How language is hiding the real internet from you
When you go online, it feels like you're accessing all the world's information. But you form social media relationships based on shared language. You search Google with the language you think in. And algorithms built to maximise attention have no reason to recommend what you won't understand. So, most of the internet remains out of sight, on the other side of a language filter – and you're missing far more than content.
Most internet activity is concentrated on a small number of large platforms, and from our linguistically siloed perspectives, it's easy to assume that everyone uses them in similar ways. But why should that be true? We expect music, literature and cuisine to vary between cultures, after all, so why not the internet?
In an upcoming paper, our team at the University of Massachusetts Amherst's Initiative for Digital Public Infrastructure has uncovered stark differences in how different cultures harness the internet. With more research, it may reshape how we think about the services that dominate the web. We're only just beginning to understand the implications.
Try Our AI Features
Explore what Daily8 AI can do for you:
Comments
No comments yet...
Related Articles
Times
28 minutes ago
- Times
What happened when we let Times readers bet $10,000 on stocks
If you could go back in time to the day of the 2024 US election — armed with $10,000 and everything you know now — how much money could you make in the stock market? That is the challenge we set Times readers last week. Buy and sell any stock in the S&P 500 or the FTSE 100 and jump forward in time, watching how the actions of the Trump administration have affected global markets. You can play it here. So, how did readers get on as stock market traders? The average player managed to turn their $10,000 investment into about $14,000 over the eight simulated months, meaning that they generated $4,000 profit — a return of 40 per cent. This is a pretty good return, even if readers knew exactly what would happen: $10,000 invested in the S&P 500 would have returned $10,787, and the same in the FTSE 100 would have resulted in $11,095 — although $10,000 in bitcoin, turning into $16,931, would have beaten the lot. Trump's tariffs resulted in a frenzy of trading, helping Barclays to a 23 per cent jump in pre-tax profits in the first half of the year. There are, of course, some objectively 'better' stocks to pick. The US AI firm Palantir, co-founded by the Trump ally Peter Thiel, has more than tripled its value since the election. Other US AI firms and computer chip manufacturers (such as Super Micro Computer Inc and Jabil Inc) have also experienced strong growth, as well as companies in AI adjacent sectors, particularly energy stocks such as the General Electrics power spin-off GE Vernova Inc, which has benefited from increased demand from AI data centres. • The cheap and easy way to invest (without the risk) A significant number of readers wisely put all their money into Palantir, yielding returns of 200 per cent. But the highest profits were for those who traded repeatedly at optimum times. Only two players turned their $10,000 into more than $100,000; at the time of writing, the top score was an eye-watering $441,681. The highest possible score, buying the best performing stock day after day, was well over a trillion dollars. Compounding returns, to paraphrase Albert Einstein, really are the eighth wonder of the world. For data privacy reasons, we don't know which stocks the top performer picked. However, on average, those who played did not opt for high-risk, high-reward US tech stocks. In fact, most readers put their money into recognisable UK companies that had a track record of doing well. Readers rightly focused on UK defence stocks such as Rolls-Royce, BAE and also Babcock, the best performing FTSE 100 company during the period. All those stocks boomed after Trump's realignment of global defence. • Bull or bear? Find out what kind of investor you are Other top performers picked out by readers included Airtel Africa, International Airlines Group (which owns British Airways) and the mining firm Fresnillo. For the FTSE at least, readers had a good sense of where the gains have been, with most of the ten best-performing stocks over the period being in the most-bought list. The above looks at which stocks had the most money invested. But the list of stocks that were bought and sold the most — that is, with the highest number of trades — shows a preference towards US growth firms. Top of this list is Coinbase, the cryptocurrency exchange, and Elon Musk's Tesla, both of which have share prices inherently tied to unpredictable and volatile entities. Readers also explored options with oil and steel stocks. Due to tariff changes, these companies were particularly volatile. Many readers bought BP, others looked at Chevron or the US steel manufacturer Steel Dynamics. There were good profits to be had with all these, but you'd need to buy and sell at specific times to really make a good profit. • I want to invest in Europe's comeback. Where do I start? These lists also feature unexpected appearances from 3M, 3i Group, AES Corporation and APA Corporation. Theses companies didn't perform particularly well over the period and their inclusion is likely to be due to their names appearing at the top alphabetically, rather than being selected as part of a broader investment strategy.
Geeky Gadgets
an hour ago
- Geeky Gadgets
ChatGPT-5 vs Claude vs Qwen : The Hidden Costs of Picking the Wrong AI Model
What if the future of app development hinged on the AI model you choose? With the rapid evolution of artificial intelligence, developers are faced with a critical question: which model delivers the best balance of speed, reliability, and cost-effectiveness? In the race to build smarter, faster applications, three contenders—ChatGPT-5, Claude, and Qwen—have emerged as frontrunners. Each features unique strengths, yet their differences could mean the success or failure of your next project. Imagine building a tool like 'Newsletter Digest,' a web app designed to summarize newsletters, only to discover that your chosen AI model struggles to deliver functional results. The stakes are high, and understanding these models' capabilities is no longer optional, it's essential. In this comparative overview, Rob Shocks explains how these AI models stack up in real-world scenarios, using the development of 'Newsletter Digest' as a case study. You'll uncover insights into their performance metrics, strengths, and limitations, as well as the hidden costs that could impact your budget. Whether you're a developer seeking reliability, a strategist prioritizing creativity, or someone chasing speed, this breakdown will help you navigate the trade-offs. By the end, you might find yourself rethinking what matters most in an AI partner: precision, adaptability, or efficiency? AI Model Comparison Summary Project Overview: Developing 'Newsletter Digest' 'Newsletter Digest' is a web application that connects to Gmail, aggregates newsletters, and summarizes key stories for users. It also allows users to view individual newsletters in a streamlined interface. The app was developed using a modern tech stack to ensure scalability, responsiveness, and ease of use: A React-based framework for building a dynamic and efficient front end. A React-based framework for building a dynamic and efficient front end. Tailwind CSS: A utility-first CSS framework that enabled clean, responsive design. A utility-first CSS framework that enabled clean, responsive design. Neon: A Postgres database solution for managing user and application data. A Postgres database solution for managing user and application data. Prisma: A database toolkit that simplified schema generation and management. A database toolkit that simplified schema generation and management. Clerk: A user authentication and management tool integrated with Stripe for seamless billing functionality. The core functionality of summarizing newsletters relied on the capabilities of Claude, GPT-5, and Qwen Coder. These models were tested for their ability to deliver a functional app, their speed, usability, and cost-effectiveness. Comparing the AI Models Claude (Opus 4.1 and Sonnet 4) Claude proved to be the most reliable and mature AI model for this project. It excelled in generating a fully functional app with minimal errors, particularly in tasks such as database schema generation and implementing app features. Its workflow required fewer corrective prompts, which significantly reduced development time. However, this high level of performance came at a premium, as Claude was the most expensive model in the comparison. For developers prioritizing reliability and efficiency, Claude offers a robust solution, albeit at a higher cost. GPT-5 GPT-5 demonstrated strong strategic thinking and versatility but fell short in implementation compared to Claude. While it made reasonable progress in building the app, it required more corrective prompts and encountered occasional errors during the setup process. The Cursor CLI, used to interact with GPT-5, is still in beta, which introduced some glitches and limited functionality. Despite these challenges, GPT-5 showcased its ability to handle complex tasks with guidance, making it a viable option for developers who can invest additional time in troubleshooting and refinement. Qwen Coder Qwen Coder stood out for its speed, delivering rapid responses during the development process. However, it struggled to produce a complete and functional app. Its outputs were often incomplete, and it frequently stopped abruptly, requiring significant manual intervention to fill in the gaps. While its speed is promising, Qwen's lack of reliability and functionality placed it behind both Claude and GPT-5 in this comparison. Developers seeking a balance between speed and reliability may find Qwen less suitable for complex projects. ChatGPT-5 vs Claude vs Qwen Watch this video on YouTube. Stay informed about the latest in AI Models Comparison by exploring our other resources and articles. Performance Metrics The three AI models were evaluated based on their overall performance in building 'Newsletter Digest.' The results revealed clear distinctions in their capabilities: Claude: Delivered the most reliable and functional app with minimal errors and a smooth workflow. Delivered the most reliable and functional app with minimal errors and a smooth workflow. GPT-5: Ranked second, offering decent progress but requiring more corrections and encountering occasional errors. Ranked second, offering decent progress but requiring more corrections and encountering occasional errors. Qwen: Ranked third, excelling in speed but failing to deliver a complete and reliable app. Claude's ability to produce a polished and functional app with fewer iterations made it the top performer, while GPT-5's versatility and Qwen's speed highlighted their respective strengths and limitations. Cost Analysis Cost is a significant factor when selecting an AI model, particularly for projects with budget constraints. Here's a breakdown of the pricing for each model: Claude (Opus 4.1): $15 per million input tokens, $75 per million output tokens. $15 per million input tokens, $75 per million output tokens. GPT-5: $0.125 per 1,000 input tokens, $0.10 per 1,000 output tokens. $0.125 per 1,000 input tokens, $0.10 per 1,000 output tokens. Qwen: Pricing varies but is generally more affordable than Claude and GPT-5. While Claude was the most expensive option, its superior performance justified the higher cost for this project. GPT-5 offered a more affordable alternative with slightly reduced effectiveness, while Qwen's lower cost reflected its limited functionality and reliability. Insights on Development Tools The development process also highlighted the importance of the tools used alongside the AI models. These tools played a critical role in streamlining the workflow and making sure a smooth development experience: Clerk: Simplified user authentication and integrated seamlessly with Stripe for billing, reducing the complexity of managing user accounts. Simplified user authentication and integrated seamlessly with Stripe for billing, reducing the complexity of managing user accounts. Cursor CLI: Provided access to multiple AI models but, as a beta tool, had limited features and occasional glitches that impacted usability. Provided access to multiple AI models but, as a beta tool, had limited features and occasional glitches that impacted usability. Neon (Postgres DB): Worked effectively for database management, especially when paired with Prisma for schema generation and maintenance. While most tools performed well, the beta status of Cursor CLI highlighted the need for further refinement to improve its reliability and feature set. Looking Ahead: Future Considerations As of August 2025, the AI landscape continues to evolve, with emerging models like Grok and Gemini showing potential to disrupt the market. Developers should remain vigilant for advancements in AI capabilities and pricing structures. For now, Claude remains the preferred choice for projects requiring maturity, reliability, and comprehensive functionality. GPT-5 offers a strong alternative for tasks that demand strategic thinking and creative input, while Qwen Coder's speed may appeal to developers working on less complex applications. Continued refinement of tools like Cursor CLI and exploration of new technologies will further enhance the app development process, paving the way for more efficient and innovative solutions. Media Credit: Rob Shocks Filed Under: AI, Top News Latest Geeky Gadgets Deals Disclosure: Some of our articles include affiliate links. If you buy something through one of these links, Geeky Gadgets may earn an affiliate commission. Learn about our Disclosure Policy.
Finextra
6 hours ago
- Finextra
Why Traditional Mobile Banking Apps Are Reaching Their Limits and How Agentic UI Will Replace Them: By Paul Shumsky
Every major shift in digital banking has started with a problem that could no longer be ignored. Moving from branches to online banking addressed the need for convenience. The move from online to mobile banking apps gave customers access to their finances anywhere and anytime. Now a new challenge is emerging. Even the best mobile banking apps are beginning to show their limits. Neobanks like Revolut and Nubank have set the standard for modern design and strong user experience. Yet as they expand into super apps offering trading, insurance, travel bookings and shopping, the trade-off becomes clear. Each new product competes for space on the screen. Menus get deeper, journeys get longer and customers struggle to find the value that drew them in. This is not a problem that affects only challengers. Traditional banks face it too. When more becomes too much To keep things manageable, many banks have split their offerings into separate apps. One for personal banking. One for businesses. Another for young customers. Another for wealth and investments. This approach makes each app less cluttered, but it creates a new problem: fragmentation. Customers need multiple logins, must switch between different designs, and face inconsistent experiences. It is also expensive. Separate apps require separate development teams, separate technology platforms and separate infrastructure. That is duplication of cost without duplication of value. The next breakthrough will not be about creating more apps or adding more menu tabs. It will be about creating a single intelligent experience that recognises each customer, understands the products they use and knows how to make their life easier. Personalisation that means something Banks often talk about personalisation in terms of letting customers change colours, switch to dark mode or rearrange widgets. In reality that is not personalisation. Studies consistently show that the vast majority of customers never change default settings. True personalisation is proactive. It guides customers through their financial journeys and anticipates what they will need next. A small business owner logging in on payroll day should see cash flow projections and quick-pay links before anything else. A frequent traveller should see currency exchange, travel insurance and lounge access in the weeks before a trip. Enter agentic UI Agentic UI represents the next step in the evolution of mobile banking. It is a conversational, adaptive, AI-powered interface that learns from each customer's profile, behaviour and context in real time. Instead of navigating static menus, customers interact with an intelligent agent. The interface adapts to who is using it, when they are using it and why they have come. This is the type of experience that Finpace helps financial institutions bring to life. With the right integration, one app can serve every customer segment and every product line without compromising on simplicity or clarity. Benefits for customers and for the business For customers, Agentic UI means fewer clicks, faster answers and the feeling that their bank truly understands them. For the bank, it brings higher engagement because customers do not have to search for value, lower support costs because there are fewer 'how do I…' queries, reduced technology spend by removing the need for multiple apps, and faster launches of new products because they can be introduced contextually to the right customers. Platforms like Finpace, Kasisto, Personetics, and Glia are already making these capabilities possible today. Finpace focuses on unifying multiple product lines into a single adaptive banking app, Kasisto enables natural language conversations that feel human, Personetics powers hyperpersonalized recommendations and goal-based journeys, and Glia integrates real-time AI assistance into the customer's digital session. Together, they are proving that Agentic UI is not a distant vision but an achievable standard for forward-thinking financial institutions. Learning from other industries Other industries have already moved in this direction. Netflix recommends what to watch without asking. Google Maps adjusts travel time based on live traffic. These are systems that adapt to the user, not the other way around. Banking can work the same way. The difference is that financial journeys are more complex, more regulated and often more sensitive. Agentic UI, enabled by platforms like Finpace, can make those journeys simple, relevant and compliant without losing the personal touch. The road ahead The case for Agentic UI is becoming stronger every day. As super apps grow and traditional banks seek to unify their offerings, one adaptive app that delivers the right experience to the right customer at the right time is becoming the logical next step. Those who start exploring this now will have a clear advantage, not only in customer satisfaction but also in cost efficiency and time to market.
