AI Insights

OpenAI Says AI Hallucinations Are Systemic, Not a Bug

Published

3 days ago

September 9, 2025

Large language models don’t just make mistakes. They sometimes invent answers with striking confidence. A new paper from OpenAI researchers Adam Tauman Kalai, Ofir Nachum, and colleagues argues that these “hallucinations” are not mysterious glitches but predictable byproducts of the way today’s artificial intelligence (AI) systems are trained and tested.

The report, “Why Language Models Hallucinate,” traces the problem to two root causes: the way models learn language during pretraining, and the way they are judged during evaluation. Together, these forces create statistical pressure to guess rather than to acknowledge uncertainty.

The first stage, pretraining, exposes a model to massive datasets. The researchers argue that even if those datasets were perfect, hallucinations would still occur because the training objective — predicting the next word — maps onto the same error patterns seen in binary classification. For example, if a model sees a celebrity’s birthday once in training, it cannot reliably reproduce it later. As the authors explain, hallucinations are simply “errors in binary classification” magnified by the task of generating fluent language.

The paper illustrates this with striking cases. When asked the birthday of one of the paper’s authors, Adam Tauman Kalai, an open-source model confidently supplied three different but incorrect dates, even though the correct answer was not in its training set.

In another test, when asked to count the number of Ds in the word DEEPSEEK, several models produced answers ranging from 2 to 7, none of them correct. These examples, the authors argue, show how models “fill in the blanks” with plausible guesses when they lack reliable information or when the task itself is poorly represented in training.

Why Post-Training Keeps Errors Alive

The second stage, post-training, is supposed to refine models and reduce errors. Yet the paper argues that evaluation systems — benchmarks and leaderboards — end up encouraging bluffing instead of honesty. Most widely used tests reward correct answers but assign zero points to uncertainty or an “I don’t know” response. That means a model that always guesses will consistently score better than one that admits gaps in its knowledge.

As the authors put it: “Optimizing models for these benchmarks may therefore foster hallucinations. Humans learn the value of expressing uncertainty outside of school, in the school of hard knocks. On the other hand, language models are primarily evaluated using exams that penalize uncertainty. Therefore, they are always in ‘test-taking’ mode.”

This framing helps explain why hallucinations remain stubborn even in the most advanced systems. Improvements in architecture, scale and alignment don’t change the fact that the scoring rules push models toward overconfidence.

The paper concludes that the solution isn’t another hallucination test but a redesign of the evaluation system itself. By modifying benchmarks to give partial credit for uncertainty, much like standardized exams that penalize wrong guesses, developers can realign incentives. The authors suggest explicit confidence thresholds, where models only answer if they are more than, say, 75% sure.

For professionals in finance, payments and other sectors where accuracy is nonnegotiable, the takeaway is sobering. Hallucinations aren’t random quirks; they are systemic. They can also be expensive for businesses and consumers alike. Insurance companies, earlier this year, started covering AI hallucination mishaps.

Unless the field changes how it measures performance, AI systems will continue to “sound right” while sometimes being wrong. But with better scoring, the researchers argue, AI could be nudged toward becoming a more trustworthy partner in high-stakes decision-making.

Source link

Up Next

2 Artificial Intelligence (AI) ETFs to Confidently Buy Heading Into 2026

Don't Miss

European AI gets boost as Dutch tech giant invests €1.3B in France’s Mistral – POLITICO

The Editors

Click to comment

AI Insights

First Trust Nasdaq Artificial Intelligence and Robotics ETF (ROBT): How to Invest

Published

23 minutes ago

September 11, 2025

The Editors

The First Trust Nasdaq Artificial Intelligence and Robotics ETF (ROBT 1.91%) offers a sophisticated way to track AI companies, with an emphasis on fundamental analysis.

It does so through an advanced index methodology designed to target companies that meet specific artificial intelligence and robotics criteria, rather than simply buying the largest names in the space.

However, with this added complexity comes higher costs. ROBT may not be the most beginner-friendly ETF. Here’s what you need to know to decide if the First Trust Nasdaq Artificial Intelligence and Robotics ETF is worth choosing over its AI-focused competitors.

Image source: Getty Images.

Overview

What is First Trust Nasdaq Artificial Intelligence and Robotics ETF?

The First Trust Nasdaq Artificial Intelligence and Robotics ETF is a thematic ETF that tracks the Nasdaq CTA Artificial Intelligence and Robotics index. It is not a mutual fund.

This benchmark takes a more nuanced approach than most AI ETFs by grouping companies into three categories:

Enablers, which develop the hardware, software, and infrastructure that form the building blocks of artificial intelligence.
Engagers, who design, integrate, or deliver AI-driven products and services.
Enhancers, which offer value-added AI capabilities as part of a broader business model, however, AI and robotics are not their primary focus.

Companies are scored across the three categories based on their level of AI involvement, with the top 30 in each category selected. However, the index is not equally split between these groups.

Engagers receive the highest weighting at 60% of the portfolio because they have the most direct AI exposure. Enablers make up 25%, while Enhancers account for15% to provide balance and diversification. Within each group, holdings are equally weighted.

The portfolio is rebalanced quarterly to reset weights and capture relative performance changes, while a semiannual reconstitution ensures the index remains current with evolving AI and robotics developments.

How to invest

Open your brokerage app: Log in to your brokerage account where you handle your investments.
Search for the ETF: Enter the ticker or ETF name into the search bar to bring up the ETF’s trading page.
Decide how many shares to buy: Consider your investment goals and how much of your portfolio you want to allocate to this ETF.
Select order type: Choose between a market order to buy at the current price or a limit order to specify the maximum price you’re willing to pay.
Submit your order: Confirm the details and submit your buy order.
Review your purchase: Check your portfolio to ensure your order was filled as expected and adjust your investment strategy accordingly.

Holdings

First Trust Nasdaq Artificial Intelligence and Robotics ETF has a heavy U.S. weighting at 64%, followed by Japan at 9.2%, then the U.K., Israel, South Korea, France, and Taiwan.

The fund holds 100 companies, with 50% classified as technology sector, 22% as industrials, and 9% as healthcare. The remaining holdings are primarily spread across the consumer discretionary and communications sectors, with small allocations to financials, consumer staples, and energy.

The First Trust Nasdaq Artificial Intelligence and Robotics ETF’s largest holdings as of late August 2025 are:

Symbotic (SYM 0.29%): 2.92%
Upstart Holdings (UPST -0.34%): 2.28%
AeroVironment (AVAV -1.81%): 2.25%
Ocado Group (OCDO -1.99%): 2.22%
Palantir Technologies (PLTR -1.39%): 2.14%
Synopsys (SNPS 13.35%): 2.07%
Recursion Pharmaceuticals (RXRX 6.72%): 2.00%
Cadence Design Systems (CDNS 4.95%): 1.91%
Gentex (GNTX 1.36%): 1.91%
Ambarella (AMBA -0.28%): 1.85%

The portfolio has a large-cap tilt, with an average market cap of $28 billion. Valuations are elevated, with shares trading around 29x price-to-earnings, 2.75x price to sales, and 17x price to cash flow.

Should I invest?

Only consider First Trust Nasdaq Artificial Intelligence and Robotics ETF if you specifically believe in its index methodology and the “engagers, enablers, enhancers” classification system, along with the resulting weighting across these groups.

There is no single “right” way to invest in AI. This is simply how ROBT’s benchmark index approaches selection and weighting compared to competing ETFs. It may under or outperform similar funds at various times.

If you appreciate a more complex, rules-based approach, you may find the fund appealing. But if you’re seeking simplicity, this ETF probably isn’t a fit.

Note that the ETF has historically lagged the broader market and has shown greater volatility than diversified index ETFs, like S&P 500 funds. Its relatively narrow portfolio means individual stock positions can have an outsized impact on performance, for better or worse.

Moreover, investing in AI and robotics carries idiosyncratic risk. These companies are often priced for growth, with high valuations that may be vulnerable to pullbacks if earnings don’t keep pace. The sector also tends to lack exposure to defensive, non-cyclical industries, which can leave long-term investors more exposed during market downturns.

Dividends

Does the ETF pay a dividend?

The First Trust Nasdaq Artificial Intelligence and Robotics ETF has a 30-day SEC yield of 0.27% as of August 2025. The ETF pays dividends semiannually in December and June. The yield is low because many AI-focused companies reinvest earnings into growth rather than paying dividends.

Expense ratio

What is the ETF’s expense ratio?

The expense ratio for First Trust Nasdaq Artificial Intelligence and Robotics ETF is 0.65%, or $65 per $10,000 invested annually. This is higher than both sector and broad market ETFs, and even on the pricey side for a thematic ETF, approaching the cost of some actively managed funds due to its more specialized index methodology.

Expense Ratio

A percentage of mutual fund or ETF assets deducted annually to cover management, operational, and administrative costs.

Historical performance

Since its inception, First Trust Nasdaq Artificial Intelligence and Robotics ETF has generally tracked its benchmark, the Nasdaq CTA Artificial Intelligence and Robotics Index, but has delivered slightly lower returns across most periods due to fee drag.

For example, over the past five years, the fund returned 6.4% annually versus 6.9% for the index.

Where the gap really shows is against the broader market: The S&P 500 has compounded at nearly 16% annually over the same period, far ahead of the ETF.

This underperformance highlights two challenges with thematic funds like this: higher volatility and sector concentration.

While the ETF has at times outpaced the S&P 500 over short stretches, it has struggled to keep up over longer horizons, reflecting the risks of a narrower, more specialized portfolio.

ROBT annualized total returns as of July 31, 2025
	1-Year	3-Year	5-Year
Net Asset Value	14.38%	9.54%	6.35%
Market Price	14.56%	9.58%	6.37%

The bottom line

First Trust Nasdaq Artificial Intelligence and Robotics ETF takes a very involved approach to index construction, breaking the AI and robotics universe into three categories and then assigning different portfolio weights to each group.

While this adds a layer of precision that some investors may appreciate, it also introduces complexity that can make the strategy harder to evaluate and follow compared to simpler, market-cap-weighted thematic ETFs.

The 0.65% expense ratio is on the higher side for a passive ETF and approaches the cost of certain actively managed funds in this space, which could make some investors question whether the additional complexity justifies the fee.

Over the long term, higher costs combined with a specialized weighting methodology may influence performance, so this fund may be best-suited for those who specifically want this unique structure rather than a broader, more conventional approach.

FAQ

Investing in First Trust Nasdaq Artificial Intelligence and Robotics ETF FAQ

What is the best way to invest in AI and robotics?

The best way to invest in AI and robotics depends on your investment style. Some investors are more comfortable with diversified thematic ETFs like the First Trust Nasdaq Artificial Intelligence and Robotics ETF, while others prefer to build their own portfolios of individual stocks.

Is ROBT a good investment?

First Trust Nasdaq Artificial Intelligence and Robotics ETF may appeal if you like its complex index approach, but its high fees and convoluted weighting can be drawbacks.

What is the best AI and robotics ETF?

The best AI and robotics ETF varies by investor goals, costs, and desired exposure, so compare options carefully. Some of the top AI and robotics ETFs by market cap include:

Global X Robotics & Artificial Intelligence ETF (NASDAQ:BOTZ)
Global X Artificial Intelligence & Technology ETF (NASDAQ:AIQ)
iShares Future AI & Tech ETF (NYSEARCA:ARTY)
Roundhill Generative AI & Technology ETF (NYSEARCA:CHAT)

Tony Dong has no position in any of the stocks mentioned. The Motley Fool has positions in and recommends AeroVironment, Cadence Design Systems, Gentex, Palantir Technologies, Symbotic, Synopsys, and Upstart. The Motley Fool has a disclosure policy.

Source link

AI Insights

Franchise AiQ™ Hosts Exclusive Denver Meetup: “AI for

Published

25 minutes ago

September 11, 2025

The Editors

Greenwood Village, Sept. 11, 2025 (GLOBE NEWSWIRE) — Greenwood Village, Colorado – September 11, 2025 –

Franchise AiQ™, the AI-powered marketing and lead activation platform for franchisors and franchisees, announced today that it will host an in-person educational event, AI for Franchises: Revolutionize Your Franchise with the Power of AI, on Wednesday, September 24, 2025 from 6:00 PM to 8:00 PM MDT at Venture X Denver Tech Center.

This high-impact session is designed to help franchise owners, multi-unit operators, and industry professionals understand how Artificial Intelligence (AI) is transforming the way franchises operate, grow, and compete.

The franchise industry is at a crossroads. Customer expectations are evolving, digital-first competitors are rising, and operational costs continue to climb. Many franchise owners are asking how they can deliver consistent service, scale efficiently, and still compete in local markets. The answer increasingly involves AI.

Artificial Intelligence is no longer a “nice-to-have” tool reserved for large corporations. It is now an essential resource for everyday franchise operations. Owners who educate themselves on AI and begin leveraging it in marketing, sales, and customer service are positioning their businesses to thrive in the years ahead. Those who delay risk falling behind competitors that embrace automation, personalization, and data-driven decision-making.

This Meetup will show attendees exactly how to make AI work for their franchise model, combining education with real-world strategies.

What franchise owners will gain from attending – Participants will walk away with practical insights, including:

Proven tactics to streamline operations and reduce unnecessary costs

Ways to personalize and enhance customer interactions using AI-driven tools

Actionable strategies to apply AI for marketing, sales, and decision-making

Real examples of how forward-thinking franchises are using AI to increase profitability

Networking opportunities with other franchise professionals committed to growth

The choice of Denver as the host city underscores the region’s growing role in business innovation and technology adoption. The Venture X Denver Tech Center provides a collaborative backdrop for entrepreneurs and franchise owners to explore how AI can be a game-changer in their industry.

Expert insights from Franchise AiQ™

Lane Houk, Co-Founder of Franchise AiQ™, will lead the session. With years of experience in digital marketing and AI-powered systems for multi-location brands, Houk is passionate about equipping franchise leaders with tools to succeed.

“Franchise owners cannot afford to ignore AI. This is not about replacing people, it is about empowering teams to deliver more with less,” said Houk. “Our goal with this event is to demystify AI and provide a roadmap for owners to adopt it strategically. When franchises learn how to use AI to automate follow-up, optimize local search, and engage leads effectively, the results can be transformative.”

About Franchise AiQ™

Franchise AiQ™ empowers franchisors with an AI-driven marketing and lead activation system that ensures every franchisee maximizes conversions and local visibility without extra workload. Powered by ScaleSynth AI™, the platform automates lead follow-up, engagement, and Google Business Profile optimization across all locations. The result is consistent branding, scalable growth, and real-time performance insights, allowing franchisees to focus on running their businesses while franchisors gain a centralized engine for market dominance.

Learn more at https://franchiseaiq.com

Event Details

Title: AI for Franchises: Revolutionize Your Franchise with the Power of AI

Date/Time: Wednesday, September 24, 2025, 6:00–8:00 PM MDT

Location: Venture X Denver Tech Center, 6400 S Fiddlers Green Circle, Ste 300, Greenwood Village, CO

Registration: AI for Franchises Meetup Event: https://www.meetup.com/ai-for-franchises/events/310526027/?eventOrigin=group_upcoming_events

Media Contact

Franchise AiQ™
Attn: Media Relations
400 S Fiddlers Green Cir, Ste 300-22
Greenwood Village, CO 80111
Phone: (833) 987-3247
Email: info@franchiseaiq.com
Website: https://franchiseaiq.com

###

For more information about Franchise AiQ, contact the company here:

Franchise AiQ
Lane Houk
(833) 987-3247
info@franchiseaiq.com
6400 S Fiddlerrs Green Cir
Ste 300-22
Greenwood Village, CO 80111

Source link

AI Insights

FTC Probes AI Chatbots’ Impact on Child Safety

Published

1 hour ago

September 11, 2025

The Editors

The Federal Trade Commission (FTC) is investigating the effect of artificial intelligence (AI) chatbots on children and teens.

The commission announced Thursday (Sept. 11) that it was issuing orders to seven providers of AI chatbots in search of information on how those companies measure and monitor potentially harmful impacts of the technology on young people.

The companies in question are Google, Character.AI, Instagram, Meta, OpenAI, Snap and xAI.

“AI chatbots may use generative artificial intelligence technology to simulate human-like communication and interpersonal relationships with users,” the FTC said in a news release. “AI chatbots can effectively mimic human characteristics, emotions and intentions, and generally are designed to communicate like a friend or confidant, which may prompt some users, especially children and teens, to trust and form relationships with chatbots.”

According to the release, the FTC wants to know what measures, if any, these companies have taken to determine the safety of their chatbots when serving as companions.

It is also seeking information on how the companies limit the products’ use by and potential negative effects on children and teens, and to inform users and parents of the risks associated with the products.

“The FTC is interested in particular on the impact of these chatbots on children and what actions companies are taking to mitigate potential negative impacts, limit or restrict children’s or teens’ use of these platforms, or comply with the Children’s Online Privacy Protection Act Rule,” the news release added.

As noted here last week when reports of the FTC’s efforts first emerged, some companies have already tried to address this issue.

For instance, OpenAI has said it would add teen accounts that can be monitored by parents. Character.AI has made similar changes, and Meta has added restrictions for people under 18 who use its AI products.

Those reports came the same day First Lady Melania Trump hosted a meeting of the White House Task Force on Artificial Intelligence Education. In a news release issued before the event, Trump said the rise of AI must be managed responsibly.

“During this primitive stage, it is our duty to treat AI as we would our own children—empowering, but with watchful guidance,” Trump said. “We are living in a moment of wonder, and it is our responsibility to prepare America’s children.”

Meanwhile, Character.AI CEO Karandeep Anand said last month he foresees a future where people have AI friends.

“They will not be a replacement for your real friends, but you will have AI friends, and you will be able to take learnings from those AI-friendly conversations into your real-life conversations,” Anand told the Financial Times.

Source link