Good morning, and happy Friday.

It’s been a long week. But the weather’s getting warmer and the weekend is almost in reach. I bet Elon Musk, for one, will be glad to see the stock market closed for a few days … $TSLA ( ▲ 5.42% ) 

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

In today’s newsletter:

• 🛰️ AI for Good: Archaeology from space

• 👁️‍🗨️ Study: AI Search still has a problem

• 🚨 Artificial intelligence is fast approaching a major inflection point

AI for Good: Archaeology from space

In 2020, Dr. Iris Kramer launched ArchAI, a startup that leverages a combination of satellite imagery, historic maps and deep learning algorithms to detect archaeological sites. 

The company has since expanded beyond simple archaeology, leveraging its same approach to study the ways in which the land has changed over time, something it is achieving in part by designing a “digital twin of the past.”

• Their idea is that the conservation efforts of today, with an eye on the future, ought to take advantage of robust and reliable historical data; in order to better understand how to protect and conserve the land, we must understand how it has changed. 

• With centuries worth of habitat-related data, conservationists can use the system to compare images of modern and historical habitats, something that can better direct conservation efforts. 

The project was funded by the U.K. Space Agency in 2021. 

Why it matters: “As we face the growing impacts of climate change and biodiversity loss,” ArchAI writes, “there has never been a more pressing need to understand where we’ve come from, in order to build a sustainable future.”

The Future of Coding: How Zencoder is Changing Software Development

Zencoder – AI coding assistant helps developers ship products faster. From code suggestions, autocompletion, unit testing, documentation and chat assistant to building Enterprise Grade Software - Zencoder helps developers effortlessly build production ready solutions, right inside VS Code and JetBrains.

Key features of Zencoder: 

• Zencoder supports 70+ programming languages and integrates seamlessly with VS Code, JetBrains and Android Studio, offering robust support for technologies and frameworks.

• Zencoder offers 20+ DevOps integrations (making us the most integrated coding agent)—including JIRA, Sentry, GitHub, and GitLab—to help you ship faster.

• Our unique Repo Grokking technology helps Zencoder better understand large codebases, generate, test, repair, and refine code, making your workflow smoother than ever. 

• Zencoder has built a unit testing agent for maintainable, high-quality code

• Zencoder provides developers to build their own custom agents that can be tailored to the teams and organizations exact needs e.g. code review agent, code refactoring agent designed to an organization's specific standards.

• Zencoder comes with enterprise-grade security, helping organizations reduce technical debt and ship products faster while being secure. 

With Zencoder’s AI Agents, speed and quality can come together; you don't have to pick one.

Learn more and start your free trial today.

Study: AI Search has a problem

As we talked about recently, the push to leverage generative AI as a leveled-up search engine goes on, with Google beginning to run experiments that would position the company to compete more directly with increasingly popular offerings from Perplexity and OpenAI. 

But the challenge with GenAI search — even when it claims to point to original sourcing — is that these systems are based in an architecture that isn’t exactly ideal for accurate information retrieval; the Large Language Models (LLMs) that power such search engines are prone to making mistakes, a phenomenon colloquially referred to as “hallucination” that, in two years of excitement over transformer-based architecture, has not been solved. 

What happened: Columbia’s Tow Center for Digital Journalism conducted a study that examined the abilities of eight popular GenAI search applications to accurately and consistently output news-related information. 

• The study largely found that chatbots were “bad at declining to answer questions they couldn’t answer accurately, offering incorrect or speculative answers instead.” The study also found that these tools “fabricated links” and cited copied versions of articles; interestingly, these inaccurate citations weren’t any better among those news organizations that have secured content licensing deals with the companies in question. 

• The study ran a total of 1,600 queries based on excerpts from articles through the eight chatbots. The researchers were looking for output to provide an accurate URL, point to the original headline and indicate the name of the publication. 

Combined, the chatbots provided incorrect answers — meaning they failed all three of those qualifiers — to more than 60% of queries. Perplexity answered 37% of queries incorrectly, while Grok 3 answered 94% of queries incorrectly.  

Why it matters: Despite incorrectly identifying 134 articles, ChatGPT “never declined to provide an answer.” More than half of the URLs produced by Gemini and Grok 3 didn’t work; Grok 2, meanwhile, had a tendency to link to a publisher’s home page, rather than the article itself. 

The study, not intended to mimic standard user behavior, sought to challenge GenAI search chatbots with something that is done competently, consistently and trivially by traditional search engines.  

• Google gets personal: Google on Thursday launched a version of its Gemini chatbot with “personalization,” which connects the chatbot to all your Google apps and services. Which means … “Gemini can now tailor its responses based on your past searches.”

• Market update: Stocks took another dive on Thursday, pushing the S&P 500 into correction territory (meaning it has fallen 10% from its previous high). Apple and Tesla, meanwhile, dragged the Nasdaq lower.

• Why the toll road text scam is out of control across the U.S., and Apple, Android can’t do anything to stop it (CNBC).

• Why AI isn’t giving Salesforce a boost (The Information).

• Elon Musk’s digital coup (Wired).

• Snapchat is rolling out AI-powered video lenses (The Verge).

• How a Chinese battery factory sparked a political meltdown in a small Michigan town (Rest of World).

Your Next Team Member is Just a Click Away

Finding the right talent doesn’t have to be a hassle. Our professionals are real, globally based experts who fit right into your team. Whether you need support in marketing, operations, creative, or finance, we’ve got you covered.

• Wide-ranging skills: From marketing to finance, find talent tailored to your needs.

• Save money: Cut hiring costs by up to 60% without sacrificing quality.

• Risk-free hiring: No fees until you find the perfect fit.

Build your team today!*

HumanX: An inflection point in the development of AI 

When he kicked off the HumanX conference on Sunday evening, CEO and co-founder Stefan Weitz expressed a clear focus on two things: trust and responsibility, two elements that are largely lacking in the field of artificial intelligence, and two things that, according to Weitz, are vitally important. 

• Shortly afterward, former Vice President Kamala Harris spoke about AI, saying that the perceived choice between safety and innovation is a false dichotomy; we can have both, she said, and we should have both. 

• She said it’s important that developers work, not just on developing a technology for its own sake, but on developing a technology that serves specific needs and solves specific problems. 

These themes were repeated with relative consistency throughout the length of the conference. 

From there, I spent the next three days talking (and listening) to professors, researchers, scientists, doctors, technologists and executives who play in all the different spaces and arenas in which artificial intelligence touches. And in everything from the enterprise to healthcare, a definite throughline emerged: that the technology itself, on its own, only gets you so far. 

Generative AI exists in many of these places at the center of a massive push that is as reliant upon trust and implementation from domain experts, proper oversight and governance mechanisms, proper levels of clear transparency and regulation, as it is on raw, technical capabilities. It is a tool, at the end of the day, that needs to get adopted in order for the business of AI to succeed. And it is a tool whose adoption raises a number of critical questions that run the gamut from legal liability to reputational risk and, quite simply, unclear return on costly investments. 

Business owners, after all, are focused on serving their clients, rather than figuring out a new tech whose value remains unclear. 

This was the topic of one of the three panels I moderated at the conference, a panel that sought to answer questions around enterprise adoption of generative AI. And the takeaway, from the experts and executives sitting at the forefront of that push, was simple: there is no “one-size fits all solution” for adoption, and for many enterprises, genuine use cases remain exceedingly vague. 

The panelists said they use generative AI themselves internally, largely for coding and research, with one describing a novel — and highly accurate — algorithmic approach to sales forecasting. 

Driving adoption: But, as they pointed out, they are technologists; they’re excited about tech that they themselves are developing. They see the use cases, and are actually deriving value from them. More accurate sales forecasting, for instance, is game-changing when it comes to tightening up business operations. 

• The problem is that, the farther this industry looks to expand beyond the tech sector, the industry-to-industry applicability remains, at best, blurred around the edges, a challenging environment that is not helped along by the overhyped narratives coming out of the industry itself. 

• These narratives make it difficult both for developers to manage expectations and for users to find the right applications.

The hype has raised awareness, but the hype is not aiding adoption.  

The technology is, in many ways, brittle and limited, something the industry doesn’t like to talk about. 

These internal issues come against a challenging macroeconomic environment, which, as we discussed during the panel, could dampen optimism around the tech, tugging the industry down and reducing interest in GenAI experiments. 

When we talk about a bubble bursting, this is what I’m referring to. 

The dot-com bubble didn’t kill the internet; it just killed a ton of companies that were overhyped and overvalued. 

The AI bubble similarly runs on an explosive combination of eye-wateringly high valuations and blistering hype that is rooted in discourse around a hypothetical general intelligence. 

Two years on from ChatGPT, we seem to be at an inflection point. The technology has been developed. The advancement is there. The innovation has happened. Sure, it can go farther, but the world needs time to actually wrap its hands around the scope and capability of the tech they’re dealing with. 

So in the midst of this rush for bigger and better models, new ones of which seem to come out every couple of weeks, barely edging out their competition on benchmarks that fail to measure real-world performance, we are at a stage where models need to become cheaper, more accessible and more secure. Reliability needs to be verifiably proven and privacy needs to be verifiably ensured in order for developers to earn trust. 

The developers that can’t overcome these issues, the developers that can’t sell a solution that reliably solves a significant problem, are likely the ones that won’t survive a bubble burst. 

And they’re quickly running out of time. 

🤔 Your thought process:

Selected Image 1 (Left):

• “There's not a reason in the world the person in Image 2 would be screwing something into a wall like that.”

Selected Image 1 (Left):

• “The drill, the attempt at looking official is good but I know too much about power tools. 😂”

💭 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 digital distrust:

Nearly half of you no longer trust any form of digital media by default.

35% of you still trust digital media, though — building distrust isn’t easy.

Constant vigilance.

Absolutely not:

• “I've always viewed any media through a critical eye, but it had never occurred to me that I needed to view digital images through such a lens.”

Something else:

• “Trust but verify.”

If you want to get in front of an audience of 450,000+ developers, business leaders and tech enthusiasts, get in touch with us here.