Good morning. OpenAI has so far kept its word; three days into “12 Days of OpenAI” and we have seen three new launches/demos.

Monday’s launch brought Sora, OpenAI’s video generation model, to Plus and Pro users of ChatGPT, some nine months after the model was first previewed.

— Ian Krietzberg, Editor-in-Chief, The Deep View

In today’s newsletter:

• 🌎 AI for Good: Modeling wildfire fallout

• 📊 Nvidia falls following announcement of antitrust investigation 

• 👁️‍🗨️ OpenAI releases Sora … finally

• 💻 UCLA’s got a new AI-generated class

AI for Good: Modeling wildfire fallout

In the wake of the deadly 2023 Lahaina, Hawaii wildfire, which killed 102 people, burned 2,170 acres of grasslands and razed thousands of buildings, researchers have sought to model zones of contamination as the recovery in Lahaina continues. 

The details: Here, we’re not talking about language models or generative AI. At the core of the modeling efforts undertaken by scientists at NOAA’s National Centers for Coastal Ocean Science sits an advanced particle tracking algorithm. 

• Such algorithms, which have been leveraged by oceanic researchers for decades, “use a velocity field to estimate the trajectories of simulated particles through space and time,” according to a recent study. 

• The researchers used this algorithm to model where stormwater might have carried ash and chemicals from the fires — they found that marine habitats within 1.25 miles of the shore are at the greatest risk of contamination. 

Why it matters: These results, according to NOAA, “can help guide local fish and shellfish monitoring efforts and inform related seafood consumption warnings.”

The #1 Rated AI Notetaker

Why are 26,000+ companies using Fathom to boost productivity?

• Records, transcribes, and summarizes meetings in less than 30 seconds

• Saves ~20 minutes of work per meeting

• Works with Zoom, Teams, Meet and integrates with your CRM

• Awarded #1 Highest Satisfaction by G2, plus it’s SOC2 and HIPPA compliant

Start using Fathom today (at zero cost).

Nvidia falls following announcement of antitrust investigation 

Shares of Nvidia were under pressure Monday following an announcement that China’s State Administration for Market Regulation would be investigating the firm for a “suspected” violation of China’s anti-monopoly laws, according to a translation of the announcement. 

The details: The potential violation here involves Nvidia’s 2019 acquisition of Mellanox for $6.9 billion, a move that was approved by Chinese regulators in 2020. 

• The investigation shortly follows the escalating technological trade war between the U.S. and China, with both countries recently instituting export bans against the other for materials related to semiconductor production. “Nvidia wins on merit,” an Nvidia spokesperson told me in an emailed statement. “We are happy to answer any questions regulators may have about our business.”

• Nvidia, whose stock has spiked some 180% this year alone, is no stranger to antitrust investigations. It comes with the territory, considering Nvidia’s dominance, not just in the GPU chips, but also in Cuda, the software needed to use said chips. Mizuho Securities estimated in June that Nvidia controls between 70% and 95% of the market for chips used to train and deploy AI models. 

Antitrust regulators in the U.S. Department of Justice and the European Union both have ongoing investigations into Nvidia. 

“I think what we have here is really more of a macro, geopolitical strain than a real probe,” Daniel Newman, CEO of the Futurum Group, a tech research firm, said. “But Nvidia’s so successful that they’re going to get probed by every regulator in the world.” 

Shares of Nvidia retreated around 2.5% on Monday.

Never Research a Company Again

Bounti’s AI saves over 10 hours/week of painful account research and meeting prep, empowering you and your team to sell more effectively through every channel and stage of the funnel.

Deliver the right message to your ideal buyer—driving real engagement at a fraction of the cost of other solutions.

Research: Get a cheat sheet for every account that matters to you. In minutes, you’ll have everything you need to make your next conversation the reason they buy.

Create: Our AI will generate custom pitches and any type of content you want, all fully customizable to what you need to land your deal.

Engage: Have the confidence to win every conversation with detailed messaging and position that your buyers actually care about.

Get Started For Free Today

• Reddit on Monday began rolling out a new, AI-powered search feature dubbed “Reddit Answers.” Unlike other AI search platforms, this one will only surface information within Reddit … hard to tell if that’s a good thing or not.

• A study into the carbon emissions of the websites of AI tools found that ChatGPT emits 260,000 kilograms of carbon dioxide every month (this is the equivalent of more than 260 flights from New York to London).

• Microsoft new sales pitch for AI: Spend less money on humans (The Information).

• Hinge CEO: Searching for a partner in real life won’t ‘hold a candle to’ meeting online with AI (Semafor).

• Friend’s AI chatbots have issues — and they want your help (The Verge).

• 404 Media objects to Texas AG’s subpoena to access reporting (404 media).

• VCs love AI, but not consumer AI. Here’s why (Pitchbook).

If you want to get in front of an audience of 200,000+ developers, business leaders and tech enthusiasts, get in touch with us here.

• Research Scientist: Anthropic, Seattle, WA

• Research Scientist: Otter AI, Mountain View, CA

OpenAI releases Sora … finally

OpenAI, on the third of its “12 Days of OpenAI” series, on Monday (finally) launched Sora, the much-hyped video generation model that the startup first previewed in February of this year. 

The details: OpenAI has spent the past few months developing Sora Turbo, an iteration of Sora that, according to the company, is “significantly faster.” This video generation system launched yesterday as a standalone product, freely accessible to ChatGPT Plus ($20/month) and Pro ($200/month) users. 

• Users can start generating videos of up to 20 seconds in length; they can also turn stills into videos and extend, remix or blend existing content with text-based prompts. Plus users can generate up to 50 videos at 480p resolution per month; Pro users get “10x more usage” and higher resolution. 

• As both OpenAI and YouTuber Marques Brownlee noted, the model has “many limitations.” Particularly, it struggles with object permanence and is unable to represent a realistic iteration of real-world physics. 

Still, the results are impressive. OpenAI said that all Sora-generated videos will include metadata to track content provenance, as well as visible watermarks by default. The company said that it is blocking the generation of child sexual abuse materials and sexual deepfakes, though added that it will soon allow users to upload and use real images of real people. 

• OpenAI did not provide details on the data used to train Sora, nor did it provide information regarding the energy cost and carbon emissions associated with the training and deployment of the model. 

• The company declined to comment on this, instead acknowledging the energy intensity of training and deploying generative AI models and saying it is always working toward improving efficiencies …

OpenAI’s system card calls out the potential for misuse, here, specifically around deepfake harassment and realistic-looking misinformation. The company says it has established a network of moderation systems, including preventing the generation of known public figures and minors, to protect against these risks. 

The release of Sora comes after other developers spent the months-long lag between OpenAI’s unveil and release of Sora to design and launch their own competitors; Meta, for instance, released an ‘open’ video generation model in early October.

UCLA’s got a new AI-generated class

A comparative literature course being offered at the University of California: Los Angeles next year is skipping traditional lectures and textbooks in favor of AI generations. In UCLA’s own words, “the textbook: AI-generated. Class assignments: AI-generated. Teaching assistants’ resources: AI-generated.”

Proud, confident and all-in on AI.

The details: The course was built around Kudu’s AI learning platform; the professor of the course — Zrinka Stahuljak — provided Kudu with course notes, PowerPoint presentations and YouTube videos she had used to teach the course in the past. 

• “With significant human input and thorough human review of the material,” Kudu synthesized that input into an AI-generated textbook, complete with AI tools designed to help the students. The resulting e-textbook will be available to students for $25. 

• UCLA said in a statement that, unlike ChatGPT and other generative AI tools, when students ask Kudu questions, the system’s answers “will draw information only from the resources Stahuljak has uploaded.”

The textbook generation process took several months to complete, owing to the time needed to review and edit the results. 

According to Stahuljak, the idea isn’t to botsource her lecturing time and expertise; it’s instead to instruct in a more customizable way: “Because the course is a survey of literature and culture, there’s an arc to what I want students to understand,” she said. “Normally, I would spend lectures contextualizing the material and using visuals to demonstrate the content. But now all of that is in the textbook we generated, and I can actually work with students to read the primary sources and walk them through what it means to analyze and think critically.”

The system can spit out AI-generated lesson plans and writing tests, all intended to ease up the professor’s time to, I suppose, teach?

Kudu — which has already been deployed in science courses at UCLA — was developed by UCLA professor Alexander Kusenko. This represents the first humanities course at UCLA to use Kudu. 

Where to begin? 

Let’s start with the AI-generated cover of the AI-generated textbook, which includes what appears to be a family tree of Romance langauges. But at closer inspection, the words — in decreasing visibility — are quite simply gibberish: “that banner at the top feels like it should read, ‘From vernacular Latin to an Evolution on Romance Languages,’ which is already odd, but it actually reads, ‘Of Nerniacular Latin to an Evoolitun on Nance Langusages,’ which is not English or any other language I know,” historian Bret Devereaux wrote. 

Not super convincing when the cover art couldn’t even get things right. Hallucinations and other mistakes in the book itself seem all too likely, something that could lead to some thoroughly confused students.

Now, how about the idea itself, of essentially using a customized textbook instead of lecturing? 

From personal experience, a lot of students simply don’t buy the textbook; the lecture is where they choose to get that information. And, as several professors wrote on Reddit in response to this, it’s really not clear what this adds to pre-existing, customized and standardized lecture notes, slides and textbooks, a trio of tools professors have been using for years. Other than, of course, an unknown and likely not insignificant quantity of carbon emissions. 

I don’t see the value of something like this, especially at the price tags associated with college education in the U.S. And I worry that these kinds of integrations, shoved uncritically and unceremoniously into educational environments, will not include an education about the context and dangers around today’s AI tools, setting up an entire generation for a future of uncritical, unquestioning adoption. 

Generative AI lacks the reliability needed for use in education. Universities shouldn’t be in such a rush to adopt this tech in these flawed iterations. It serves no one.

🤔 Your thought process:

Selected Image 1 (Left):

• “Image 2 seemed to have random stems not attached to certain strawberries.”

Selected Image 2 (Right):

• “Image 1 has berries that are bending/ squishing over top of each other.”

💭 A poll before you go

Thanks for reading today’s edition of The Deep View!

We’ll see you in the next one.

Here’s your view on Grok:

Half of you don’t use Grok, or aren’t on Twitter. The rest were pretty evenly split between “it sucks,” “it’s great” and “it’s okay, but there are better options out there.”