Good morning, London. It's Friday, and while the skies are partly cloudy, the world of artificial intelligence remains anything yet dim. Alan Turing once remarked "I have such a stressful job I do not have much time for anything else yet work." A sentiment that resonates deeply today as we navigate the complexities of artificial intelligence—a field that demands our attention and clarity amidst the noise. Here at Turing's Torch, our mission is to demystify this rapidly evolving landscape, ensuring you stay informed without the overwhelm. Tired of drowning in artificial intelligence headlines? Ready for clarity, insight, and a direct line to the pulse of innovation? Welcome to Turing's Torch: artificial intelligence Weekly! I'm Jonathan Harris your host and I'm cutting through the noise to bring you the most critical artificial intelligence developments explained analysed and delivered straight to you. Let's ignite your understanding of artificial intelligence, together. We seem to be swimming in a sea of artificial intelligence announcements this week as a result let's try to make some sense of it all. There's a lot of hype, as always, yet also some genuine shifts happening under the surface. Let's begin with the sheer scale of investment that's being talked about. SoftBank, never a firm to shy away from a bold move, is reportedly considering putting up to thirty billion dollars into OpenAI. That's on top of existing plans to raise a hundred billion, potentially giving the company a valuation north of eight hundred billion. To put that in perspective that's an enormous amount of money even by Silicon Valley standards and from a firm whose previous artificial intelligence investments haven't always been…shrewd. What does that kind of funding actually mean? Well it gives OpenAI considerable resources to pursue research and development perhaps even some blue-sky thinking without the immediate pressure of commercial return. It also acts as a signal to other investors potentially driving up valuations across the board and accelerating the hunt for talent and intellectual property. Yet it also raises some serious questions about the concentration of power within the artificial intelligence sector. A single company with that level of funding has the potential to shape the direction of research development and deployment in ways that might not necessarily be in the best interests of everyone. We've seen similar patterns in other industries. The history of computing is littered with examples of dominant players shaping markets and stifling innovation. What makes this different is the sheer speed at which it is all happening, and the pervasive nature of the technology itself. This isn't simply a better word processor or a faster spreadsheet it's a fundamental shift in how we interact with information create content and even make decisions. One wonders, of course, what SoftBank expects in return for such a vast sum. Perhaps they see a clear path to profitability or maybe they're simply betting on the long-term potential of the technology regardless of immediate financial returns. Either way, it's a high-stakes gamble, and one that will be watched closely by everyone in the industry. And while we're talking about large language models Alibaba has announced a new one called Qwen3-Max-Thinking which they claim offers advanced reasoning capabilities. What this means in practice is that the model is designed to think more deeply about a problem by controlling the depth of its reasoning process. It also has built-in tools for searching remembering and even executing code all intended to help it handle more complex tasks without needing external software. The idea is to create a system that can actively engage with tasks and make better decisions. Now, the potential impact here is considerable. Businesses are constantly looking for ways to automate complex decision-making and if this model lives up to its promise it could significantly alter how they use artificial intelligence. It's part of a wider trend toward more sophisticated models that combine computational power with practical functionality. Of course, the question remains whether it can live up to the hype. We've seen many artificial intelligence models touted as revolutionary only to find their limitations exposed when faced with the messy realities of real-world data. One wonders if the name itself is perhaps just a little too on the nose. Still, it's worth watching to see whether it delivers or fades into the background noise. Google's DeepMind the outfit that previously brought us AlphaFold for predicting protein structures has now unveiled AlphaGenome a model intended to decipher the human genome. While AlphaFold concentrated on the three-dimensional shapes of proteins AlphaGenome attempts to understand the relationships between DNA sequences and their actual biological functions. This involves using hybrid transformers and U-Nets, which are types of neural networks, to interpret complex genomic data. The aim is to move beyond simply analysing static DNA sequences to understanding how those sequences actually behave. The potential impact here is substantial. It could lead to breakthroughs in understanding genetic diseases, evolution, and potentially pave the way for personalized medicine. If successful it would bridge a significant gap in biological research offering insights into how genetic sequences interact and function within biological systems. This could drive the development of new therapies and a deeper understanding of human biology. Of course, with such powerful tools come questions of ethics and potential misuse. The ability to decode and interpret the human genome raises questions about privacy, data security, and the potential for unintended consequences. It is easy to imagine scenarios where such technology could be used for nefarious purposes or where biases in the data could lead to discriminatory outcomes. One hopes the discussion of those risks is as energetic as the development of the technology itself. Ant Group the financial services people have announced a new artificial intelligence model called LingBot-VLA intended for controlling robots specifically those with two arms. The idea is to allow these robots to perform more complex tasks in real-world environments. LingBot-VLA is what they call a "foundation model," meaning it's been trained on a massive dataset – in this case around 20,000 hours of recordings showing people operating dual-armed robots. This data is meant to allow the artificial intelligence to understand both what it sees and what it's told to do as a result it can then manipulate objects in a useful way. Think of it as trying to teach a robot to assemble things or sort items, based on spoken instructions. The significance here is in the potential for increased automation. If robots can be easily trained to perform a wider range of tasks particularly in environments shared with humans then their adoption across various industries could accelerate. Logistics, manufacturing, even healthcare, could see a shift towards more automated processes. This has implications for jobs, of course, and also for the kinds of skills that will be valued in the future workforce. The development reflects a broader trend: the push to make artificial intelligence systems more adaptable and user-friendly. The more intuitive these systems become, the more readily they can be integrated into existing workflows. That said, it's worth remembering that the jump from a lab demonstration to a reliable, real-world application is often a considerable one. We've seen many artificial intelligence systems that perform impressively in controlled. settings yet struggle when faced with the unpredictability of the actual world. Will LingBot-VLA be able to cope with the inherent messiness of real-world tasks or will it prove to be just another overhyped demo? Only time will tell if it can truly bridge the gap between human intention and robotic action yet for now it's another step in the ongoing effort to automate the physical world. Nvidia the chip manufacturer best known for powering computer games has launched a suite of open-source artificial intelligence tools aimed at improving weather forecasting. They're calling it Earth-2. Now, what does that actually mean? Well, for decades, weather prediction has relied on massive supercomputers run by governments, performing incredibly complex calculations. Nvidia's move essentially puts artificial intelligence-powered forecasting into the hands of a much wider group. Think smaller companies, research institutions, even individual developers. The potential impact here is considerable. Imagine a small farming collective gaining access to the same level. of predictive power that was once only available to national weather services. They could fine-tune planting schedules, optimise irrigation, and generally make better decisions based on more accurate, localised forecasts. This could have significant implications for food security, resource management, and even disaster preparedness. And, of course, it could create new commercial opportunities for those who can build and deploy these models effectively. This move also fits into a larger pattern we're seeing, a democratisation of powerful artificial intelligence tools. The question, as always, is whether this increased access will truly lead to better outcomes. It's easy to imagine scenarios where biased data or poorly designed models lead to inaccurate or even harmful predictions. Just because you can forecast the weather with artificial intelligence doesn't necessarily mean you should at least not without proper oversight and validation. One wonders whether we'll soon be facing a deluge of conflicting weather forecasts, each claiming to be more accurate than the last. The irony, of course, is that we might end up more confused than ever. That's quite a lot of development, and that's just the tip of the iceberg. Let's shift gears and talk about how these technologies are actually being used, and some of the challenges that are emerging. There's been some interesting work published on differentiable computer vision, using a library called Kornia. Now, "differentiable computer vision" sounds rather grand, doesn't it? What it really means is making the entire image processing pipeline – from. initial image manipulation to feature extraction and matching – amenable to gradient descent. Gradient descent is simply a method of finding the minimum of a function. In this case it allows the system to learn and improve how it transforms and understands images based on feedback about its performance. The Kornia library allows developers to do this within PyTorch, which is a popular platform for machine learning research and deployment. A key aspect highlighted is the use of GPU acceleration which speeds up the processing considerably and the ability to synchronise images masks and keypoints – which are essentially tagged features of interest. The significance here lies in the potential for more accurate and efficient computer vision systems. Think about applications like autonomous vehicles, robotic surgery, or even advanced image editing software. The ability to fine-tune how a system "sees" and interprets visual data is crucial for reliability and precision. Furthermore, building these systems from the ground up within PyTorch gives researchers more control and a deeper understanding of the underlying processes. This is particularly important as computer vision becomes more integrated into our daily lives. It also reflects a trend we're seeing across various domains: a push towards greater customisation and control over artificial intelligence systems. Instead of relying on pre-built black-box solutions developers are increasingly looking for tools that allow them to tailor the technology to their specific needs. The claim, of course, is that it allows researchers and practitioners to experiment and innovate without being bogged down by rigid frameworks. Which sounds ideal, until you factor in the time and expertise required to build everything from scratch. One can't help yet wonder if for many applications a slightly less customisable yet more readily available solution might still be the more practical choice. DeepSeek a company you may not have heard of has released a new version of its optical character recognition software imaginatively named DeepSeek OCR 2. Now the original version was apparently quite good at extracting text from documents and making them smaller yet struggled with anything remotely complex such as reading order or interpreting multi-column layouts. Optical character recognition, or OCR, is essentially teaching a computer to ‘read' a scanned image of text. Think of it as the digital equivalent of teaching a child to read. The software analyses the image, identifies the letters, and then converts them into editable text. This is incredibly useful for digitising old documents, processing invoices, or any situation where you need to extract text from an image. Yet, as anyone who has tried to copy text from a PDF knows, it is not always reliable. The problem arises when documents have complex layouts, with multiple columns, tables, or unusual formatting. Previous OCR software often got confused, garbling the text or misinterpreting the order in which it should be read. DeepSeek's update seems to address these issues aiming for a more human-like reading experience which is to say it attempts to understand the layout of the document and extract the text in the correct order. This matters because industries like law, finance, and academia rely heavily on accurate document processing. Imagine a lawyer trying to extract information from a contract with complex clauses and tables or a researcher analysing a historical document with faded text and unusual formatting. Inaccurate OCR can lead to errors, wasted time, and potentially even legal or financial consequences. If this new version works as advertised, it could significantly improve efficiency and reduce the need for manual correction. The broader trend of course is towards automation and the increasing reliance on artificial intelligence to perform tasks that were once the domain of humans. As these systems become more sophisticated, they have the potential to transform industries and reshape the way we work. It's worth remembering though that even the most advanced algorithms are still prone to errors and a healthy dose of scepticism is always warranted. I'll believe it when I see it correctly interpret my grandmother's handwritten recipes. The UK government is experimenting with artificial intelligence to help citizens navigate public services. The Department for Science, Innovation, and Technology has commissioned Anthropic, one of the artificial intelligence companies, to build an artificial intelligence assistant. Now, when they say 'artificial intelligence assistant', think of it as a sophisticated chatbot. The idea is that instead of wading through endless websites and forms you could simply ask the artificial intelligence a question and it would guide you to the right information or service. This is meant to streamline things like applying for benefits, renewing a passport, or understanding tax requirements. The implications are considerable. If successful, this could save citizens time and reduce frustration. It could also free up government employees to focus on more complex tasks. That said, there's a risk that it could create new problems. What happens if the artificial intelligence gives incorrect advice? Who is responsible? And how do you ensure that everyone, regardless of their digital literacy, can access and use the system effectively? There's also the issue of bias. These systems are trained on data, and if that data reflects existing inequalities, the artificial intelligence could perpetuate them. We have seen other attempts at deploying artificial intelligence in government services and many have stalled in the pilot phase never making it to full deployment. The choice of Anthropic which has a reputation for responsible artificial intelligence development suggests the government is trying to get ahead of some of the ethical and safety concerns. Ultimately, this project will be judged on whether it actually improves the citizen experience. It is not enough for it to be technologically impressive; it has to be genuinely useful and accessible. One hopes it is more helpful than the average automated phone menu. The big retail chains appear to be diving headfirst into artificial intelligence, even if it means handing over some control. We're seeing major players like Walmart Target and Etsy partnering with the likes of Google and Microsoft to integrate artificial intelligence platforms directly into their online operations. What this means in practice is that artificial intelligence is increasingly handling tasks previously done by humans things like suggesting products managing inventory and personalizing the shopping experience. Instead of browsing a website curated by a team of merchandisers, or getting recommendations from a salesperson, you're interacting with an algorithm. The stakes are considerable. For retailers the promise is greater efficiency better data analysis and the ability to scale personalized services to a much larger customer base. Yet it also means ceding control over the customer relationship, and potentially handing over valuable data to third-party artificial intelligence providers. For consumers it could mean a more streamlined and convenient shopping experience yet also one that feels less personal and perhaps more susceptible to manipulation. This trend aligns with the broader push towards automation in various sectors, and the increasing reliance on artificial intelligence to drive decision-making. It also raises familiar questions about data privacy, algorithmic bias, and the potential for job displacement. We've seen similar moves across other industries, each with its own set of trade-offs. One can't help yet wonder if we'll eventually find ourselves nostalgic for the days when shopping involved, you know, actual human beings. It seems that even retail is now determined to re-engineer itself as a data optimisation problem. The recent cold weather in the United States has shown how airlines are using artificial intelligence to manage disruptions. Rather than relying on people alone, some airlines are using artificial intelligence systems to predict and respond to weather-related problems in real-time. In practice, this means analysing huge amounts of data, including historical flight patterns, weather forecasts, and current operational information. The goal is to forecast delays and cancellations before they actually happen, allowing airlines to adjust schedules and reroute flights in advance. artificial intelligence-powered chatbots are also being used to handle customer service inquiries, freeing up human agents to deal with more complex problems. The impact is potentially significant. Airlines can minimise disruption to passengers, improve customer service, and operate more efficiently. Given the increasing frequency of extreme weather events, it's becoming essential for airlines to adopt these technologies. One wonders though whether this apparent embrace of artificial intelligence is simply a way for airlines to offload responsibility for inevitable delays and cancellations onto a supposedly neutral algorithm. After all, blaming the machine is far easier than admitting to poor planning or inadequate staffing. Databricks, the data and artificial intelligence company, suggests that businesses are starting to favour as a result-called agentic systems in their artificial intelligence deployments. Now, what does that mean? Well, initially, many companies jumped on the generative artificial intelligence bandwagon, expecting a revolution. What they often got were chatbots that didn't quite work and pilot projects that never really took off. "Agentic systems," in theory, offer a more sophisticated approach. Instead of just spitting out text or answering simple questions these systems are designed to operate with more autonomy making decisions and adjusting workflows on their own. Think of it as moving from a helpful assistant to a more proactive, self-managing colleague. The real impact here is about control and efficiency. If these systems can truly integrate into decision-making processes, they could streamline operations and potentially give companies a competitive edge. The shift also suggests a growing maturity in how businesses view artificial intelligence. They're moving beyond mere experimentation and looking for tangible results. This move towards more autonomous systems also raises questions about transparency and accountability. If artificial intelligence is making decisions, how do we ensure those decisions are fair and unbiased? How do we audit these systems to understand why they made a particular choice? It seems we're less concerned with whether artificial intelligence can do something, and more concerned with whether artificial intelligence should do something. Artificial intelligence agents are spreading rapidly inside corporate networks, and this is causing headaches for chief information officers. These aren't just pieces of software, yet autonomous entities that can make decisions and take actions on their own. To put it plainly, different departments are adopting these technologies at speed, yet without any central control. The risk is that these digital assistants, operating independently, can cause operational problems and security vulnerabilities. It's a bit like the early days of cloud computing when employees were installing unapproved applications without oversight except this time the applications can think for themselves. The core issue is that existing governance frameworks simply weren't designed for this level of autonomy. What was once a manageable IT landscape is becoming a sprawling, decentralised network of digital actors. And if these actors are not properly monitored if their actions aren't logged and audited then the company is exposed to potential risks ranging from data breaches to regulatory non-compliance. The stakes are high because as companies become more reliant on artificial intelligence to improve productivity any failure to govern these agents properly can quickly turn into a serious liability. This situation highlights a familiar pattern: technological innovation outpacing our ability to manage its consequences. There's a rush to embrace the benefits of artificial intelligence, yet not enough attention is paid to the potential downsides. One might even say that some executives are as a result dazzled by the. promise of increased efficiency that they're willing to overlook the inherent risks. The world of cybersecurity is seeing increased interest in as a result-called defensive artificial intelligence. Essentially defensive artificial intelligence involves using algorithms to sift through enormous quantities of data much faster than any human team could manage looking for unusual patterns that might indicate a threat. This isn't about replacing human analysts; rather, it's meant to augment their abilities. The systems are designed to learn and adapt as new threats emerge, rather than relying on pre-programmed responses to known attack signatures. This matters because the volume and sophistication of cyberattacks are constantly increasing. Traditional

Well, another week, another deluge of artificial intelligence news. Sorting signal from noise becomes more vital with each passing day. If you'd like a daily digest of the most important developments stripped of the usual hyperbole you can sign up for my free artificial intelligence briefing at jonathan-harris.online. And for those seeking a more considered exploration of artificial intelligence's transformative potential particularly in the context of space exploration you might find my book "Beyond Earth: How artificial intelligence Is Transforming Space Exploration" a worthwhile read. It's available at books.jonathan-harris.online/ai-space. That's it for this week's Turing's Torch. Keep the flame burning, stay curious, and I'll see you next week with more artificial intelligence insights that matter. I'm Jonathan Harris—keep building the future.