Good morning. It’s finally happened. OpenAI has secured nearly $7 billion in new funding at a post-money valuation of $157 billion (it was last valued at $86 billion).

The ChatGPT startup is now one of the largest private companies — by valuation, but certainly not by revenue — in the country.

We’ve got the details below.

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

In today’s newsletter:

• 🧠 MBZUAI Research: Themes, fables and natural language processing

• 💰 OpenAI now valued at $157 billion

• 🏛️ German court dismisses copyright lawsuit

MBZUAI Research: Themes, fables and natural language processing

One of the significant limitations of large language models (LLMs) is that, though they might be referred to as artificial intelligence, the systems themselves aren’t intelligent. The architecture performs predictive text generation at scale, meaning that reading comprehension and context-based inference remain out of reach. 

Recent research from the Mohamed Bin Zayed University of Artificial Intelligence aims to figure out just how far out of reach it is. 

The details: Researchers created a dataset — called EduStory — designed to test interpretative and inferential comprehension in language models. The dataset is loaded up with “educational” stories, like fables, which have a very clear theme or lesson. 

• The researchers explored four key areas: theme keyword identification, story-theme matching, reading comprehension and theme generation. The models performed poorly in matching pre-defined theme keywords to specific aspects of the stories, something the researchers said is likely due to a lack of relevant training data. 

• Fine-tuned language models performed well in the other three categories. 

Models remain limited to the quality and scale of their training data. 

This work represents the first dataset specifically designed to evaluate theme interpretability in language models, something that the researchers said will support future work on this “challenging and significant” task. 

To learn more about MBZUAI’s research visit their website.  

• Peak is approaching, but is your customer service team ready?

  • DigitalGenius is the AI Concierge used by Reebok, Olipop, Honeylove, and AllSaints to keep customers happy by fully resolving queries in seconds, at any time and in any channel. Speak to us today to be fully ready for peak.*

• Boost your software development skills with generative AI. Learn to write code faster, improve quality, and join 77% of learners who have reported career benefits including new skills, increased pay, and new job opportunities. Perfect for developers at all levels. Enroll now.*

• OpenAI, Italian state lender CDP team up for AI use in startups (Reuters).

• Someone put facial recognition tech on Meta’s smart glasses to instantly dox strangers (404 Media).

• Tesla stock slips after EV maker misses estimates on deliveries (CNBC).

• America’s emerging AI leaders: Amazon, Walmart and Wells Fargo? (Semafor).

• How OpenAI is keeping the startup flushed with cash (The Information).

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.

OpenAI now valued at $157 billion

After weeks of rumors, OpenAI said Wednesday that it has successfully closed a $6.6 billion funding round. The company said that it is now valued at $157 billion. 

The details: While OpenAI didn’t disclose the names of the investors who participated in this latest round, CNBC reported that the round included participation from Thrive Capital, Softbank, Nvidia and Microsoft. Thrive reportedly committed $1 billion.

• OpenAI said in a blog post that 250 million people use ChatGPT every week. 

• “The new funding will allow us to double down on our leadership in frontier AI research, increase compute capacity and continue building tools that help people solve hard problems,” OpenAI said.

This — which makes OpenAI one of the most valuable privately-held companies in the country — comes despite OpenAI’s lack of profitability; the company, according to CNBC, earned $300 million in revenue last month, but expects to lose $5 billion this year. 

It also comes amid the departure of more key executives, namely including former CTO Mira Murati, and internal considerations to restructure the company into a for-profit enterprise. 

Axios reported that, if OpenAI doesn’t restructure into a for-profit within the next two years, investors can ask for their money back.

German court dismisses copyright lawsuit

German photographer Robert Kneschke discovered last year that “heaps” of his images had been used — without his consent, knowledge or compensation — to train generative AI models through the LAION-5B dataset, which is owned by the Large-scale Artificial Intelligence Open Network (LAION) non-profit and contains some 5.8 billion images. 

Kneschke first asked LAION — whose datasets have been used for training by companies including Stability AI — to remove his images from the dataset. When they refused (and charged him nearly $1,000 for making the request), he filed suit against them. 

What happened: The Hamburg Regional Court recently dismissed the case. 

• The court ruled that LAION’s use of Kneschke’s images meets the standard for copyright infringement exception under section 60(d) of German copyright law. This section establishes copyright exceptions for the purposes of scientific research. 

• “The mere fact that individual members of the defendant, in addition to their activities for the association, also pursue paid activities with such companies is not sufficient to attribute the activities of these companies to the defendant as his own,” the judge said. 

The legal firm representing Kneschke said in response that “the court assumed that the defendant was engaged in research,” adding that this “alleged research has not yet been proven in the slightest.” 

“The court sets the bar for this copyright barrier very low,” they wrote. “As a result, this would mean that it would be irrelevant who subsequently uses the created training data (commercially and non-commercially) as long as the copyright-relevant act of reproduction is carried out in advance by an association such as the defendant.”

Kneschke said that he is considering appealing the decision.

It remains to be seen whether this case will actually establish any precedent regarding global copyright law. Dissimilar to the other ongoing class-action cases faced by OpenAI, Stability and the like, the dismissal of this case dealt not with the copyright technicalities of training an AI model, instead focusing on research exceptions. 

If it holds up after appeal, it still doesn’t answer the question of whether training constitutes copyright infringement, and it doesn’t answer the question of whether for-profit corporations have access to those same exceptions. 

“I don’t love the decision but I don’t see it as justifying commercial use,” cognitive scientist Gary Marcus said. 

I think the case might have gone differently if Kneschke had sued Stability instead of LAION. 

🤔 Your thought process:

Selected Image 2 (Left):

• “Paragliders would never be so close together and in a single line as in image #1, it's too dangerous.”

Selected Image 1 (Right):

• “Fooled by the apparent lower altitude, which seemed more realistic.”

💭 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 OpenAI’s structure:

A third of you said OpenAI was bad even with the non-profit; a quarter of you said the restructuring isn’t a good thing. 20% of you don’t care and 13% think it’s great.

So bad:

• “A transition that probably has been under the radar for longer than has been reported.”

Something else:

• “For-profit tends to accelerate progress faster than non-profit, so time will tell.”