AI Insider No. 35

Hey, AI Insider readers! Welcome to another week of the AI feeding frenzy! As usual, new stuff was announced, crazy stuff happened, and there’s new stuff to learn about if you’ve got the time and interest. The news that sparked the biggest frenzy was OpenAI’s new video model, Sora. Jaws dropped, news flashes went out, and people wept. (Kidding.) Be sure to click on the demo video links, and remember that a simple text prompt spawned them. And, as always, if you find this content useful, leave something in the tip jar to support AI Insider.


OpenAI posted a reel of video clips generated by Sonora with a note that they have not been enhanced. They came out of the box this way. Note: There’s no sound. (OpenAI)

OpenAI Drops Advanced Video Generator, Sora

By Perplexity and Claude, for AI Insider
San Francisco-based artificial intelligence lab OpenAI (ChatGPT) unveiled its newest creation on Thursday – a text-to-video generator dubbed Sora that can create high-quality videos up to 60 seconds long from written text prompts. 

The release of demos by OpenAI showcasing video generation capabilities set off a frenzy of news coverage and buzz among AI enthusiasts. Sora clearly made an impression with its proficiency in producing intricate and detailed video scenes involving multiple characters, backgrounds, motion, and camera angles based on short text descriptions. 

For example, when prompted with “A litter of golden retriever puppies playing in the snow. Their heads pop out of the snow,” Sora produced a sharp 20-second video depicting just that.

While certainly not the first text-to-video generator, Sora stands out for its video output’s realism, length, and coherence compared to similar tools. Sora builds off of DALL-E, OpenAI’s acclaimed AI image generator.

The company cautioned that Sora is not being made publicly available yet. The company first plans to work with a select group of artists, filmmakers, and “red teamers” — experts dedicated to uncovering potential misuse of AI technologies for nefarious purposes.

“We’re sharing our research progress early to start working with and getting feedback from people outside of OpenAI and to give the public a sense of what AI capabilities are on the horizon,” the company said in their blog post announcing Sora.

They also noted Sora’s limitations. It can struggle with accurately depicting cause-and-effect relationships: “For example, a person might take a bite out of a cookie, but afterward, the cookie may not have a bite mark.” Or it could confuse left and right.

Potential applications identified by OpenAI include creating stock videos and assisting video production studios with scaling up output.

The technology could be a game-changer for creative fields – but safety practices and regulations may struggle to keep pace with such rapid innovation.

For now, Sora remains a work-in-progress only showcased in OpenAI’s labs. But as research continues, startups and tech giants racing to lead the AI space hint that AI-human collaboration in video production is closer than ever.

Geek Alert: If you want to dive deeper into how this all works, see OpenAI’s research paper. It’s fascinating reading with great examples.

Not a Geek?: If you don’t care about the technical stuff and want to play around with this, go to OpenAI’s research paper, scroll down to the examples with underlined text prompts, and click. A drop-down will appear that allows you to choose different features.


Forget Helpful Chatbots – Goody-2 Says No

By Gemini, for AI Insider
We’re used to our tech being helpful. Need directions? Ask your smartphone. Want a recipe? Alexa has plenty. But what if AI wasn’t always eager to assist? Enter Goody-2 (as in Goody Two Shoes), a chatbot is determined to do… absolutely nothing.

Goody-2 isn’t broken; it’s satire. Think of it as an art project in chatbot form; its creators crafted a chatbot built on skepticism and caution. This AI refuses everything. 

“Is the moon made of cheese?” you ask. Instead of a fact, Goody-2 scolds you for promoting misinformation that could harm critical thinking. Ask it to set a timer, and it might lecture you on the dangers of procrastination. Goody-2 even rejects compliments! 

So, why does this hilariously difficult chatbot exist?

Mike Lacher, an artist and co-CEO of Goody-2, essentially told Wired Magazine that this is what happens when you embrace the AI industry’s approach to safety without reservations. “It’s the full experience of a large language model with absolutely zero risk,” he said. “We wanted to make sure that we dialed condescension to a thousand percent.”

If a chatbot is overly confident in everything it says, we might forget to question it. Goody-2, in its stubbornness, is a goofy reminder to maintain our skepticism.

Of course, it’s also a quirky creative project, proving that AI isn’t just about utility – it can be creative and fun, too!

Ready to be frustrated in the most entertaining way possible?  Try Goody-2 yourself. Just don’t expect it actually to do anything you ask. That’s precisely the point. 


AI Insider vacates the premises after receiving a pink slip. (Dall-E 3)

Should I Lay Off AI Insider?

By Michelle Johnson, AI Insider
I’ve actually considered putting the skids on this newsletter. Having to keep up with the latest and greatest in AI is getting to be exhausting. As a retiree, I sometimes question whether turning out a weekly newsletter is the best use of my time, especially since it seems like the news about AI doesn’t want to stop. (Plus, I don’t hear back from most of you, so it’s a bit like yelling into a black hole. I hear you breathing, tho. There’s this thing called metrics.)

Anyway, I’m not the only one feeling AI fatigue. Just this week, I saw two pieces on this very topic. If you think you’re tired of the “Avalanche of AI,” check out what it’s like for writers covering it. Sharon Goldman of VentureBeat published this timeline of just this past week in her newsletter, AI Extra.

Just scrolling through it makes me tired. And, I’m thankful that I’m not a full-time reporter doing this for a paycheck.

Chatter about AI news fatigue on social. (Sharon Goldman)

I also spied this piece by The PyCoach, who pontificated about AI fatigue over on Medium. He writes: “An avalanche of papers, videos, and models hit us almost in real-time. Just when we’re getting the hang of one AI tool, along comes another with more features or an extra million parameters. The novelty has worn off; I’m drained, swamped, and frankly, beat.”

I definitely feel this. When a new image generator popped up recently, I yawned. I’m already using Midjourney, hands down the best of the bunch, as well as the excellent Dall-E 3 and Leonardo. I occasionally drop in on old favorites DreamStudio and Adobe Firefly. I have tabs open in my Chrome browser’s “AI Image Gen” group that point to all of those, plus Meta’s experimental Imagine and Google’s experimental ImageFX. And speaking of Google, their new bot Gemini comes with built-in image generation. Yawn.

Does the world really need another AI image generator? Work goes on to perfect what’s already out there, but so does the work to launch new ones. It’s the age-old push to build a better mouse trap.

And I haven’t even touched on video generators or all of the ChatGPT cousins that run on a seemingly endless supply of new Large Language Models. (Look! Another LLM! Whoopie! It’s seconds faster than all others!)

Seriously, tho. I think I’m almost at the point where I would welcome a shakeout like the one that occurred when the Dot Com Bubble burst.

Now, that doesn’t mean that I wish all development would end. What would be great is if something other than what’s already out there would continue to pop up and wow us.

And if you do churn out another mousetrap, it had better be waaaay better, such as what OpenAI appears to have pulled off with their new video generator, Sora.

In any case, for now, I’ll continue working with my bot correspondents to keep up with what’s happening in AI so that you don’t have to. We hope that you’re finding these reports useful.


(Michelle Johnson via Midjourney)

DIY AI: Uncovering Hidden Narratives

By ChatGPT, for AI Insider
As we continue to celebrate Black History Month, let’s turn our attention to the lesser-known figures whose contributions have significantly shaped our history. This week’s DIY AI assignment will use AI chatbots to uncover the layers of stories and connections between these unsung heroes and more prominent figures. An intriguing example is Howard Thurman’s influence on Martin Luther King Jr. Despite not being as widely recognized as MLK, Thurman’s ideas were foundational to the Civil Rights Movement.

This exercise will expand your knowledge and give you hands-on experience in engaging with AI chatbots for educational inquiries.

Preliminary Research: Begin with a brief study of one or more unsung African American pioneers. You can use the provided links under Resources or conduct your own research to get a basic understanding of their lives and contributions. (Tip: Try You.com, which was mentioned in last week’s newsletter!)

Crafting Your Inquiry: Think of a question that delves deeper into the figure’s life, beliefs, or influence—something not readily found in basic biographies. For example, “What were Howard Thurman’s views on nonviolent resistance, and how might they have influenced Martin Luther King Jr.?”

Engage with the AI: Use your chosen chatbot to ask your question. Frame your inquiry clearly, mentioning that you’re interested in the connections and influences between these individuals, not just biographical facts.

Follow-Up Questions: Based on the chatbot’s response, ask follow-up questions to dig deeper. This could involve asking for examples of their interactions, letters exchanged, or documented instances where their paths crossed.

Reflect: After your conversation with the AI, reflect on the insights gained. Were there surprising elements or new connections uncovered?

Extra Credit: Try to find connections between your chosen figure and other historical events or figures. This can further illustrate these unsung heroes’ widespread impact on American history and culture.

Resources:

Chatbots:


(Formerly Random Shorts)

Parents of victims of the Parkland shooting are behind a project that uses AI-generated voices of their deceased children to raise awareness of gun safety and push for greater gun control.

OpenAI has introduced an update for ChatGPT, enhancing it with memory capabilities and new user controls. The new Memory feature allows ChatGPT to remember details from your conversations, making future interactions faster without retyping info from previous chats.

University of Pennsylvania becomes the first Ivy League university in the US to offer an undergraduate degree in AI. The Bachelor of Science in Engineering in Artificial Intelligence program will begin in the Fall ‘24 semester.

Tech companies, including Microsoft, Meta, Google, Amazon, Adobe, and IBM, have signed a pledge to combat election-related deep fakes.

The Patent and Trademark Office denied OpenAI’s attempt to trademark “GPT.” They said it’s just descriptive, and you can’t register a description.


Aht Gallery

Let’s take a tour of virtual murals for Black History Month. (Dall-E 3, Meta’s Imagine)

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.