AI Impact: The Limits of Scale

Editor's note: This is AI Impact, Newsweek's weekly newsletter where each week, we will explore how business leaders are unlocking real value through artificial intelligence.

Tap here to get this newsletter delivered to your inbox.

Core Intelligence

...

Gary Marcus has spent years challenging big claims about artificial intelligence, and in a recent conversation with Marcus Weldon, Newsweek contributing editor for AI, he explained why today’s large language models impress people while still falling short of real understanding.

A former child prodigy turned cognitive scientist, Marcus said these systems can sound sharp because they pull patterns from text—but not because they grasp meaning. As he put it, “LLMs are like a box of chocolates, you never know what you're going to get.”

Marcus explained that humans learn through built-in abilities and real-world experience that help us form mental models of how the world works. LLMs, he said, don’t have that grounding. “Language is a compressed version of reality and that is its utility, but that is also its weakness,” he noted, arguing that text alone can’t reliably capture the rich structure humans use to understand objects, intentions, physical space and cause and effect.

He also warned that simply making today’s systems bigger will not solve these weaknesses. Marcus believes AI must combine pattern-recognition methods like deep learning with symbolic reasoning and “world models” that reflect how things actually work. “We need world models as a centerpiece rather than an afterthought,” he said, pointing to examples like AlphaFold and GNoME that succeed by embedding scientific structure into their designs.

Despite his critiques, Marcus isn’t pessimistic about AI’s long-term potential. He thinks smarter, more reliable systems will come—but only if researchers move past scaling and begin incorporating the same kinds of structure and understanding that humans rely on.

You can read Weldon’s full analysis of the discussion here: The bittersweet, Forrest Gump-like world of AI, according to Gary Marcus.

Upcoming Webinars

How AI is Reshaping Business

...

As artificial intelligence moves from experimentation to execution, business leaders are shifting from asking whether AI should be adopted to how it should reshape their organizations. That question sits at the center of Newsweek’s upcoming webinar, “How AI Is Reshaping Business,” featuring Gopi Kallayil, chief business strategist for AI at Google, and Suraj Srinivasan, professor at Harvard Business School and member of the Newsweek Advisory Board. Both bring deep experience working with C-suites, navigating the operational and strategic pressures created by the accelerated adoption of generative AI.

Kallayil works with executive teams across industries as they rethink the role of AI in core operations, customer experience and organizational design. Srinivasan’s research focuses on how artificial intelligence is transforming governance, performance and decision-making. He noted that “in the last few years, we have seen data create immense value for organizations,” and that generative AI is now “turbocharging it” by making the creation of insights, software and design increasingly accessible.

Together, their perspectives frame a critical moment for businesses: the shift from piloting AI tools to integrating intelligence into how companies operate, compete and grow. “How AI Is Reshaping Business” will take place on Monday, December 1, at 1 p.m. ET.

The Myths and Realities of AI for SaaS

...

Dr. Ranjit Tinaikar’s recent op-ed on Newsweek, "AI and the New SaaS Buyer: Myths and Realities of the Impact of AI," argues that the true transformation of AI in the SaaS world is happening where few are looking. The biggest productivity gains, he writes, are coming from the engineering side, where AI is accelerating code generation, refactoring and modernization rather than in customer-facing tools or new revenue streams. Despite the hype, the path from AI implementation to measurable business impact remains uneven, and most of the value still sits behind the scenes.

Tinaikar warns that while AI is becoming table stakes for SaaS providers, buyers aren’t paying more for it. Instead, he sees the near-term opportunity in legacy modernization and automation, where ROI is tangible and adoption less risky.

He’ll discuss these findings with Gabriel Snyder, editorial director of Newsweek Nexus and co-host of the AI Impact Interview series, in a live webinar, The Myths and Realities of AI for SaaS, on December 4 at 11 a.m. ET.

AI Impact Awards & Summit

The Newsweek AI Impact Awards seek to identify and recognize uniquely innovative AI solutions that solve critical business problems in different industry segments, or significantly advance capabilities. Recognition comes not from ideas, but from measurable IMPACT on business operations.

Prompt Injection

...

“Customer data is the multiplier that determines the real ROI of AI. Models are powerful, but they’re only as smart as the data feeding them. Investing in clean, unified data is how companies turn AI from potential into performance.”

Have your own lesson to share? Email us at [email protected]

Run Log

By Adam Mills

At Novartis, scientists don’t just wrestle with biology; they wrestle with data. The company runs thousands of clinical studies and draws on information from hundreds of millions of patient records, lab results and research programs. For years, those datasets sat in silos, slowing even basic scientific questions and forcing teams to pull information from a patchwork of disconnected systems.

Birgit Schoeberl, head of Data42 at Novartis, said the goal is straightforward: “to put the full breadth of our scientific data into the hands of the people who need it most, so they can make better decisions, faster.” Achieving that required rethinking how the company organizes its scientific knowledge.

Novartis built an AI-driven research backbone that brings together more than 700 million patient lives, 3,000 clinical trials, molecular data and scientific literature in a single environment. Instead of navigating separate databases, scientists can now explore how diseases, compounds, biomarkers and outcomes connect—and surface insights that once required days of manual data gathering.

Human dose prediction shows how dramatic the shift has been. Historically, the modeling and expert review took close to a week. With the new system, researchers can compare prior studies, run automated analyses and generate a dose estimate in about two hours. Beyond speed, the process has become more consistent and reproducible, tightening the link between early discovery and first-in-human studies.

“Instead of starting from scratch, we’re building on everything we already know,” Schoeberl said. “It lets our scientists spend less time gathering data and more time interpreting it.”

For Novartis, the promise of AI isn’t just acceleration. It’s clarity, turning an unwieldy mass of scientific data into a coherent, searchable foundation for discovering new medicines and identifying where new therapies might have the best chance to succeed.

Context Window

■ Red Bull partnered with Scania and PlusAI to deploy two autonomous trucks moving in perfect sync, carving a sub-one-second window for a stunt bike jump, demonstrating that the same precision tech powering commercial freight can now headline high-risk spectacle. [Newsweek]

■ Jeff Bezos is stepping into a formal operational role as co-CEO of a new AI startup (Project Prometheus) that has already raised around $6.2 billion and aims to build AI for engineering, manufacturing and aerospace. [The New York Times]

■ A prominent tech-industry investor (Klarna’s co-founder) warned that the pantheon of mega-scale AI infrastructure spending may be reaching a “bubble” stage, arguing models are increasingly efficient and questioning whether the vast data-center build-out justifies its cost. [Financial Times]

■ A venture capital firm exclusively focused on legal technology has just closed a $110 million second fund, underscoring the growing belief that AI can automate even highly skilled professional work, in this case, corporate legal workflows. [The Wall Street Journal]

■ Major banks are quietly moving past AI hesitation, deploying agents and automation in customer-service, onboarding and compliance workflows and framing the shift as augmentation rather than job cuts. [Newsweek]

Transfer Protocol

Yang Song, Ph.D., formerly director of AI at Apple and VP of AI research at IBM, has been named senior vice president and chief data science & AI officer at Equinix, reporting to Harmeen Mehta and leading the company’s global AI strategy.

Arun Chandra, previously vice president of operations at Meta and a longtime operations leader at Apple, is joining NICE Ltd. as chief operating officer, where he will head the company’s new Global Customer Operations division and support its AI-first CX strategy.

Sachin Katti, formerly chief technology and AI officer at Intel, has left the company to join OpenAI, where he will focus on building next-generation compute infrastructure to support AGI development.

Paul Dillahay, previously CEO of NCI Information Systems and a veteran of AI-enabled government services, has been appointed president of the AI division at DigitalNet.ai.

AJ De Rosa, formerly executive vice president at Intellectus Partners, has been named chief executive officer of Intellebox.ai, a newly spun-out platform focused on AI-driven infrastructure for financial advisors and wealth-management firms.

Know someone on the move in AI? Send job change info to [email protected]

Magic Moment

...

"My parents are in their mid-80s and they call me for everything [because] they know I'm in the tech space... I'm almost like full-time I.T. support, and usually after I help them out. I always say, 'You should just Google it.' So, instead of doing that because it never worked, I actually got them an account on ChatGPT, and they don't ever call me about stuff anymore...they're much more comfortable utilizing it. I don't get IT support type of questions anymore from my parents... I want to hear from my parents, of course. I don't want to spend 90 percent of that time explaining how their iPad works." Experience some AI magic? Tell us about it at [email protected]

CEO Circle

Before you go, be sure to take a look at “Why Centralizing Health Care in the Age of AI Is a Mistake” –– where Newsweek CEO Circle member Lawrence Rosenberg (president & CEO of the Integrated Health & Social Services University Network for West-Central Montreal) argues that while health care systems worldwide are consolidating budgets, data and decision-making, the very architecture required for AI to deliver transformational value is decentralization.

Click here to get this newsletter delivered to your inbox.

Request Reprint & LicensingSubmit CorrectionView Editorial & AI Guidelines