Category: Uncategorized

Artificial intelligence systems like ChatGPT, Claude, Perplexity, Gemini, Copilot, and similar large language models (LLMs) have become, perhaps, a bit too present in our daily quest for feedback and answers. We enjoy using them because they generate original, easy-to-read, and ‘easy to believe’ responses. This versatility, however, masks a very profound limitation: these systems struggle to acknowledge the boundaries of their knowledge. They rarely say, ‘I don’t know,’ and when they do not know, they generate content that seems to have the same confident tone as what we believe to be reliable responses. As researchers, users, and developers of AI-based tools, the authors of this article aim to understand why AI systems exhibit this ‘overconfidence problem,’ or ‘omni-knowledge,’ what it means for users, and what can be done to address this significant usage limitation. The Architecture of Omni-knowledge: How Large Language Models Work To understand why AI systems struggle with accepting their own limitations, we must first understand how these systems work. Large language models are not knowledge databases in the traditional sense; they are pattern-matching and pattern-searching mechanisms that are trained on large amounts of data to predict outcomes given the relationship and proximity of words. This allows them to predict which words should answer a question. Otherwise said, the answer comes from the closeness of the question to the words available to answer it .1 When they do not know, they generate content that seems to have the same confident tone as what we believe to be reliable responses. However, this association of ideas finds its limits when the words and documents available are not sufficiently close to the question asked. This is when the phenomenon of AI hallucination appears, which means that AI generates plausible but entirely fabricated information .2 Remember that uncle who had ‘travelled the world and had all the answers’? This is an analogy of how this hallucination phenomenon happens: your uncle may have never been to India, but, given his experience in Pakistan, he projects the similarities and differences between one and the other based on information he has recollected from reading books, and he answers questions as if he had been there. This type of issue is acknowledged by OpenAI, saying that ‘hallucinations remain a fundamental challenge for all large language models’ even as capabilities improve, occurring when models ‘ confidently generate an answer that isn’t true ‘.3 The real issue is that theoretical work has demonstrated that hallucination is not merely a technical problem to be solved through better training, but rather ‘an innate limitation’ of LLMs because these hallucinations, as explained before, are at the basis of the functioning of the algorithms used by AI; it’s about finding close relationships and delivering an answer, forcing it to create inexistent realities 4 that are like the relationships it has found. This mathematical truth means that no amount of engineering can eliminate hallucination; it can only be reduced. This point is also reinforced by recent work from OpenAI, Why Language Models Hallucinate , which shows that hallucinations arise even when models are well trained, because prediction-based systems must generate a plausible continuation when certainty is low. What Can Be Done: Addressing AI Omni-knowledge Mitigating the confidence problem in AI systems requires coordinated action across multiple stakeholders: AI developers improving systems, users developing critical engagement practices, and institutional frameworks establishing appropriate use boundaries. From a technical perspective, AI could be given certain limitations, although none will completely cover all potential subjects and issues likely to be generated after a prompt is issued. Researchers at Oxford University have developed methods to detect when a

Posted in: Comics, Mad Cave Studios | Tagged: 51, Bludwire, Is Ted Okay, Junk Punch, Land Of Never, Last Starfighter Mad Cave June 2026 Full Solicits: Junk Punch, Land Of Never & Blüdwire Mad Cave Studios’ June 2026 solicits include Junk Punch #1, Land Of Never #1, Blüdwire #1, 51 #4, Is Ted Okay #5 and The Last Starfighter Article Summary Mad Cave Studios launches Junk Punch, Land Of Never, and Blüdwire with new #1 issues in June 2026. Fan favorites like 51 and Is Ted OK? return with new issues packed with mystery and action. Ray Fawkes brings The Phantom Vs The Red Dragons one-shot for a high-stakes South China Sea showdown. The Last Starfighter movie adaptation is reissued, plus continuing titles like Gatchaman and Flash Gordon. Mad Cave Studios’ June 2026 solicits and solicitations include the launches of Junk Punch #1 by Paul Tobin, Carlos Javier Olivares, Land Of Never #1 by Steve Orlando and Miguel Mora, Blüdwire #1 by Paul Allor and Ermitis Blanco, and The Phantom Vs The Red Dragons one-shot by Ray Fawkes and Lynne Yoshii… as well as the usual Gatchaman, Dick Tracy, Flash Gordon, Speed Racer, War Wolf, Dog Tag, Exit City, Honor And Curse, Terrorbytes, Barbarian Behind Bars, Exploit, Pretty Hate Machine, Planet Atmos, Racer X, and certain comics that have got all sorts of people excited, 51 and Is Ted OK? as well as a republication of Marvel’s The Last Starfighter movie adaptation by Bill Mantlo, Bret Blevins and Tony Salmons…… BLUDWIRE #1 (OF 5) CVR A VICTOR IBANEZ (MR) (W) Paul Allor (A) Ermitis Blanco (CA) Victor Ibanez Ace and Zora were built for men’s pleasure. Now they’re building a life of their own, one bullet at a time. When sex bot Ace mounts a dangerous rescue mission, she isn’t starting a revolution. She’s trying to save the woman she loves. But Ace and Zora are valuable property, and a corporate kill squad is hot on their trail. Bullets fly. Metal screams. Rivers run red with blood and coolant. From critically-acclaimed writer Paul Allor and artist Ermitis Blanco, Blüdwire is a tender love story surrounded by a hail of bullets. Love is love, but liberation is violent. $4.99 6/24/2026 JUNK PUNCH #1 (OF 5) CVR A CARLOS JAVIER OLIVARES (MR) (W) Paul Tobin (A) Carlos Javier Olivares, Colleen Coover (CA) Carlos Javier Olivares In a near-future world, peculiar new compulsions have arisen to afflict individuals in bizarre ways, such as a need to paint all pigeons pink or the unfortunate craving to tell the truth on dating apps. One such individual is Clara Castanelle, aka Junk Punch, who “suffers” an addiction to punching people square in their junk! From Eisner-winning author Paul Tobin and Probably Even More Talented Artist Carlos Olivares comes the story of one woman’s quest to solve a bizarre series of thefts, such as kisses stolen from willing and waiting lips! JUNK PUNCH! It’s entertainment with IMPACT! $4.99 6/3/2026 LAND OF NEVER #1 (OF 6) CVR A MIGUEL MORA (W) Steve Orlando (A/CA) Miguel Mora Six months ago, Jim Hoke’s daughter Wendy disappeared from her room. All Jim caught was a glimpse of a hulking figure at the window—the kind of late night vision that’s easily written off as a hallucination. No one believes Jim’s story—and six months later, the retired pathologist is just looking for anyone who’ll still listen. And even if he finds someone, should they believe him? Jim already retired early for planting evidence against an alleged killer. His word doesn’t mean much—and it means even less with him as a suspect in Wendy’s

From 19 to 20 March, Xi’an Jiaotong-Liverpool University (XJTLU) hosted the 2026 Postgraduate Research Symposium, bringing together nearly 550 PhD students from 13 universities and research institutes in China and abroad. The event featured 614 presentations, including 329 posters and 285 oral presentations across multiple disciplines. With support from the UK-Jiangsu World Class University Consortium, the symposium provided an international platform for PhD students to present their work, exchange ideas and receive feedback from academics, industry experts, and fellow researchers. Poster presentation at the 2026 Postgraduate Research Symposium The event concluded with an award ceremony on 20 March, which recognised outstanding presentations. According to XJTLU Graduate School, award-winning students from XJTLU will have the opportunity to represent the University at the second XJTU-XJTLU-UoL Doctoral Wisdom Convergence Camp in April. Sharing ideas across borders Raj Roy, a PhD student in Chemistry at New York University Shanghai, says he joined the symposium to exchange ideas with researchers working on similar problems. ‘I don’t believe research is a win-or-lose game. When a problem gets solved, whether by me or by someone else, it’s a win for humanity. It’s a small world in research, and it’s very likely that many people are working on similar problems. ‘Symposiums like this bring such people together, and gives an opportunity to exchange perspectives, discuss different approaches, learn from each other’s mistakes, and have meaningful debates that help move the field forward,’ he says. Raj Roy Roy, whose research explores ways to enhance the effectiveness of radiation therapy, says interacting with researchers from different cultural backgrounds is one of the most valuable aspects of academic events. ‘I believe cultural differences can also have an impact on research, because people from different cultures often have their thinking shaped by different experiences. As a result, they may approach the same research problem in unique ways. Seeing these different perspectives is valuable because it helps us think about problems in ways we might not have considered before,’ he says. ‘AI for Research’ A highlight of the symposium is the themed exhibition, ‘AI for Research’, which took place on the afternoon of 20 March. The exhibition showcased how PhD students integrate artificial intelligence into their research across diverse fields. Lok Hang Cheung, a PhD student at XJTLU’s Design School, who presented at the exhibition, focuses on the application of AI in architectural design. In his view, the emergence of AI has allowed many researchers without a computer science background to cross traditional boundaries. Lok Hang Cheung ‘AI for Research’ exhibition ‘In the past, I mainly explored – on a theoretical level – how designers interact with computer tools. But now, I can spend just a few days building a small application to test whether my ideas actually work. My research now has become inseparable from AI,’ he says. AI is also opening up new possibilities in life sciences, says Yu Lu, a PhD student in biological sciences at China Pharmaceutical University. She studies the molecular mechanisms involved in the initiation and progression of breast cancer. ‘Life science and biomedical research involve enormous amounts of data, from tumour gene sequencing to protein interaction analysis and drug molecule screening,’ she says. ‘Traditional research methods can be time-consuming and inefficient. AI, with its ability to process large datasets and make accurate model predictions, can significantly shorten research cycles and reduce research costs.’ Yu Lu While many see AI as a powerful tool, others also emphasise the importance of reflecting critically on its use. Vladimir Milić, from Serbia, is a PhD student at XJTLU’s School of Humanities and Social Sciences, studying China’s engagement with Central and