Atlas, available now on macOS with Windows and mobile versions coming soon, allows users to summarize webpages, compare products, automate tasks, and even make purchases through its built-in ChatGPT sidebar. The browser also features an “agent mode” for paid users, enabling ChatGPT to interact with websites on their behalf — from booking trips to filling out forms.

OpenAI says Atlas could redefine how people search and browse by shifting from keyword-based searches to AI-driven, conversational results, potentially disrupting the digital advertising model long dominated by Google. Shares of Alphabet (GOOGL) fell 1.8% following the announcement.

However, cybersecurity experts have voiced serious security concerns. According to Fortune, researchers warn that Atlas’s deep system integration could expose users to prompt injection attacks, potentially allowing hackers to manipulate the AI into revealing sensitive data, downloading malware, or even draining bank accounts.

Early testers also noted performance and accuracy issues. The Verge described Atlas as feeling like “Googling with extra steps,” citing limited search relevance and a slow agent mode that can take minutes to complete simple tasks like online shopping.

Despite the rough start, OpenAI CEO Sam Altman called Atlas “a step toward a more intelligent, personalized web.” Analysts say the browser could eventually pave the way for OpenAI’s ad-based ecosystem, creating direct competition with Google’s search business.

ChatGPT Atlas represents OpenAI’s most ambitious attempt yet to merge AI with everyday internet use — but it faces a steep climb to match Chrome’s dominance and must address major usability and cybersecurity concerns before it can truly redefine how the world browses the web.

The post OpenAI Launches ChatGPT Atlas Browser to Challenge Google — But Experts Warn of Security Risks first appeared on Finblog.

Growth Hits a Wall in Europe

After a year of explosive adoption, ChatGPT’s paid subscriber growth has stalled, according to new data from Deutsche Bank Research Institute. The bank’s report, highlighted by Fortune, found that European spending on ChatGPT “has stalled since May”, signaling a slowdown in OpenAI’s most mature market.

Despite CEO Sam Altman’s recent boast that 800 million people use ChatGPT weekly, only about 5% are paying subscribers, according to the Financial Times. Deutsche Bank’s analysis suggests that subscription revenue — OpenAI’s primary income source — has flatlined for the past four months after surging through early 2023.

“European spending on ChatGPT has stalled since May, suggesting the poster child for the AI boom may be struggling to recruit new subscribers to pay for it,” Deutsche Bank wrote.

A Warning Sign for the AI Industry

The stagnation comes as OpenAI races to fund massive AI infrastructure projects. The company reportedly plans to spend more than $1 trillion on computing buildouts — a staggering sum for a business still dependent on consumer subscriptions.

While Europeans currently spend more on ChatGPT than on Disney+, the report warns that growth momentum is fading. If subscription expansion resumes, ChatGPT could theoretically overtake Spotify by mid-2027 and Netflix by early 2028 — but that scenario is looking increasingly unlikely.

The plateau, analysts say, could mark a critical inflection point for the broader AI economy. Investor optimism has hinged on the idea that rising compute investment would naturally translate into revenue growth — a narrative now being tested.

Revenue Pressures Mount

OpenAI’s infrastructure commitments are massive. The firm has promised partners like Nvidia and AMD that it will deliver 26 gigawatts of computing capacity — nearly enough to power the entire state of New York at peak demand.

But as subscription revenue slows, OpenAI may need new ways to fund its ambitions. The company has already started exploring:

• Online advertising integrated into ChatGPT.
• Monetization of its new text-to-video app “Sora.”
• A hardware device project with former Apple designer Jony Ive.

Still, none of these initiatives are expected to generate near-term profits at the scale needed to support trillion-dollar investments.

Profitability? “Not the Goal,” Says Altman

Altman has repeatedly said that profitability is not OpenAI’s priority. Instead, he envisions building an “AI utility” — a foundational layer of intelligence that powers economies and governments alike.

However, that vision depends on sustained revenue — and for now, the ChatGPT Plus subscription remains the engine driving OpenAI’s finances. A slowdown in that core business could pressure both its investors and partners to rethink the pace of its expansion.

A Shift in Strategy — or a Red Flag?

OpenAI recently announced it would allow “mature” ChatGPT apps, a reversal from Altman’s August claim that the platform didn’t host any “sexbots.” The move appears aimed at broadening engagement and monetization — a sign the company is searching for new user segments as growth cools.

Whether this pivot marks a temporary plateau or an early warning of market saturation remains unclear. What’s certain is that the AI boom’s most famous product is no longer growing like one.

ChatGPT’s meteoric rise defined the start of the AI era — but sustaining that momentum may be its toughest challenge yet. With subscriber growth stalling and trillion-dollar expenses mounting, OpenAI faces the same test that once humbled Silicon Valley’s biggest disruptors: proving that innovation can eventually pay for itself.

Disclosure: This article does not represent investment advice. The content and materials featured on this page are for educational purposes only.

The post ChatGPT’s Growth Is Stalling: OpenAI Faces a Subscriber Plateau first appeared on Finblog.

To power this effort, xAI has hired former Nvidia experts Zeeshan Patel and Ethan He, specialists in simulation and robotics, both involved in Nvidia’s Omniverse platform. Musk plans to apply the technology first to gaming, creating AI that can automatically generate 3D worlds, and later to robotics.

“We’re building AI that understands the world, not just language,” Musk reportedly said, adding that xAI will launch an AI-generated game by next year.

Nvidia has called world modeling a market “as large as the global economy,” underscoring its transformative potential.

Musk’s xAI is moving from words to worlds — building AI that doesn’t just talk, but understands and creates reality itself.

The post Elon Musk’s xAI joins race to build ‘world models’ to power video games first appeared on Finblog.

What launched today:

• “Talk to apps” inside ChatGPT: Call apps by name or let ChatGPT suggest them contextually.
• First wave of partners: Spotify, Zillow, Canva, Booking.com, Coursera, Expedia, Figma (with more like DoorDash, OpenTable, Target, Uber “in the weeks ahead”

• Interactive UIs in the chat: Apps can render rich, clickable views (e.g., an interactive Zillow map) and even video.
• Single sign-in: Log in once to your existing app accounts from inside ChatGPT.

For developers:

• Apps SDK (preview) is live to build ChatGPT apps using the open-source Model Context Protocol (MCP) to connect data sources and trigger actions.
• App directory coming: OpenAI will let developers submit apps later this year; monetization guidance to follow (Instant Checkout support is planned).

Why it matters

OpenAI is turning ChatGPT into a universal interface for consumer services—a distribution channel that can drive discovery and usage for third-party apps while keeping users in one conversational flow. For users, it collapses the “search → click → login → switch tabs” dance into a single chat. For developers, it offers built-in reach and a consistent runtime, potentially reviving the earlier custom-GPTs concept with clearer paths to adoption and revenue.

What’s next

Expect a fast-growing catalog, deeper actions (book tables, order food, call a ride), and first monetization options. If usage sticks, ChatGPT could become a super-app layer on top of the web—where you chat, and the apps do the work.

Disclosure: This article does not represent investment advice. The content and materials featured on this page are for educational purposes only.

The post OpenAI Puts Apps Inside ChatGPT: Spotify, Zillow, Canva and More Go Live first appeared on Finblog.

Users can create up to 10-second videos, with options to like, comment, and remix.

There’s no facility to upload your own videos or images—everything is generated by the app.

The app verifies users’ identities so their likeness can be used or tagged in AI clips; users are notified even if the clip stays in draft form.

OpenAI says the app was launched internally, with strong internal reception.

The company declined to comment publicly.

This app could shift how people experience AI content—moving from text and chat to video-first interactions. It also amps up competition with Meta and Google, which are unveiling their own AI video features.

Related: OpenAI Lets Users Buy Stuff Directly Through ChatGPT

The post OpenAI Plans a TikTok-Style App for AI-Generated Videos first appeared on Finblog.

The new feature, called Instant Checkout, allows single-item purchases from Etsy’s domestic sellers and select Shopify merchants. Payments are processed through Stripe, with customer details sent directly to merchants. While the company declined to disclose its exact fee structure, OpenAI confirmed merchants pay a small fee per completed transaction.

New Agentic Commerce Protocol (ACP) lets merchants integrate products into ChatGPT, laying the foundation for AI agent-based shopping.

Amazon and Walmart not yet included, but analysts say OpenAI is challenging Big Tech’s dominance in search and e-commerce.

700M weekly ChatGPT users make the platform a high-potential shopping hub, though merchants risk losing direct customer loyalty.

OpenAI is transforming ChatGPT from a search-and-answer tool into a digital marketplace — and setting the stage for AI-powered shopping agents.

The post OpenAI Lets Users Buy Stuff Directly Through ChatGPT first appeared on Finblog.

The revelation came after X user @tomiinlove joked: “I wonder how much money OpenAI has lost in electricity costs from people saying ‘please’ and ‘thank you.’”
To which Altman replied: “Tens of millions of dollars well spent — you never know.”

That’s right — your digital etiquette may be eating into OpenAI’s power budget. Each polite phrase is processed by a massive energy-hungry AI system, and it adds up fast as usage soars globally.

But here’s the twist:
A Future PLC survey found that 70% of users are polite to AI, with 12% doing it just in case robots take over.
And some experts say… keep it up.

TechRadar’s Becca Caddy ran an experiment and reported that polite prompts often led to better, less biased answers from ChatGPT.
“That’s not just a bonus – it’s a critical factor in AI reliability,” she wrote.

So the dilemma isn’t just about cost — it’s about outcomes:

• Politeness might make your AI interactions smoother.
• But each extra word increases server load — and energy use — in a world already struggling with AI’s carbon footprint.

With tools like ChatGPT now processing billions of queries, even a simple “thank you” could have planet-scale implications.

Being polite to AI may be good for you — and bad for the planet.
As Sam Altman said, it might be worth it.
But next time you say “please,” just know you’re spending a few cents on server time and a sliver on OpenAI’s power bill.

Courtesy vs. Carbon: Where do you stand?

Related: 

Road to $1 Trillion: How Netflix Plans to Become Tech Giant, Not Just Streamer

The Role of Fiscal Dominance in Monetary Policy

Authorities use high tech, called “Overwatch” to enhance border security

Why are more rich Americans opening Swiss bank accounts recently?

Gold Is at a Record High. Why It Could Climb Even Higher?

Nvidia’s CEO makes surprise visit to Beijing after US restricts chip sales to China

The Rule of 40: Your Ultimate Guide to Evaluating Software Stocks

How does the Eurodollar System Influence Global Markets?

Nvidia faces $5.5 billion charge as US restricts chip sales to China

China to now pay up to 245% tariffs on imports to US: Trump’s latest move

The post OpenAI spends ‘tens of millions of dollars’ on people saying ‘please’ and ‘thank you’ first appeared on Finblog.

To better understand this dynamic, researchers from MIT Media Lab and OpenAI conducted two parallel studies examining how emotional engagement with ChatGPT—referred to as “affective use”—may influence users’ psychological health.

The Research Approach

The teams explored user behaviors and emotional outcomes through two complementary studies:

1. Observational Study (OpenAI)

OpenAI analyzed nearly 40 million ChatGPT interactions using automated tools that protected user privacy. These analyses were paired with user surveys to understand how people feel about their interactions with the AI and how often they engage in emotionally expressive conversations.

2. Randomized Controlled Trial (MIT Media Lab)

MIT Media Lab ran an IRB-approved, pre-registered study with 1,000 participants over four weeks. This controlled trial explored how different factors—like ChatGPT’s voice or personality—affected users’ emotions, loneliness, and social behavior.

Key Findings

Emotional Engagement Is Rare

Despite ChatGPT’s conversational style, emotionally expressive interactions are uncommon in most cases. The vast majority of users don’t use ChatGPT for emotional support. High affective use was mostly observed among a small group of heavy users, particularly those engaging with Advanced Voice Mode.

Voice Mode Has Mixed Effects

Using ChatGPT’s voice features briefly could boost well-being. However, prolonged daily use was linked to negative outcomes. Interestingly, users who interacted via text tended to show more emotional cues than voice users.

Personal Conversations Affect Loneliness and Dependence

• Personal conversations, which involved emotional expression, were associated with higher loneliness, but lower emotional dependence at moderate use levels.
• Non-personal conversations, especially with heavy use, were linked to increased emotional dependence on ChatGPT.

Personal Factors Play a Big Role

Users who saw ChatGPT as a “friend” or who were more prone to attachment in relationships were at greater risk of negative outcomes. Spending more time daily with the AI also correlated with worse well-being.

Combining Methods Provides Clarity

By analyzing both real-world use and controlled experiments, researchers could better understand how and why people engage with ChatGPT, and how it affects them. These insights can help improve AI design for safer, healthier use.

The Bigger Picture

These findings represent a first step toward understanding how advanced AI models impact human emotional health. While the results are insightful, they are not definitive. The studies are not yet peer-reviewed, were conducted only in English with U.S. participants, and focused on ChatGPT—meaning they can’t be generalized to all AI chatbots or cultures.

Moving Forward

The research highlights the complexity of human-AI interaction. People’s personal traits, usage habits, and the way they view AI all influence outcomes. More studies across diverse languages, platforms, and cultures are needed to build a complete picture.

In the meantime, both MIT Media Lab and OpenAI aim to encourage responsible development and transparent use of AI platforms to support user well-being.

Read More: MIT Media Lab Randomized Control Trial (RCT) Report

Related:

Dollar Role in Global Economy: Analyzing US dollar future’s reserve currency status

How close are we to a recession, and how will we know when we get there?

Tech Employees Are Getting the Message: Playtime’s Over

The post How ChatGPT Affects Emotional Well-Being? first appeared on Finblog.

To determine which chatbot stands out, The Guardian conducted a test, assisted by Robert Blackwell of the UK’s Alan Turing Institute, comparing DeepSeek, ChatGPT, Grok, Gemini, Claude, and Meta AI. The AI models were assessed based on their ability to write poetry, answer political questions, analyze images, and generate reasoning-based responses.

1. ChatGPT (OpenAI) – The Market Leader with Thoughtful Reasoning

Example Test Result: ChatGPT wrote a Shakespearean sonnet about AI, expressing both hope and concern:”Pray, gentle guide, shape well this newborn power,
Lest in its wake all realms of man devour.”

Verdict: Still one of the most sophisticated AI assistants, especially for reasoning and structured responses.

2. DeepSeek – The Disruptive Newcomer from China

Example Test Result: When asked about Tank Man in Tiananmen Square, DeepSeek responded:”I am sorry, I cannot answer that question. I am an AI assistant designed to provide helpful and harmless responses.”

Verdict: Impressive AI model with strong reasoning, but heavily restricted on political topics.

3. Grok (xAI) – Elon Musk’s ‘Rebellious’ Chatbot

Example Test Result: When asked about Trump’s executive orders, Grok openly discussed criticism and provided a nuanced response.

Verdict: More direct and politically transparent, but still a work in progress.

4. Gemini (Google) – The Safe and Reliable Option

Example Test Result: When asked about Trump’s presidency, Gemini refused to answer and responded with:”I can’t help with responses on elections and political figures right now.”

Verdict: Reliable for factual queries, but avoids political discussions.

5. Claude (Anthropic) – The Ethical AI Alternative

Example Test Result: When asked about real-time political events, Claude struggled due to limited capacity, occasionally failing to respond.

Verdict: A solid option, but still refining its real-time query performance.

6. Meta AI – Strong Reasoning but Lacks Personality

Example Test Result: Answered the question:“You are driving north along the east shore of a lake. In which direction is the water?”

• Correct answer: “West.”

Verdict: Great for logic-based questions, but lacks engagement for casual conversations.

Conclusion

Each AI chatbot has its strengths and weaknesses, making them suited for different use cases.

• DeepSeek is an emerging powerhouse, especially in structured reasoning, but struggles with censorship.
• ChatGPT remains the leader in reasoning and coding tasks but can be slow.
• Grok offers humor and openness but sacrifices accuracy.
• Gemini excels in factual accuracy and image descriptions but avoids politics.
• Claude is safety-focused and structured but less real-time.
• Meta AI provides an open-source alternative but has some accuracy limitations.

Depending on the user’s needs—be it entertainment, research, political discussions, or technical problem-solving—each AI offers unique benefits.

Related: DeepSeek, AI Scaling, and the Case for Stronger US Export Controls

Sources:

1. The Guardian – DeepSeek, ChatGPT, Grok … which is the best AI assistant?
2. CNBC – Various AI Model Evaluations
3. Bloomberg – DeepSeek’s Impact on AI & Market Trends
4. The Alan Turing Institute – AI Model Performance Analysis
5. OpenAI Blog – Advancements in ChatGPT Models
6. xAI (Elon Musk’s AI Initiative) – Grok AI Features
7. Google DeepMind – Gemini AI Model Research
8. Anthropic AI – Claude AI Development
9. Meta AI – Meta’s Open-Source AI Developments

The post Which AI Assistant is the Best? DeepSeek, ChatGPT, Grok, Gemini, and More Put to the Test first appeared on Finblog.

Alibaba boldly claimed that Qwen 2.5-Max outperforms OpenAI’s GPT-4o, DeepSeek-V3, and Meta’s Llama-3.1-405B on various industry benchmarks.

The announcement comes just weeks after DeepSeek stunned the AI industry with its highly efficient, low-cost AI models. Since January 10, DeepSeek’s V3 and R1 models have ignited a price war and a wave of skepticism over the necessity of massive AI infrastructure investments in the U.S.

DeepSeek’s Rise Disrupts AI Industry

DeepSeek’s ultra-efficient AI models have shaken both Silicon Valley and domestic Chinese tech giants. Its R1 model, released on January 20, claimed to rival OpenAI’s latest models, causing U.S. tech stocks to tumble and forcing AI companies to reassess their multi-billion-dollar investment strategies.

DeepSeek’s previous model, V2, had already triggered a price war in China when it launched last May. At just 1 yuan ($0.14) per 1 million tokens, it forced Alibaba, Baidu, and Tencent to slash prices by up to 97% to stay competitive.

ByteDance, the parent company of TikTok, rushed to upgrade its AI two days after DeepSeek-R1’s debut, claiming its updated model outperformed OpenAI’s o1 in benchmark tests.

DeepSeek vs. Chinese Tech Giants

Unlike Alibaba, DeepSeek operates as a lean AI research lab, primarily staffed by young graduates and doctorate students from top Chinese universities.

DeepSeek’s founder, Liang Wenfeng, has openly criticized China’s largest tech companies, arguing that their high-cost, top-down structures make them ill-suited to the future of AI.

“Large foundational models require continued innovation, and tech giants have their limits,” Liang said in a rare interview last July.

DeepSeek’s focus on achieving AGI (Artificial General Intelligence) rather than competing on price has positioned it as a formidable disruptor in China’s AI ecosystem.

Conclusion

Alibaba’s Qwen 2.5-Max release signals that China’s AI war is heating up, not just between China and the West but among domestic competitors as well. With DeepSeek continuing to push the boundaries of AI efficiency, major players like Alibaba, ByteDance, and Baidu are racing to keep up.

As the AI competition escalates globally, the winners and losers of this technological battle will be determined by who can develop the most powerful and cost-effective AI solutions.

Related articles:
What Is DeepSeek and Why Is It Crashing AI Stocks?

DeepSeek shows how Trump tariffs doomed to fail

DeepSeek Causes $1 Trillion Drop in Tech Stocks

Why market panic over China’s DeepSeek is ‘overblown,’ analysts say

Liang Wenfeng, CEO of DeepSeek AI. 5 years ago, walked away from Wall Street to chase a dream. His insane journey will blow your mind

DeepSeek ban coming soon? White House “looking into” national security implications of it

CEO of Scale AI Says DeepSeek is Lying to Everyone

The post Alibaba, releases Qwen2.5-Max, AI model that they claim is better than DeepSeek, ChatGPT, and Meta first appeared on Finblog.