BUSINESS

Do you remember the “early days” of live streaming TV? You know, back before cord-cutting was the hip thing to do. If so, you probably recall that PlayStation Vue was one of the leading premium pay options.

That service, which Sony launched all the way back in 2015, shut down in 2020 as mainstream brands entered the space with better pricing and better smart TV connectivity as cable TV alternatives.

But PlayStation hasn’t given up on the dream of being your ultimate entertainment machine. Instead of going back to the premium subscription model, Sony has done the opposite to try to regain streaming momentum:

It has launched a new free streaming TV service (FAST) called Live TV on PS5.

And, since we love a good free streaming service here at Team Clark, I decided to check it out so we could give you all the details.

Live TV on PS5: What You Need to Know

On August 31, PlayStation announced its return to the streaming TV world with the launch of Live TV on PS5.

Here’s a quick rundown of what you need to know from my quick tour of this new FAST service:

How Do You Access It?

There’s no subscription fee for this one, but you do need a PS5 gaming console to access it.

If you have a PS5, you can access Live TV on PS5 from the console’s home screen by following these steps:

Go to the Media tab:

Select the Live TV on PS5 icon. 

Select “Sign In With My PlayStation Account” to connect your PlayStation account and start watching.

To view content, Live TV on PS5 requires you to link your PlayStation account to create a Sony Pictures CORE account or link your existing account to access the service.

Note: I was unable to view the content via the PlayStation phone app at this time, and there was no indication in the company’s official release that it would be available this way.

What Channels Are Available?

This free service offers more than 100 live channels and on-demand content. Like its competitors in the space, it does not feature “pay-level” channels like ESPN or TNT. Instead, it offers a blend of niche-focused FAST channels.

News and Weather

NBC News Now

ABC News Live

LiveNOW from Fox

Scripps News

BBC News

Fox Weather

AccuWeather Now

Sports

NBC Sports Now

Fox Sports (not FS1)

Fubo Sports Network

Scripps Sports

NFL Channel (not NFL Network)

UFC Channel (not live UFC events)

PGA Tour Channel (not Golf Channel)

Entertainment

Ion

Walking Dead Universe

MGM Presents Action

Crunchyroll

A&E Crime 360

First 48 by A&E

Sony One Action Hits

The on-demand movie section features a predictable blend of offbeat and popular dated movies. These are likely to rotate regularly, as they do on most FAST services.

At the time of publication, some recognizable stars featured in films on the service were: Adam Sandler, Bruce Willis, Paul Rudd, Aubrey Plaza, Danny McBride, Jessica Alba, Nicolas Cage, Joaquin Phoenix, Kristen Stewart, Dwayne Johnson, Arnold Schwarzenegger, and Jason Statham.

Is the User Experience Any Good?

If you’re a regular gamer, using your PS5 to access a FAST service and a PS5 controller to change channels may feel natural.

But for most people, this experience is a little clunky in the smart TV era.

PlayStation gained streaming market share with its Vue product in the era during which you needed a third-party device attached to your television (such as a streaming dongle) to access non-cable or satellite content. This process was acceptable back then.

But now that many of these FAST channels are built native into modern smart TV (or available from the homepage via a quick app download) it seems like using your video game console to access these channels is a bit of a nuisance.

Bottom Line: The content selection is pretty decent for a freebie service, but using your video game console to access FAST channels seems like an unnecessary step in 2026.

I like PlayStation, but there is nothing here that separates this free service from a crowded marketplace that features some superior alternatives like Tubi TV, Pluto TV and The Roku Channel.

Will you be trying PlayStation’s newest streaming service? We’d love to hear your thoughts in the Clark.com community.
The post Streaming TV Returns to PlayStation and This Time It Is Free appeared first on Clark Howard.

Over many decades, the belief that the vast majority of Americans have never left the country and do not hold a valid passport has become deeply ingrained into certain stereotypes about the country.

That number was indeed at 17% in 2000 and just 5% in 1990. However, annual data shared by the U.S. State Department shows that it has crept steadily upward as the evolution of airfare costs has significantly democratized international travel.

The latest number is that there are currently 183 million, or between 48% and 53% of the American population, passport holders in the country. With 27.3 million of these passport issued last year, 2025 was also a record year for Americans applying for passports.

Why only 18.5% of Japanese citizens have a valid passport

The countries with the lowest passport ownership rates are generally developing nations where foreign travel is inaccessible for reasons ranging from weak currency to stringent visa requirements. But Japan has long bucked this trend as having the lowest number of citizens with passports out of all developed nations.

According to the latest Henley Index, Japan tied with South Korea and the United Arab Emirates as having the second-most powerful passport in the world.

Related: U.K. and Ireland crack down on some entry rules for travelers

Numbers released every year by the Japanese Ministry of Foreign Affairs, meanwhile, show that 14.73 million Japanese, or 18.5% of the population, has a valid passport in 2026.

Amid a government initiative to encourage more Japanese citizens to apply for a passport and a lowered application cost, this number rose from 17.5% in 2025. However, it still remains significantly lower than in other Asian nations, such as Taiwan and South Korea, where the number is upwards of a respective 60% and 40%.

The Okinawa Prefecture is introducing a new accommodation tax amid a dramatic spike in international travelers.Shutterstock

Multiple factors keep Japanese citizens from traveling abroad

With the country also welcoming a record 42.7 million international travelers last year, Japan is in the middle of a tourism explosion that is not being reciprocated by Japanese traveling abroad at nearly the same scale.

The reasons for low Japanese travel abroad come down to a combination of a high cost of living that makes foreign travel an out-of-reach luxury for many, a work culture that does not leave much holiday time, and other cultural factors such as low foreign language skills and certain intergenerational traditions around domestic travel.

More Travel News:

Airline to launch unusual new flight to Cayman Islands from the U.S.

There is a very cool Irish version of swimming pigs in The Bahamas

Unexpected country is most luxurious travel destination for 2026

Low-cost airline launches easier way to get to Sri Lanka

“The reason is the same as why Americans don’t travel abroad very often,” one Japanese commenter wrote in a discussion on the subject in a Reddit forum. “Domestic travel is their main focus. Foreign countries involve language barriers, and safety concerns are higher than in Japan. This trend existed even before the yen depreciated.”

The Japanese yen hit a 40-year low earlier this year, according to CNN. Financial concerns hit a significant portion of the population and are increasingly affecting travel decisions.

While over 20 million Japanese traveled abroad in 2019, the numbers are slumping by between 28% and 35% of their pre-pandemic levels.

Related: The latest ultra-luxury train will come to Portugal

When people are worried about their income, their job security, and the cost of everything from rent to food rising, they tend to become careful about spending.

“Consumer sentiment confirmed its early month reading, falling about 6% from last month and landing about 11% below a year ago amid continued worries that inflation will remain elevated for the foreseeable future,” according to University of Michigan Surveys of Consumers Director Joanne Hsu.

Consumers are worried, but some people are more worried than others.

“Groups who are typically less-equipped to absorb increases in cost of living also exhibited stronger decreases in sentiment, including older consumers, lower- and middle-income consumers, and those with no stock holdings,” Hsu shared.

That, however, has proven to be good news for discount retailers with dollar in their name including Dollar General and Dollar Tree.

Dollar General and Dollar Tree reported strong results

Dollar General and Dollar Tree both reported strong second-quarter results.

Dollar General’s Q2 net sales jumped 5.2% year over year to $11.3 billion, according to a company press release.

Its 3.5% same-store sales increase reflected a 2% increase in traffic and a 1.5% bump in average transaction amount.

Dollar Tree reported total sales grew 7% year over year to $4.9 billion, according to its second-quarter earnings release.

Same-store sales increased 3.7% due to a 3.3% increase in ticket and a 0.4% rise in traffic.

“Both chains have continued to attract shoppers looking for value, and Dollar General has also been seeing more middle- and higher-income customers,” according to Coresight Analyst Sujeet Naik.

Dollar General sees strength in weakness

Dollar General CEO Todd Vasos noted that his chain keeps growing its customer base.

“This marks the fifth consecutive quarter of growth in customer traffic as we continue to build on the momentum in our business with both new and existing customers,” he shared during Dollar General’s second-quarter earnings call.

He admitted a sales weakness among lower-income customers, but also believes that economic conditions will lead to more shopping by that group.

“Our core customers continue to be financially constrained with a variety of factors impacting their budget,” he said.

He believes gas prices will impact his chain’s customers’ willingness to visit Walmart, Target, Costco, and grocery chains that might be located farther away.

“As customers have continued to reduce trips and shop closer to home, Dollar General is uniquely positioned to meet their needs with more than 21,000 stores located within 5 miles of approximately 75% of the U.S. population,” he added.

GlobalData Managing Director Neil Saunders also saw Dollar General’s footprint as an advantage.

“The sales lift was driven by both traffic and a slight uplift in basket values,” he told RetailDive. “The former dynamic got a small boost from higher gas prices which, especially in rural areas, make the proximity and shorter drive-times of Dollar General stores a bit more attractive. We have seen this dynamic play out before and, essentially, it acts as a recruiting tool for Dollar General.” 

Dollar Tree’s and Dollar General’s large footprints help drive traffic from nearby shoppers.Shutterstock

Dollar Tree sees wealthier customers

Dollar Tree CEO Michael Creedon saw similar consumer behavior to what Vasos reported, and he talked about it during his company’s second-quarter earnings call.

“The consumer environment remains dynamic. Customers continue managing household budgets carefully, shopping with purpose and prioritizing value and affordability,” he said.

Those economic challenges, he noted, have driven wealthier customers to Dollar Tree.

“Our data shows we grew sales across all income cohorts. Households we serve were up nicely year-over-year with gains skewing to the middle and higher-income households,” he said.

Creedon noted that the challenging economy has hit lower-income customers hardest.

“First, the inflationary backdrop continues to pressure all household budgets, particularly for lower-income consumers. As our customers look for ways to stretch their dollars, they are increasingly turning to Dollar Tree for everyday essentials at compelling opening price points and pack sizes that help them manage their budgets,” he added.

Kroger, Albertsons, and regional grocery chains are the targets

Consumers have reasons to keep going to Walmart, Costco, and other discount chains for larger trips. Those chains are cheaper than traditional grocery stores, according to data from Consumer Reports’ Most and Least Expensive Supermarkets.

“Dollar Tree and Dollar General’s recent sales surge isn’t taking a bite out of Walmart or Costco. They’re going straight after the regional full-priced grocers,” RTM Nexus CEO Dominick Miserandino told TheStreet.

He thinks that traditional grocery chains face a significant problem in the current economy.

“Kroger, Albertsons, Publix, and local supermarkets are getting squeezed from both sides. Low-income shoppers are trading down to dollar stores for quick fill-in trips to manage weekly cash flow, while middle-income consumers are sliding in for basic consumables like milk, cereal, and paper towels,” he added.

Walmart and Costco, he explained, are already cheaper and provide enough value to cash-strapped consumers that it’s worth spending the gas money to get there.

“Walmart already owns the absolute low-price full basket, and Costco has a higher-income demographic locked into bulk unit economics. Neither of them is losing their core weekly trip. The dollar chains are stealing the mid-week fill-in run — the exact three-item trip for bread and detergent that traditional grocers relied on to pad their high-margin center aisles,” he shared.

Miserandino’s analysis is supported by data from Dunnhumby.

“Affordability challenges are reshaping U.S. consumer buying habits for grocery shopping, according to the latest Dunnhumby Consumer Trends Tracker (CTT). Mass-market and value retailers like Walmart, Dollar General, Dollar Tree and Family Dollar have achieved 79% penetration, matching the reach of traditional supermarkets for the first time,” Retail TouchPoints reported.

ALSO READ: Costco makes bold move into beauty and self-care

The company’s backlog has reached $95 billion as demand accelerates for its AI servers

Oil prices on Tuesday settled at their highest levels in nearly six weeks, after the U.S. military said it struck Iranian targets in the Strait of Hormuz in response to Tehran’s overnight attacks on ships transiting the vital waterway.

TheStreet aims to feature only the best products and services. If you buy something via one of our links, we may earn a commission.

Why we love this deal

While summer offers long, warm days and bright evenings, as we move toward fall, the days get shorter and sunlight wanes. Preparing your outdoor space for shorter, cooler days is important if you want to continue to get the most out of your space, instead of ignoring your outdoor space for half the year. A nice setup with cozy lighting and a warm centerpiece can help liven up the place, making it easier to stay outside for longer.  

If you need an affordable option to light up your patio, the Aodledy 50-Foot Remote-Controlled Patio Lights are a great choice. At 52% off, they’re long enough to cover a good portion of your outdoor space, and affordable enough to buy more if you need it. Shoppers can get this set of lights for just $14 at Amazon.

Aodledy 50-Foot Remote-Controlled Patio Lights, $14 (was $30) at Amazon

Courtesy pf Amazon

Shop at Amazon

Why do shoppers love it?

These lights feature 40 feet of lighting, including 19 lights and a replacement bulb, and a 10-foot plug that can easily reach an outdoor outlet without having to use extension cords. The bulbs are plastic, preventing shattering in any weather, and the whole set is waterproof, so it can be left out even in harsh conditions. They use 24-volt energy-saving lights that can be dimmed if needed, offering any type of lighting for your outdoor activities, whether you’re playing games or just relaxing. 

Related: Walmart’s $75 fire pit can be used as a table and a beverage tub

The lights feature three modes, including steady lighting, breath, and flash for different types of gatherings, and a lighting timer. It can be controlled via the included remote, making it easy to turn on or change without having to get up, allowing you to turn them off whenever you’re done, instead of using solar options that stay on all night and may irritate your sleep if they shine through your window. They are easily installed using tacks to hang them up, or specific light hangers that keep them in place during windy conditions. They can also be sat on the ground to line the patio or steps, or wrapped around your deck bannister for a fun and easy vibe.

Details to know

Length: This set has 40 feet of lights plus a 10-foot cord for convenience.  

Modes: This set can be dimmed and features three visual modes, including steady on, breath, and flash. 

Weatherproof: These lights are waterproof and shatterproof, allowing you to use them in all weather. 

“These lights are perfect,” said one reviewer. “The bulbs are plastic, so you don’t have to worry about them shattering. They light up our camping canopy perfectly. They’re bright enough to sit and play cards at night, and they dim as well. It’s a great product with great value. I would definitely recommend them.”

Another shopper said, “These are great quality and look great. I love the quality, appearance, and ambiance it creates on my patio.”

Shop more deals

Jmexsuss 50-Piece Hanging Light Hooks, $10 at Amazon

Brightown Solar String Light 2-Pack, $9 (was $10) at Amazon

Joomer 65-Foot Solar Outdoor String Globe Lights, $14 (was $24) at Amazon

Whether you need bright light for outdoor games after dark, or want a cozy vibe to relax to, the Aodledy 50-Foot Remote-Controlled Patio Lights offer dimmable, remote-controlled lights that allow you to use your deck whenever you want. They’re easy, affordable, and offer different settings for any occasion. At just $14, this is a great set to stock up on for the whole house, whether you want more light by the front door, garage, or backyard. 

Getting to the root of Trump’s Washington D.C. beautification.

The U.S. Air Force Global Strike Command is replacing Vietnam-era Hueys with the ultramodern Grey Wolf helicopter as the U.S. overhauls key nuclear missile launch sites.

For many companies, timing their biggest product launches right before earnings is no accident. Two days ahead of its fiscal third report, Broadcom Inc. (AVGO) used a conference in Las Vegas, an event most investors ignore, to unveil a sweeping AI governance platform.

The timing matters less than what the announcement reveals about a bet Broadcom has already made with its balance sheet.

That conference was VMware Explore 2026 in Las Vegas. Broadcom has run the VMware business since its $69 billion acquisition closed in 2023, and it has mostly used the platform to sell software bundles to existing enterprise customers. This week, the pitch changed.

Broadcom introduced VMware Private AI Cloud, a platform meant to let companies run AI inference and autonomous software agents inside their own data centers instead of renting capacity from a public cloud provider, according to a Broadcom press release.

That distinction is important. Most of the AI infrastructure story so far has centered on hyperscalers building bigger data centers and chip suppliers financing that buildout.

Broadcom just told the market that a larger share of AI demand may never touch hyperscalers at all.

Broadcom estimates that 56% of enterprises are already running or planning to run production AI inference on a private cloud, according to the company’s announcement.

That figure is the real thesis behind this week’s reveal. A majority of large companies want AI running on infrastructure they control, not infrastructure they rent from someone else.

Broadcom’s AgentMinder governs autonomous AI agents

Broadcom paired the private cloud pitch with tools built to police AI agents, not just run them. VMware AI Factory can run more than 150 open-source and commercial models, and it works with accelerator hardware from Nvidia and AMD, the company said.

Broadcom also introduced AgentMinder, a system built to track and restrain what autonomous AI agents can do inside a company’s systems. It’s a feature most enterprises would be interested in.

More AI:

Nvidia just made a move Wall Street wasn’t ready for

Microsoft just took sides in AI policy fight

OpenAI just disclosed something genuinely alarming

Marvell vs. Broadcom: the custom silicon shift

According to Broadcom CIO Alan Davidson, the tool provides “chain of custody capabilities between developers and multiple agents” across the company’s own operations.

Dell, HPE, and Nvidia’s own enterprise reference architecture have pitched similar on-premises AI stacks for two years, with limited uptake.

So what does Broadcom have that they don’t? VMware’s installed base: Tens of thousands of enterprises already running vSphere and vSAN won’t need to rip out infrastructure to adopt a new platform.

Broadcom unveiled VMware Private AI Cloud two days before its Q3 earnings.SOPA Images / Getty Images

Broadcom stock heads into earnings already moving

Broadcom shares closed at $370.34 on Aug. 31, up 0.42% for the day. The stock trades well below its all-time closing high of $480.77, set in June, reflecting a broader pullback in AI infrastructure names since early summer.

The bigger event lands Wednesday, Sept. 2. Broadcom reports fiscal third-quarter results after market close that day, and analysts expect adjusted earnings of $3.24 per share on revenue near $29.4 billion, up from $15.95 billion a year earlier, according to a consensus estimate reported by Seeking Alpha.

Investors are also watching a separate story. Broadcom is negotiating with lenders to raise up to $80 billion in debt to help finance AI chip deployments for companies including Anthropic, according to CNBC.

Broadcom’s strategy at a glance:

More than 150 open source and commercial AI models can run on VMware AI Factory, giving enterprises flexibility to switch providers without rebuilding infrastructure.

Certification with servers from Cisco, Dell, Supermicro, and Lenovo means companies can deploy the platform without replacing existing hardware vendors.

A financing arrangement worth up to $80 billion in debt, reported by CNBC, would fund AI chip deployment for external customers such as Anthropic, separate from the enterprise software push.

Broadcom deal finances custom AI chips

Most of the market’s attention on Broadcom sits with the debt financing story, because the numbers are enormous and the structure is unusual. That deal, routed through a special purpose vehicle, finances custom chips for a handful of AI labs racing to build frontier models.

This concentration risk is precisely why the enterprise push matters. It gives Broadcom a second, more diversified AI revenue stream that doesn’t depend on a handful of frontier labs remaining solvent.

The VMware announcement is a different bet aimed at a different customer. It targets the much larger and more conservative pool of enterprises that run payroll and manage customer data, not frontier models, and that have spent decades worrying about compliance and control.

The AI infrastructure story has mostly been told as a conversation between chimakers and hyperscalers.

Broadcom’s VMware push is a bet that a third group, the ordinary enterprise, ends up mattering just as much to how AI actually gets deployed.

If that bet pays off, the winners in AI won’t only be the companies with the biggest data centers. They will be the ones that gain access to normal sectors by making private AI infrastructure boring enough for a bank, hospital, or insurer to trust.

Related: BMO sees writing on the wall for Broadcom stock after earnings

When large language models (LLMs) hallucinate, developers typically assume the model lacks the required facts. Engineering teams diagnose the error as missing knowledge. The standard response is to increase model size, expand training data, or build complex retrieval architectures.A new study by researchers at Google Research and Technion demonstrates that the knowledge is often not missing. The model has the information encoded parametrically but fails to surface it during generation. Their experiments show that frontier models like GPT-5 and Gemini-3 encode 95-98% of tested facts. This indicates that in many cases, recall, rather than encoding, is the primary bottleneck for factual accuracy. By understanding how to unlock existing knowledge through inference-time computation, engineering teams can build more reliable applications without necessarily relying on larger models or external databases.Knowledge profiling: measuring what models actually knowTo map this gap between storage and retrieval, the researchers propose shifting the evaluation focus from question-level accuracy to fact-level profiling. Instead of simply scoring whether an LLM answers an isolated prompt right or wrong, fact-level profiling tests a single underlying piece of information across multiple conditions, evaluating whether the fact is stored in the model’s parameters at all, whether it can be queried from different directions and phrasings, and what computational effort is required to retrieve it.This framework distinguishes between whether a fact is parametrically “encoded” and whether it is “known”. A model encodes a fact if it can accurately reproduce it when primed with its original training context. A model knows a fact if it can reliably answer questions about it across varied phrasings and directions.”Encoding and recall failures are indistinguishable under accuracy metrics, yet they imply different limitations and solutions,” the researchers write. “Encoding failures call for pre-training interventions, such as scaling model size or data coverage. Recall failures suggest post-training interventions that often improve how models utilize what they already encode.”The paper illustrates this using a sample fact: Oasis played their first gig at the Boardwalk club. Based on how models process this information, the study categorizes knowledge into five distinct profiles:Direct recall: The model encodes the fact and readily accesses it to answer direct questions without extra inference compute.Encoding failure (empty shelves): The model neither encodes nor knows the fact. It cannot complete a Wikipedia-style sentence about Oasis’s early days, nor can it answer questions about the event. This signals a need for more pre-training data or greater model capacity.Recall failure (lost keys): The model has the fact encoded but cannot access it. It can seamlessly complete the original training text about Oasis, but fails to answer “Where did Oasis play their first show?” even when given time to think.Recall with thinking: The fact is encoded, but inaccessible to direct generation. It is only successfully recalled when the model uses inference-time computation, such as Chain-of-Thought, to bridge the gap. The researchers refer to this mechanism as recall facilitation. The model might initially fail to answer the direct question. By generating intermediate thoughts about the band’s early history in Manchester, it structurally primes itself to locate and recall the locked answer.Inference without encoding: The model never explicitly encoded the Oasis fact. Instead, it successfully answers the question by making an educated guess or reasoning across other encoded facts it does know. It might deduce the answer by chaining together separate data points, such as “Oasis formed in Manchester,” “the Boardwalk was a famous 90s music club there,” and “the Boardwalk hosted early gigs by emerging bands.”Scaling illusions, long-tails, and tip-of-the-tongue recoveriesThe researchers evaluated 13 LLMs on over 4 million responses. They used WikiProfile, a benchmark containing 2,150 facts extracted from Wikipedia, testing each fact across formats ranging from exact context completion to multiple-choice verification.For frontier models like GPT-5 and Gemini-3, encoding is nearing saturation. These models successfully encode 95-98% of the tested facts. However, they still fail to directly recall 26-34% of those encoded facts without thinking. Inference-time thinking acts as a vital recovery mechanism. Providing models with extra computational effort successfully retrieves 40-65% of the encoded facts that models initially fail to directly recall. The researchers compare this to the human tip-of-the-tongue state, where deliberate effort, such as mentally retracing context, eventually helps remember the information. Scaling up model size does not automatically resolve this gap. For example, they found that scaling the Gemma3 model from 1 billion to 27 billion parameters decreased encoding failures from 85% to 23%. But at the same time, the share of recall failures increased, peaking at 40% without thinking. This suggests that scaling mainly solves the storage problem rather than the access problem. As the model memorizes vastly more facts, a larger pool of knowledge becomes trapped in an “encoded but inaccessible” state. The bulk of model errors shifts from missing data to failed recall.”Our findings suggest that recall is tightly coupled to the conditions under which facts were learned, degrading when queries diverge from training-time patterns,” the researchers write. How a user asks a question directly dictates whether the model can unlock the stored answer.For example, the experiments showed that rare facts are encoded at rates similar to popular facts. Yet they found a large recall gap between long-tail and highly popular facts that exceeds 25% for frontier models. Similarly, models struggle to generate answers to reverse questions (i.e., asking for the subject instead of the object). For example, a model might easily answer that Oasis played their first gig at the Boardwalk club, but fail to answer who played their first gig at that same club. At the same time, the same models show that they know the correct answer when given the same question in multiple-choice format.”Whereas these failures are often interpreted as limitations of memorization or bidirectional encoding, our results suggest a different picture: rare facts are often encoded but inaccessible, and reverse facts can be recognized even when they cannot be generated,” the researchers write. “This reframes both phenomena as recall failures rather than ‘missing knowledge.'”The ROI of thinking and tips for developersThe high encoding rates of frontier models require a shift in how developers approach factuality and pipeline architecture.Don’t treat every factual failure as a retrieval problem: The default enterprise reaction to hallucinations is often to deploy Retrieval-Augmented Generation (RAG), scale up vector databases, or ingest more domain documents. While necessary for real-time updates or entirely absent data, this study shows up to 95-98% of standard facts are already parametrically encoded. Many failures can be recovered without external retrieval.Use inference-time reasoning selectively: Thinking recovered 40–65% of encoded facts that models failed to directly recall. However, thinking is computationally expensive. A practical architecture involves a fast first-pass model call, followed by a higher-reasoning-effort retry when confidence is low. Teams can optimize costs by dynamically routing challenging conditions, like rare entities or reverse questions, to thinking models.Test semantic access, not just benchmark accuracy: Standard accuracy metrics mask underlying model capabilities. Evaluation sets should probe the same underlying fact across different phrasings, contexts, and directions to truly understand what a model knows versus what it can reliably access.Leverage query reformulation and retries: Because recall is highly context-dependent, query framing dictates success. Changing the structure of a prompt, generating relevant intermediate context, or prompting the model to generate a reasoning chain before answering are legitimate reliability mechanisms that surface information direct prompts miss.Limitations and practical takeawaysThe WikiProfile benchmark relies on encyclopedic Wikipedia facts. These findings might not perfectly generalize to proprietary or highly specialized enterprise domains. A model’s ability to store and recall a niche internal company metric may behave differently than its handling of public encyclopedic data.Fully profiling a frontier model on the WikiProfile suite costs approximately $500. Developers can significantly reduce this cost by omitting multiple-choice variants or using fewer response samples per question. Teams can access the WikiProfile benchmark on Hugging Face to evaluate their own systems and build custom knowledge profiles. Identifying whether an enterprise application suffers from empty shelves or lost keys is the first step toward fixing it.