI developers implement a "kill switch" to shut the model down in case of an emergency. We ask again, not a hard ask, right?

Why is the bill important? It’s going to answer questions like: Could AI turn into an evil robot overlord and take over the world? Does making rules for tech always throw a wet blanket on cool new inventions? But seriously, Washington is in a deadlock over bringing in any real AI regulation to the country at a national level. The state of California—home of Silicon Valley and where a lot of AI companies set up shop—along with other states are taking the baton, so to speak, to regulate AI somehow in their own states.

What’s next for the bill? First introduced in February 2024 and passed in May by the California State Senate, Senate Bill 1047, authored by Scott Wiener (D-San Francisco), will now go to the Assembly Appropriations Committee in early August. If it passes that committee, it will go to a full floor vote later that month.

…And another one bites the dust

Time magazine has signed a deal with OpenAI to allow the AI company to use its archive of articles to train ChatGPT. The chatbot, when citing information from a Time article, will link back to the source on the Time website. The magazine has also inked a deal with ElevenLabs to create audio-accessible content for readers, announced ElevenLabs in a blog.

This comes after many other news publications—Associated Press, Vox Media, The Atlantic, and Semafor, among others—also partnered with OpenAI to train its models.

OpenAI released a new model called CriticGPT…

…and yes, it’s supposed to do exactly what its name implies. To criticize the work of ChatGPT. The model will catch errors in ChatGPT's code output.

“We found that when people get help from CriticGPT to review ChatGPT code they outperform those without help 60% of the time,” said OpenAI in a blog.

Who would have thought that Toys-R-Us is still around?

Especially after a major chunk of their market share has been encroached upon by the likes of Amazon, Walmart, and Target. But apparently, they are, and they have made a sort of comeback into the sphere of relevance by using OpenAI’s text-to-video model Sora for an advertisement.

The said advertisement has garnered mixed reviews. While some find it jarring, others are exclaiming that the age of AI-powered ads is finally here. But we, of course, want you to form your own opinion.

Google also made a bunch of announcements this week. One of them is that Gemini 1.5 Pro has a 2 million context window. This means it can consider up to 2 million tokens—words, parts of words, punctuation—from the input text at once.

To put that into context, for example, you can upload the entirety of George RR Martin’s A Song of Ice and Fire series and have conversations with Gemini about it. It will be able to remember and reference details from a much earlier part of the book.

In the follow up to our big story last week, Amazon is finally looking into the charges leveled against Perplexity AI, a company Jeff Bezos has heavily invested in, that is scraping content from websites without permission.

This comes after a WIRED investigation suggesting Perplexity violated site paywalls, a charge the company has denied. Perplexity uses Amazon Web Services (AWS), which prohibits abusive practices.

In other news, a worthy competitor to Meta’s Ray-Ban smart glasses emerges. Solos has announced AirGo Vision—which has a camera and a microphone and is powered by OpenAI’s latest offering—GPT4o.

We often find ourselves glued to our devices, frantically searching for information as if we are on a caffeine high. What AirGo Vision (is the name inspired by Apple Vision Pro?) promises is a hands-free future, like a diligent butler who anticipates our needs.

It can recognize people, objects, and landmarks, which is similar to Meta’s Ray-Ban sunglasses. The smart glasses not only display maps as you wander through the city but also send texts on your behalf. Need to tell your boss you're "working from home" while you’re actually at the beach? Done. You can read more about it here.

Lastly, after facing backlash last week from photographers for tagging real pictures with ‘Made with AI’ label, Meta is now changing the label to ‘AI info.’ The company said that its previous approach was too narrow as it did not provide information as to whether the image with the tag is fully created by AI or the creator merely used AI-powered tools in the editing process, said the company in its blog.

And last, here’s wishing all the readers of AI Logs a happy 4th of July. Now go fire up the barbeque and watch some drone-powered fireworks.

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

• In Zemeckis' new movie, camera never moves, AI used to de-age actors

  The 104-minute story is based on a 2014 graphic novel by Richard McGuire.

• Ukraine: AI drone swarms that recognize Russian army uniforms unveiled

  While the drones can operate autonomously, they order to fire must be made by a human.

• AI bias solved? New study proposes radical fairness framework

  A new study proposes a way to avoid biases in decisions made by AI like deciding who gets a new mortgage or hired for a job.

• World’s 1st smart glasses with GPT-4o identify objects, answer queries

  Solos smart eyewear announces AirGo Vision, the first glasses to incorporate GPT-4o technology.

MUST-READ

World's fastest AI supercomputer can be 30x more powerful than Frontier

In the realm of supercomputers, there are efforts to develop a high-performing machine that would be 30x more powerful than the current fastest supercomputer, Frontier.

As per reports, AMD, a California-based semiconductor company, has been approached by a client to create an AI training cluster of 1.2 million GPUs.

Only a few thousand GPUs connected via high-speed interconnect across several local servers make current training clusters. Notably, Frontier has 37,888 Radeon GPUs.

AI PICTURE OF THE WEEK

As June came to an end, and with it the pride month, a photo on X caught our attention, as well as that of 20 million others. A non-profit called Pride, going by the handle @prideukorg on X, posted an AI-generated image honoring the deceased who had been part of the LGBTQI+ rights movement.

However, people felt that the organization should have instead used human-created art and specifically art created by queer people. Visual artist Karen Hallion drove the point home for many when she reposted @prideukorg’s post with her own art, which honored influential figures who have made significant contributions to civil rights and the fight against LGBTQ+ oppression

On to the results of the previous week’s AI picture challenge. We asked you to differentiate between photos of a real chicken burger and an AI-generated one.

The results were fascinating: 46% of you chose the AI-generated image as the real one. And to the majority of you who got it right, give yourselves a pat on the back—the first image is the real one, and your burger-spotting skills are top-notch.

We also went through what you all had written about the images. A respondent who selected the AI-generated image as the real one said this about the real image: “The bread looks fake. The tomatoes look fake. The chicken looks like Kimchi Cabbage.” Ah, mate! :/

Thank you to everyone who participated and shared their thoughts. We will do more of these challenges in the future.

AI TOOL OF THE WEEK

In most newsrooms, there is a team of dedicated graphic designers who create images and graphs to explain articles better. But if you’re a blogger or a content creator who neither has the skills to create an image nor the time to click a photo, BannerGPT can come to their rescue. 

1. Accessing BannerGPT

• Head over to BannerGPT’s website.

• There will be an option labeled ‘Get started for free’ on the landing webpage

• Once you click on it, the website will ask you to log in using Google

• After creating an account, you will arrive at a web page titled ‘Blog Banner Image Generator’

2. Crafting your prompt

• BannerGPT gives you the option of adding an overlay text to your image, so you can type in the text you want to appear on your image

• In the section right below it, there is a window to paste your text or your blog, on the basis of which the image will be created.

3. Choosing an artistic style

• BannerGPT gives you five options for the style of the image: realistic, sketch, cartoon, painting, and ‘surprise me.’

• It also gives you five aspect ratios to choose from.

5. Generating the image

• Once you're happy with your prompt, click on the "Generate" button.

• Two banner images will be generated based on the content you provided.

6. Downloading and sharing

• Once processing is complete, you can preview the generated image

• If you like it, you can download the files.

That’s all folks! Feel free to reach out at [email protected] to give your suggestions as to what you’d like to see more or less of.