Ansa Zulfiqar

ChatGPT Codex emerges with the biggest upgrade to how developers write code. This isn’t your average software update. OpenAI’s newest upgrade instead turns their AI instrument into an entirely new beast: an independent coding partner, capable of thinking, planning, and carrying its plans out on its own. The days of dumb code completions are over. We are entering the age of agentic coding, in which artificial intelligence not only recommends snippets of code but prepares entire development projects from beginning to end. ChatGPT Codex 2025: What’s New in OpenAI’s Next-Gen AI Coder OpenAI’s latest is an RPC/incorefication of all our lives. We don’t deserve OpenAI, which just proved Elon Musk to have one of the worst wrongs in the history of disastrously stupid people doing disastrous stupidity. And now the ChatGPT Codex functions as a cloud-based development AI that can multitask multiple parallel tasks at once. Core Architecture Improvements If you want to compare it this way, the OpenAI you knew is as dead and gone as a thumbprint on a freeway. This new Codex-1 Model Architecture brings with it enriched functionality that is going to shape how we think of automated coding. The system is now a real programming aid rather than just a suggested tool. The AI chatbot now understands project context across entire codebases. It remembers previous conversations, learns your coding style, and maintains consistency throughout long development sessions. This development AI can handle complex refactoring operations while maintaining code quality standards. READ ALSO: Budget-Friendly Triple-Screen Laptop Launches at $700 with Intel Alder Lake Power — A Portable Workstation for Multitaskers What Makes This Different Old-style coding software needed an all-the-time human touch. You’d input a prompt, receive a response, and then you would guide the next step. ChatGPT Codex completely breaks this cycle with its act-based coding method. The current system functions more like a collaborator rather than a code-writing tool. It gets business needs, turns them into tech specs, and delivers a full solution with very little direction. Feature Old Codex New ChatGPT Codex Task Handling Single requests Multiple parallel tasks Context Awareness Limited Full project understanding Integration API required Direct ChatGPT account access Collaboration One-way prompts Two-way conversation Agentic Coding 101 — How Codex Redefines the Future of Development Agentic coding is a fundamental revolution of automated coding. This is a departure from the traditional programming assistance tools in that the AI is actually empowered to open its own doors and do complex development tasks. Understanding the Agentic Difference Traditional AI is something akin to a smart autocomplete. You begin typing, and it completes your thought. ChatGPT Codex works differently. It’s like having a junior developer who can take high-level requirements and translate them into working code. This isn’t just a system that regurgitates prompts….and takes an active part in the coding operation itself, raising queries, suggesting alternatives, and spotting potential stumbling blocks before they trip you up. The Four Pillars of Autonomous Development This is the basis of ChatGPT Codex. The AI Evolution system is an AI development, decision, and learning system that is truly evolutionary. Evolution divides AI design into “what to do” and “how to do”; the how is evolutionary. Instead of sitting around waiting for detailed instructions, it draws its own map from your initial description. Self-Directed Execution enables the coding tool to operate on its own once it reads your situation. It writes fully-featured features, it handles errors, and it adheres to best practices without constant watching. Such a process could be made even more effective by allowing an opportunity for the automated coding systems to test their own output using Continuous Self-Correction. When it stumbles upon bugs or inefficiencies, it automatically fixes them before presenting the ultimate result. Context Preservation ChatGPT Codex never forgets anything about your project. It knows how new code impacts existing features and keeps the overall codebase clean and finite. READ ALSO: Samsung Galaxy S26 Ultra Leak Reveals Exclusive Snapdragon Customization That Could Outrun Every Android Flagship Real Impact on Developers The revolution in automated coding has transformed daily routines significantly. “Top engineers tell us they can now get their work done — design, code, and ship beautiful software without months of planning. Ideal for tasks of all types, including competitive analysis. This has led to higher-quality code, as we’re always following best practices and general industry standards via the AI tool. The system is also a great programming tutor, explaining its reasoning and teaching new tricks as you play. From Prompt to Pull Request: Codex Autonomy in Real-World Action From Prompt to Pull Request: The complete development lifecycle is reshaped by ChatGPT Codex. The following is a step-by-step description of the AI-run coding initiative, from beginning to end, demonstrating the kind of agentic coding powers that can only be achieved through its usage. Phase 1: Requirements Analysis In there, you say what you’d like, in plain English. AI in development. The development AI writes tech specs, automatically maps dependencies and potential conflicts, estimates the timeline with resources, and alerts when a project is likely to miss the timeline. This coding device doesn’t simply accept orders quietly. It asks smart questions, offers good enough alternatives, and generally helps spec out stuff before writing actual code. Phase 2: Autonomous Development And here is where ChatGPT Codex actually really shines as an automated coding answer. It takes care of the entire feature development flow – both the steps of doing it the right way (error handling, TDD best practices, generating well-documented code) and of doing the mundane ones, like opening a PR with a detailed description in it. Coding conducted by AI is patterned after the existing codebase but incorporates up-to-date best practices. “It’s like having a development AI that doesn’t get tired and always writes code according to the normal standards.” Phase 3: Integration & Deployment The automated coding system doesn’t stop at writing the code. It generates an automatic pull request, works with CI/CD to merge changes faster, detects performance improvement, and proposes

ChatGPT just got smarter – and now you get to decide exactly how smart it should be. OpenAI’s new water-testing feature, dubbed “Thinking Effort,” allows you to change the level of intellectual horsepower your AI assistant devotes to each subsequent answer. It’s kind of akin to opting between a short answer and a professor’s long explanation. This groundbreaking control manual system arose after GPT-5’s release was mocked for being too slow and over-engineered. Users said they were struggling to obtain basic information without receiving needlessly complicated responses. Today, ChatGPT introduces a new set of 4 tiers of level of effort, putting even more fine-grained control into how much computational budget is applied to your question. It is a sea change in strategy for A.I. performance, the feature. Instead of generic replies, you decide whether ChatGPT should sprint to a response or marathon through deep consideration. This UX enhancement helps resolve some significant reliability issues, and at the same time, creates new opportunities for casual and professional users. READ ALSO: Latam-GPT Gets a New Brain: Latin America’s Open-Source AI Built to Understand Local Languages & Culture What “Thinking Effort” Actually Does THE THINKING EFFORT CHATGPT2 feature will transform the way ChatGPT handles your input. Rather than processing with a fixed amount of computational intensity, the system now performs processing steps with your selected effort level. Here’s how the computational budget system works: Effort Level Computational Units Processing Time Best For Light 5 units 1-2 seconds Quick facts, simple questions Standard 18 units 3-5 seconds General tasks, conversations Extended 48 units 8-12 seconds Complex analysis, research Max 200 units 15-30 seconds Expert-level problems As you choose a higher level of thinking effort, ChatGPT is then doing more reasoning steps behind the scenes. Your invisible layer. When people say that what really happens is this black-box model that’s consuming all kinds of data, what they’re usually talking about is a huge computational process – things like fact-checking multiple solutions, or self-verification: that it all hangs together, and that’s all invisible to you. The system takes advantage of what OpenAI refers to as “reasoning tokens,”  the unseen thinking steps that occur before generating your visible response. As a result, at Max, ChatGPT may produce thousands of these hidden tokens, essentially arguing with itself about the best answer. This method is a way to address a key performance challenge in AI. Earlier versions either thought too little (missing subtleties) or too much (devoting time to easy questions). Now ChatGPT thinks only as hard as your question requires. Light → Max: Pick the Right Thinking Level Once you know what each setting has to offer, picking the right “effort” level is a no-brainer. The trick is to match the level of computation to the level you really need. Light Thinking is great for the immediate stuff. Use it when you need ChatGPT to quickly look something up, make rudimentary computations, or work through basic questions. The reaction time is near instantaneous, and the conversations are as organic as they can be. Standard Thinking does almost all of everyone’s daily interactions just fine. And here the well-balanced users got each time a reasonable response each time without unnecessary delays. It’s great for jotting out emails, light researching, and general problem-solving, where you want a proven, accurate answer but don’t want to wait. READ ALSO: Scientists Unveil the World’s First 6G Chip: Blazing 100 Gbps Speeds for Future Networks Complexity Thinking Enables a Deeper Analysis of Complex ProblemsExtended thinking is the contrast of the second type of thinking, known as complexity thinking. Select this level when solving multi-step problems or when planning strategies for solution methods or detailed analysis. From there, ChatGPT will take additional time to think through multiple perspectives and different types of solutions. Find out more Max Thinking Max Thinking is expert advice to give you real confidence in answering even your most challenging questions. This level of computational power works great for high-level coding, scientific/numerical analysis, and in-depth research where quality is more important than speed. The intelligent plan is to begin low and increase if necessary. If the first answer from ChatGPT looks bad, just ask him the same question with a higher level of thought. This performance approach cuts down your timing and guarantees the right depth, whenever required. Why OpenAI Added an Effort Dial Thinking effort was a direct result of users after the GPT-5 release provided this feedback. Early versions erred on the side of analytical thoroughness at the expense of the user experience, leaving users demanding fast answers to easy questions. The first try was marred by reliability issues. ChatGPT tended to overcomplicate rather than simplify user input, which would result in poor user experiences. The AI appeared incapable of properly calibrating the computing power it devoted to the intensity of the task. OpenAI’s reply to its critics was about fine-grained control, not automation. Rather than attempting to read their users’ minds, they opted to give manual controls directly to users. This fine-tuning places control in the hands of users, the ones who know their own demands and deadlines best. The feature also responds to performance concerns coming from enterprise customers. Businesses require predictable AI competency across various tasks. A law firm reviewing contracts should want maximum analysis depth, whereas a customer service team would want to achieve the fastest response time for daily questions. This user-led philosophy of control is symptomatic of a larger transformation occurring within AI. Instead of building a series of ever-more-complicated robot overmen, OpenAI has opted to let users take the wheel of policy creation for them. The outcome enhances performance as well as user response in a variety of use scenarios. Real-world Uses: When to Make ChatGPT Think Harder Different professions and tasks benefit from specific effort level selections. Understanding these patterns helps you maximize AI effectiveness while minimizing wait times. READ ALSO: Artificial SuperIntelligence and the Control Dilemma: What No One Talks About Professional scenarios demanding Max’s effort Legal contract analysis

Latam-GPT is a new large language model being developed in and for Latin America. The project, led by the nonprofit Chilean National Center for Artificial Intelligence (CENIA), aims to help the region achieve technological independence by developing an open-source AI model trained on Latin American languages and contexts. “This work cannot be undertaken by just one group or one country in Latin America: It is a challenge that requires everyone’s participation,” says Álvaro Soto, director of CENIA, in an interview with WIRED en Español. “Latam-GPT is a project that seeks to create an open, free, and, above all, collaborative AI model. We have been working for two years with a very bottom-up process, bringing together citizens from different countries who want to collaborate. Recently, it has also seen some more top-down initiatives, with governments taking an interest and beginning to participate in the project.” The project stands out for its collaborative spirit. “We’re not looking to compete with OpenAI, DeepSeek, or Google. We want a model specific to Latin America and the Caribbean, aware of the cultural requirements and challenges that this entails, such as understanding different dialects, the region’s history, and unique cultural aspects,” explains Soto. READ ALSO: Scientists Unveil the World’s First 6G Chip: Blazing 100 Gbps Speeds for Future Networks Latam-GPT’s New Brain: Built for Latin America New Look & Feel Latam-GPT has a New Brain; it is now based on a completely revamped architecture, tailored to the Latin American context. While other models have Spanish as a secondary language, which treats regional languages of secondary importance, this AI model makes regional languages the first priority. The technical foundation runs on supercomputing infrastructure spread across Chile, Colombia, and Mexico. These data centers process training data that includes everything from Mexican telenovela scripts to Argentinian tango lyrics. The model understands that “¿Qué tal?” means different things in different countries. CENIA is responsible for technical development along with partner institutions such as the Universidad de Chile and Tecnológico de Monterrey. Their machine learning system prioritizes cultural and not just linguistic context. The model learns from local news sources, social media discussions, and academic papers penned by Latin American scholars. Feature Latam-GPT Global Models Languages Supported 15 indigenous + Spanish/Portuguese variants Limited regional support Training Data Sources 70% Latin American content <10% regional content Cultural Context Processing Native understanding Translation-based Open Source Availability Full access Restricted APIs The language model training phase lasted 18 months and used more than 2 million hours of computation. The researchers fed conversations from street vendors in Guatemala, academic papers from Brazilian universities, and poetry written in Quechua into the system. This shared AI model also has the effect of immediately benefiting everyone, creating the opportunity to build products people love and need even faster. If Mexican developers improve the machine understanding of business language, Colombian users get those improvements automatically. The AI community is more powerful when the wisdom of the crowd is harnessed. Why Latam-GPT Matters — Language, Culture, Sovereignty Latin America’s technology has always depended on solutions built elsewhere. When Microsoft’s Cortana couldn’t understand Chilean slang or Google Translate butchered Peruvian idioms, the region accepted these limitations as normal. Latam-GPT Gets a New Brain challenges this acceptance completely. Digital autonomy means more than just having local servers. It means having AI that understands why Mexicans celebrate Día de los Muertos differently than Guatemalans, or why Brazilian Portuguese sounds nothing like European Portuguese. These cultural requirements go far beyond simple translation. The tech sovereignty argument becomes clearer when you consider data privacy. Every conversation with ChatGPT or Claude sends Latin American data to US servers. Every business query reveals competitive intelligence to foreign companies. Latam-GPT gets a New Brain, keeps this data within the region. READ ALSO: Artificial SuperIntelligence and the Control Dilemma: What No One Talks About Cultural Intelligence That Actually Gets It Regional culture understanding shows up in countless ways: Recognizing that “empanada” recipes vary dramatically across countries Understanding political sensitivities around historical events Processing humor, sarcasm, and wordplay that doesn’t translate well Respecting indigenous spiritual concepts that Western AI models often misinterpret Cultural adaptation extends to business contexts, too. The model knows that Mexican business meetings start with a personal conversation. It understands that Brazilian negotiations involve relationship-building that other cultures might consider inefficient. Self-reliance in AI development creates jobs for regional talent. Instead of sending the brightest minds to Silicon Valley, Latam-GPT Gets a New Brain creates opportunities at home. Computer science graduates can work on cutting-edge NLP models without leaving their countries. How Latam-GPT Was Built: Data, Partners, and Infrastructure Building Latam-GPT Gets a New Brain requires unprecedented strategic partnerships across the region. The Chilean National Center coordinated efforts, but the real work happened through tech cooperation spanning borders. The regional database contains over 800 billion tokens of Latin American text. This massive data repository includes news articles from major regional outlets like Clarín, El Universal, and Folha de S.Paulo. Academic publications from 50+ universities provide scholarly depth. The biggest roadblock was a lack of high-performance computing infrastructure. This was no enterprise-level tech-giants-with-no-limits budget, so the project had to share resources. Specifically, Chile donated licenses of its National Laboratory for High Performance Computing. Colombia offered the use of its national supercomputing center. Mexico provided cloud infrastructure as part of its digital modernization efforts. The Collaboration Network University consortia on joint ventures formed the basis of model preference. Country-specific research groups focused their research on: Brazil: Portuguese language variants and cultural context Mexico: Indigenous language integration and preservation Argentina: Literary and artistic expression processing Chile: Technical infrastructure and algorithm training Joint ventures involved formal accords among governments. Regional development banks provided a total of $45 million for the project. Further support also resulted from partnerships with the private sector, such as telecommunications enterprises and education organizations. It was important that the data pool was curated well to balance quality with cultural representation. Millions of snippets of text were manually screened by teams to eliminate bias and ensure authenticity

Artificial superintelligence (ASI) is a hypothetical software-based artificial intelligence (AI) system with an intellectual scope beyond human intelligence. At the most fundamental level, this superintelligent AI has cutting-edge cognitive functions and highly developed thinking skills more advanced than any human. While ASI is still a hypothetical future state, there are several technological processes we have today that form the building blocks of ASI. But first, to illustrate how far off ASI is from this moment, it bears mentioning that the current level of AI is often referred to as Artificial Narrow Intelligence (ANI), weak AI, or narrow AI technologies. Weak AI is good for particular tasks such as playing chess or translating languages, but it cannot learn new skills or absorb profound knowledge of the world. It is based on a priori coded algorithms, data, and human intervention for its operation. Not all thinkers even agree on the possibility of something like an ASI. Human intelligence has a certain evolutionary origin and does not need to be ideal or universal. What’s more, we still don’t really understand how the brain works, which makes it hard to replicate it in software and hardware. READ ALSO: British AI Tool Could Revolutionize Manufacturing: Here’s What It Brings to Industry 4.0 What is Artificial Superintelligence? Artificial superintelligence (ASI)  where machines surpass the cognitive performance of humans in every intellectual domain, including creativity, general wisdom, and social skills. Unlike the AI tools that exist today, generative AI, where the tools are very good at very narrow things, ASI is a sea change in cognitive capability. Think of it this way: If existing AI is like a calculator that is really good at math, artificial superintelligence would be a combination of Einstein, Darwin, and Shakespeare, but thousands of times more intelligent. The Three Types of Superintelligent AI Type Definition Example Speed Superintelligence Human-level intelligence, but millions of times faster Processing years of research in minutes Collective Superintelligence Networks of human-level AIs working together Thousands of AI researchers are collaborating instantly Quality Superintelligence Intelligence that exceeds humans qualitatively Discovering solutions humans can’t even comprehend The tone of this piece intentionally confuses AGI with artificial superintelligence, a key distinction between the two being one of scope. Figure AGI is tied to human-level competence in variety of contexts. ASI transcends it entirely. Is Artificial Superintelligence Possible? Overall, the scientific consensus has changed drastically. Newer surveys of AI researchers find that 80 percent of the respondents think that a superintelligent machine is likely to arrive within just 50 years. More concerning? 25 percent expect it to occur in 20 years or sooner. Already, to experts, neural networks and deep learning have proved surprising. There are a number of ways GPT-4 can show emergent capabilities that were not specifically programmed, such as solving mathematical problems it was never shown during training. Current Evidence Supporting ASI Development Scaling Laws: Larger machine learning models consistently show improved performance Cross-Domain Transfer: Modern AI applies knowledge across unrelated fields Emergent Behaviors: New capabilities appear at specific parameter thresholds However, skeptics raise valid concerns. Consciousness is the basis of cognitive functions in biological brains, which we don’t understand, let alone know how to reproduce in artificial systems. READ ALSO: Apple MacBook Air With M4 Chip Launches: Faster, Smarter, and Sleeker Than Ever The Latest AI Trends Driving Superintelligent AI The rise of artificial superintelligence isn’t happening in a vacuum. Present AI breakthroughs are being set up as of now. Generative AI models, GPT-4 and Claude kind, are examples of advanced NLP. They are not merely manipulating text; they are reasoning, generating, and problem-solving across domains. Key Developments Accelerating ASI Multi-modal Integration: AI systems processing text, images, and audio simultaneously Agentic AI: Systems that can plan, execute, and adapt strategies independently AI-to-AI Communication: Machines developing their own protocols for collaboration Recursive Self-Improvement: AI systems optimizing their own architectures Neural networks are becoming more efficient through evolutionary computation approaches. These learning algorithms mimic natural selection, producing better AI designs automatically. Stuart Russell, AI safety expert, warns: “The real risk is not malice but competence—a superintelligent system pursuing goals misaligned with ours.” Pathways to Artificial Superintelligence But there could be many paths to artificial superintelligence. Knowing that these are some of the pathways helps us prepare for what’s to come. The Scaling Pathway Such a brute-force strategy presumes that larger models with more data will automatically get to artificial superintelligence. Today’s machine learning trends back that argument up. Timeline: Conservatively, 2040-2050, if scaling, management, and regulation remain at their present levels. The Brain Emulation Pathway Human brains are mapped and simulated in silico, and superintelligent AI arises by direct copying of the cognitive function. Timeline: Perhaps around 2050 to 2070, depending on the development of neuroscience and computing power. The Hybrid Pathway An intermixing of the two could potentially go a long way toward bootstrapping ASI. Timeline: Already starting with brain-computer interfaces and AI-aided research. Benefits of Artificial Superintelligence And artificial superintelligence could solve the most vexing problems facing humanity. The potential benefits are staggering. Scientific Acceleration Superintelligent AI might be able to condense a few centuries of scientific and technological progress into a few years. The various hard problems in physics, chemistry, or biology that perplex our brightest and sharpest minds might become the easiest to solve. Real Example: Alpha Fold used machine learning techniques to solve protein folding, a 50-year-old biological puzzle. Medical Breakthroughs Personalized medicine could become reality. A general artificial superintelligence could produce personalized treatments based on genetic, lifestyle and environmental factors. Climate Solutions A superintelligent AI could restructure the world in sustainable ways. From carbon capture to the distribution of renewable energy, ASI might help organize planetary-scale solutions. Economic Transformation Post-scarcity economics is a situation where artificial superintelligence works to allocate resources and organize production. READ ALSO: This New Windows Laptop Might Finally Beat MacBook Air in Portability and Power The Control Dilemma: The Hidden Crisis Here’s something that no one is discussing: The control problem isn’t something that’s coming up in the