Good morning. Perplexity attracted yet another lawsuit on Monday. The New York Post and Dow Jones filed a joint lawsuit against the AI search startup alleging massive copyright infringement.

The litigation here is interesting considering that Dow Jones parent News Corp inked a massive partnership with OpenAI back in May. Clearly, News Corp isn’t allergic to lawsuits.

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

In today’s newsletter:

AI for Good: Underwater robots

While artificial intelligence is good for parsing vast quantities of data, it needs a way to access that data. And when it comes to oceanic research and marine conservation, there is a vast, unexplored sea of data, information and observations waiting to be gathered. 

What happened: Researchers at the Minnesota Interactive Robotics and Vision Laboratory have developed an AI-powered solution, called the Low-Cost, Open-Source, Autonomous Underwater Vehicle, or LoCO AUV. 

The project has received nearly $1 million in funding from the U.S. National Science Foundation. 

Why it matters: "Our project is about making underwater robots more effective tools for scientists and conservationists,” Junaed Sattar, a lead investigator on the project, said in a statement. “With improved vision and localization, these robots can better understand and protect our underwater environments, which is crucial for ecological balance and human prosperity.”

Meet your new AI assistant for work

Sana AI is a knowledge assistant that helps you work faster and smarter.

You can use it for everything from analyzing documents and drafting reports to finding information and automating repetitive tasks.

Integrated with your apps, capable of understanding meetings and completing actions in other tools, Sana AI is the most powerful assistant on the market.

Microsoft, Google and AI Agents

Just a few weeks after announcing a host of new upgrades to its Copilot software, including the ability for customers to build custom “AI agents,” Microsoft said Monday that it is rolling out these agentic capabilities to the public (in a public preview) next month. 

The details: Microsoft additionally announced that it is adding 10 new autonomous agents to its reserves for customers to choose from. 

At the same time, Honeywell — one of the Copilot testimonials Microsoft mentioned in its blog — announced a partnership with Google Cloud to build AI agents trained on Honeywell’s Internet of Things platform Honeywell Forge. 

Honeywell customers will be able to build customized agents using Google’s tech to automate tasks and assist with the resolution of maintenance issues by 2025. 

The thing that neither company mentioned in their promotional announcements is the fact that these agents are based on the language models they have developed, and language models are known to exhibit biases, hallucinations and general reliability issues. 

Microsoft’s only concession to this point was that customers could build guardrails into their agents to govern their processes. 

NVIDIA, Databricks, HP, and more share GenAI tips

Learn how NVIDIA, Databricks, Twilio, HP, and ServiceNow get their GenAI apps into production!

Don't miss GenAI Productionize 2.0 for GenAI best practices, including:

Free Registration

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.

IBM releases small new AI models (and actually tells us the details)

IBM on Monday released Granite 3.0, the third generation of its Granite language model family. 

The details: The models, which specifically target enterprise users, have exhibited strong performance across retrieval augmented generation (RAG), classification, summarization and tool use. 

What makes this different: Developers are always releasing new AI models. But often, details about these models are slim to nonexistent. That’s not the case here. Both in an extensive research paper and technical report and responsible use guide, IBM provided details on its training data and data curation process, in addition to details on the energy use and carbon emissions associated with the models. 

Training the family of four models took a total of 1.2 million GPU hours, resulting in the consumption of 945.6 megawatt hours (MWh) of electricity and the emission of 368.7 tons of carbon dioxide. For context, the average U.S. household consumes about 10.5 MWh of electricity each year, and per-capita carbon emissions in the U.S. amount to 14.9 tons annually. 

Since the models are “relatively lightweight, they can be run on a single GPU,” meaning the energy use of operation shouldn’t be enormously consumptive. 

In terms of training data, the models were built on a mix of synthetic data and scraped internet data with “permissible licenses.” That data was then carefully analyzed and cleaned before being used for training. 

Europeans would let an AI vote for them

IE University’s Center for the Governance of Change (CGC) unveiled its annual European Tech Insights report on Monday. A key focus of the report centered around increasing integrations of generative artificial intelligence into societal institutions. 

And Europeans are really mixed about the whole thing. 

AI in elections: Around 40% of Europeans, and nearly 75% of public servants, are concerned about the misuse of generative AI in elections, with a third of those surveyed saying that they believe AI has already influenced their voting behavior. 

AI in public integrations: 60% of those polled had no idea that their governments are actively using AI to deliver public services. 

And right around 75% of Europeans support the use of AI for police and military operations, “e.g. using facial recognition and biometric data for surveillance.”

At the same time, 64% of those polled don’t support the use of AI to determine eligibility for parole and 50.6% are against the use of AI to determine immigration visa eligibility. 

And 70% support the passage of laws to limit automation in order to protect their jobs. 

“European citizens’ attitudes toward adopting emerging technologies can be summed up as ‘yes, as long as core values are respected,’” the report reads. “This is evident in their cautious optimism about AI in public life, including elections, where many trust its potential but worry about misuse.”

The report surveyed 3,006 adults across Estonia, France, Germany, Italy, the Netherlands, Poland, Romania, Spain, Sweden and the U.K. 

AI literacy is clearly lacking and needs to improve. 

The juxtaposition here of people prepared to “trust” AI while worrying about misuse is a hard one to grapple with. I see the potential of AI; I do not trust its output. 

To trust an AI tool to vote on their behalf, while at the same time acknowledging concern about electoral misuse, seems based on anything but an understanding of the technology. 

Likewise, to support the use of AI in a variety of government operations, including military and police operations, while wanting laws that prevent automation to protect their jobs ignores the real, active harms and threats posed by the integration of hallucinatory, biased systems in high-stakes environments. (Also, I find it unlikely that governments will pass such laws while taking advantage of these systems themselves). 

There is a risk here of regulatory efforts being swayed by civilians and government officials who are vastly misinformed about the genuine capabilities, promises, risks and active harms of current AI, something that could enable the entrenchment of faulty systems at scale, increasing threats of unfettered surveillance, enhanced predictive policing and ingrained biases and critical mistakes in crucial operations. 

🤔 Your thought process:

Selected Image 2 (Left):

Selected Image 1 (Right):

💭 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 fake news:

The bulk of you (50%) just cross-check your information across a few different trusted sources. It really is that easy.

The rest aren’t really sure what to do about it.

Cross-check:

Something else: