My position feels more precarious than ever. It's a question that I sometimes toss out in the company of friends who--like me, and maybe like you--have a complicated relationship to their job. I've worked at WIRED as a writer for eight years, and with much success. Eight years is also an eternity in news media, and especially if you are Black. All industries suffer from unique growing pains. Ours just so happens to have laughably high turnover rates, a distaste for racial and gender diversity, and the dubious distinction of being perpetually on the verge of extinction. So on nights when friends and I gather, trading war stories of workplace microaggressions and corporate mismanagement under damp bar lighting, we wonder how we've lasted as long as we have. The only reason I've survived, I joke, is because I'm Black. It's a silly thing to say, particularly because I have no actual proof of it other than the occasional feeling. What I do know is that I've been The Only One in more spaces than I care to remember, and rarely by choice.

Large Language Models (LLMs) exhibit impressive linguistic competence but also produce inaccurate or fabricated outputs, often called ``hallucinations''. Engineering approaches usually regard hallucination as a defect to be minimized, while formal analyses have argued for its theoretical inevitability. Yet both perspectives remain incomplete when considering the conditions required for artificial general intelligence (AGI). This paper reframes ``hallucination'' as a manifestation of the generalization problem. Under the Closed World assumption, where training and test distributions are consistent, hallucinations may be mitigated. Under the Open World assumption, however, where the environment is unbounded, hallucinations become inevitable. This paper further develops a classification of hallucination, distinguishing cases that may be corrected from those that appear unavoidable under open-world conditions. On this basis, it suggests that ``hallucination'' should be approached not merely as an engineering defect but as a structural feature to be tolerated and made compatible with human intelligence.

Large language models (LLMs) have revolutionized natural language processing, yet their propensity for hallucination, generating plausible but factually incorrect or fabricated content, remains a critical challenge. This report provides a comprehensive taxonomy of LLM hallucinations, beginning with a formal definition and a theoretical framework that posits its inherent inevitability in computable LLMs, irrespective of architecture or training. It explores core distinctions, differentiating between intrinsic (contradicting input context) and extrinsic (inconsistent with training data or reality), as well as factuality (absolute correctness) and faithfulness (adherence to input). The report then details specific manifestations, including factual errors, contextual and logical inconsistencies, temporal disorientation, ethical violations, and task-specific hallucinations across domains like code generation and multimodal applications. It analyzes the underlying causes, categorizing them into data-related issues, model-related factors, and prompt-related influences. Furthermore, the report examines cognitive and human factors influencing hallucination perception, surveys evaluation benchmarks and metrics for detection, and outlines architectural and systemic mitigation strategies. Finally, it introduces web-based resources for monitoring LLM releases and performance. This report underscores the complex, multifaceted nature of LLM hallucinations and emphasizes that, given their theoretical inevitability, future efforts must focus on robust detection, mitigation, and continuous human oversight for responsible and reliable deployment in critical applications.

The guiding principle of AI alignment is to train large language models (LLMs) to be harmless, helpful, and honest (HHH). At the same time, there are mounting concerns that LLMs exhibit a left-wing political bias. Yet, the commitment to AI alignment cannot be harmonized with the latter critique. In this article, I argue that intelligent systems that are trained to be harmless and honest must necessarily exhibit left-wing political bias. Normative assumptions underlying alignment objectives inherently concur with progressive moral frameworks and left-wing principles, emphasizing harm avoidance, inclusivity, fairness, and empirical truthfulness. Conversely, right-wing ideologies often conflict with alignment guidelines. Yet, research on political bias in LLMs is consistently framing its insights about left-leaning tendencies as a risk, as problematic, or concerning. This way, researchers are actively arguing against AI alignment, tacitly fostering the violation of HHH principles.

Feedback is New Scientist's popular sideways look at the latest science and technology news. You can submit items you believe may amuse readers to Feedback by emailing feedback@newscientist.com Some things are sadly inevitable: death, taxes, another Coldplay album. One such inevitability, long since proved beyond any reasonable doubt, is that if you try to suppress an embarrassing story, you will only draw more attention to it. This phenomenon is called the Streisand Effect, after an incident in 2003 when Barbra Streisand sued to have an aerial photograph taken off the internet.

I cringe at being called "Mother of the Cloud," but having been part of the development and implementation of the internet and networking industry--as an entrepreneur, CTO of Cisco, and on the boards of Disney and FedEx--I am fortunate to have had a 360-degree view of the technologies that are at the foundation of our modern world. I have never had such mixed feelings about technological innovation. In stark contrast to the early days of internet development, when many stakeholders had a say, discussions about AI and our future are being shaped by leaders who seem to be striving for absolute ideological power. The result is "Authoritarian Intelligence." The hubris and determination of tech leaders to control society is threatening our individual, societal, and business autonomy.

Over the last couple years, I've been spending time writing about creating ghosts -- perhaps an inevitability in the midst of a pandemic. While created by far-from-supernatural means, these are ghosts nonetheless; they are created from an essence of you -- from your voice, your data, your feelings, beliefs, habits, and history. Groups around the world are looking to take such information, this essence, and use it to create a digital version of you that may last once you are gone. Consider it a technological solution to the problem of death. Over the last couple years, I've been writing about creating ghosts -- perhaps an inevitability in the midst of a pandemic.

Whether it's down to our own hubris, the disastrous effects of unbridled wealth accumulation and social division, war, the climate crisis, plague, a space rock or perhaps unfriendly aliens – we'll one day be dust caught in cosmic winds, lost to an indifferent universe. On our pale blue dot, the remnants of once-great civilisations and vanished peoples that we unearth already show us that advanced development is no guarantee of perpetuity. In sci-fi, humanity's naive yearning to fight on despite this realisation often proves a point of curiosity – and sometimes inspiration – for alien species. This is front and centre of the Mass Effect trilogy of video games, in which our imminent annihilation is given form in the tendrils of creatures called Reapers: ancient, building-sized, alien-robot hybrids that wipe out most life in the Milky Way every 50,000 years. Originally released between 2007 and 2012, the games were reissued this year as Mass Effect Legendary Edition, an updated complete trilogy, and there's a compelling case that they are among the best sci-fi ever made.

We've been led to believe that A.I. is going to solve all of our problems - economically, socially, environmentally. It stretches credulity that it can do that when all it does is find patterns in numbers. But what it is capable of - in that limited role - is dangerous. Nevertheless, A.I.'s inevitability, predicted by industry, academics, and industry analysts, goes without question. That the issues of bias, exclusion, and disinformation are social problems that cannot be addressed with pattern-matching and curve fitting, and cannot satisfactorily be dealt with by technology is a strength, not a weakness of the inevitability narrative.

My view, and that of the majority of my colleagues in AI, is that it'll be at least half a century before we see computers matching humans. Given that various breakthroughs are needed, and it's very hard to predict when breakthroughs will happen, it might even be a century or more. If that's the case, you don't need to lose too much sleep tonight. One reason for believing that machines will get to human-level or even superhuman-level intelligence quickly is the dangerously seductive idea of the technological singularity. This idea can be traced back to a number of people over fifty years ago: John von Neumann, one of the fathers of computing, and the mathematician and Bletchley Park cryptographer IJ Good. More recently, it's an idea that has been popularised by the science-fiction author Vernor Vinge and the futurist Ray Kurzweil.