Latest news with #LargeLanguageModels
The Guardian
4 days ago
- Entertainment
- The Guardian
As we race headlong into our glorious AI-powered future, are we on the road to Idiocracy?
As we race headlong into our glorious AI-powered future, a long-forgotten flick from the naughties is gaining cult status by posing a simple question: are we on the road to Idiocracy? In the tradition of Bill and Ted, Idiocracy, a 2006 American sci-fi comedy, depicts an average guy transported 500 years into the future to find he is the smartest man on earth. A wrestler is president, people worship 'the profit' and the world's biggest corporation is a sports drink company whose market strategy has taken humanity to the brink of collapse. Idiocracy is not a flawless film; it's jarring class eugenics and casual sexism seem more dated than its 20 years would suggest. But what it does brilliantly is challenge the assumption that the human race is on an uncontested journey to higher consciousness. This is a timely counterpoint to the zeitgeist about the self-evident benefits of machine learning built by large tech companies under the conceit of 'artificial intelligence'. Already we are seeing academic studies suggesting that Large Language Models are linked to cognitive decline, with a recent MIT study finding lower brain engagement among students using GPTs in writing essays. We've also seen AI chatbots accused of inducing vulnerable people to suicide and cheery predictions about the hollowing out of entry level knowledge economy jobs which loom as our next inter-generational betrayal. The question is whether these dumb outcomes are features or bugs of our emerging information ecosystem. Evidence is mounting it is the former. Exhibit A is OpenAI's Economic Blueprint for Australia, an embarrassing document seeking rapid government adaptation, investment and minimalist regulation for 'the most significant economic and strategic opportunity of our time' You could drive a truck through the holes in OpenAI's analysis: $115bn in predicted productivity improvements based on crude calculations of hours saved at work, with no trade-off for the costs of the jobs already being destroyed. To realise this promised dividend OpenAI says the government would need to lean into resource-sucking datacentres, socialising the costs of energy and water, steamrolling communities and putting new pressures on the grid and fast-tracking development at the cost of the energy transition. As our creativity is being stolen and repurposed, trading off our collective empathy for automated culture, OpenAI wants us to 'streamline and update copyright law', with industry groups already pushing hard for a general right to mine our data. But these self-serving policy asks are not the worst of the OpenAI blueprint; it is the very fact that this massive corporation purports to set our future agenda at all, defining rather than responding to our collective needs. This design principle could represent the inflection point between a smart future and an impending Idiocracy. It is true this technology carries amazing power to synthesise information and challenge higher order thinking in new and profound ways. OpenAI is right to describe the technology as 'like electricity' something that can illuminate the night sky. But would you get a power company to set the rules for electrical safety? The truth is OpenAI is nether open or intelligent: it's a play to dominate a new technology on commercial terms for its material benefit, using their copious venture capital as a shield against competitors and a sword against government to create a policy environment to suit them. We are witnessing the next phase of their corporate history. ChatGPT is a compelling shop front, but what if it is intellectual heroin? It tricks us to feel smarter, more seen and even loved, while actually providing the opposite by convincing us to commoditise our collective intelligence. Short of returning to a genuine not-for-profit mission, OpenAI can never be a good faith partner. Theirs is an operating model to be resisted but that relies on us having the time, the understanding and, yes, the leadership to do this intelligently. In the latest episode of my podcast Burning Platforms one of the godfathers of artificial intelligence, Prof Toby Walsh, differentiates between the richness of distributed intelligence and the homogeneity of the concentrated intelligence that chatbots serve up. High quality data is earned not stolen property; it is used mindfully to address problems humans identify, not commoditised to fill some market niche. While the tech broligarchs battle for world domination, maybe the smart money should be on the design and value of small data models, designed for purpose not for producing mainstream slop and brain rot, that chew up less energy and eradicate fewer jobs. Because of the power of the tech sector, all of whom have well-paid and well-positioned Canberra lobbyists, we also need to resist. Since shouting out the Luddites in my last column I've been delighted to discover there are people already doing this. Ben Zhao, a University of Chicago computer scientist has developed programs like Glaze which protects private photos being harvested to train facial recognition technology and Nightshade, a filter for artists that tricks AI into seeing a cat as a dog, like putting ink in a bag of stolen bank cheques. And Cloudfare, one of the dominant cybersecurity companies, has announced it will ban AI web crawlers from scraping content from their sites without paying compensation to the owner of those sites. As Cloudfare CEO, Matthew Prince, says: 'I go to war every single day with the Chinese government, the Russian government, the Iranians, the North Koreans, probably Americans, the Israelis, all of them who are trying to hack into our customer sites. And you're telling me, I can't stop some nerd with a C-corporation in Palo Alto?' In the battle for our future intelligence, we need to deploy all the grey matter at our disposal: workers' intelligence, cultural intelligence, collective intelligence and the power of technologists in the face of the artifice. Spoiler: in the movie the sports drink company, 'Brawndo' extends its market dominance in electrolytes by expanding into agriculture, poisoning the land in pursuit of 'The Profit'. Open AI's Blueprint for Australia would be a similar triumph of Idiocracy. Peter Lewis is the executive director of Essential, a progressive strategic communications and research company that undertook research for Labor in the 2025 election and conducts qualitative research for Guardian Australia. He is also the host of Per Capita's Burning Platforms podcast
Bloomberg
5 days ago
- Business
- Bloomberg
Multiverse Computing Plans to Transform the AI Inference Market
Spanish AI startup Multiverse Computing says it has managed to compress Large Language Models (LLMs) by 95% and without sacrificing performance. So far, Multiverse Computing has $215 million to scale its quantum-inspired AI model compression tool. CEO and co-founder Enrique Lizaso spoke to Bloomberg's Tom Mackenzie about his company's plans to improve energy efficiency and compete with larger AI providers. (Source: Bloomberg)
Business Wire
5 days ago
- Business
- Business Wire
CyberCube and Munich Re: Joint experts publish report to advance the insurance industry's understanding of systemic cyber risks
LONDON--(BUSINESS WIRE)--CyberCube and Munich Re, both leading providers in their field of cyber risk, analytics and insurance, have published the main findings of a joint study on severe cyber accumulation events and the relative resiliency of organizations to systemic events due to effective mitigation measures. The survey gathered insights from 93 seasoned cybersecurity professionals. The results provide a nuanced view of how systemic cyber events might unfold and of the factors that drive wide variation in risk exposure across firms: Widespread Malware Risk According to the majority of responding experts, a severe malware event could infect a quarter of all systems worldwide, but they agreed in that case only 15% may be fully compromised. Experts do not see an event where more than 50% of the world's systems are completely compromised. Based on the experts' judgement, another event on the scale of WannaCry and NotPetya would not be seen as surprising. Patch management, network segmentation, and data backups are identified as the most effective mitigations that organizations have against widespread malware attacks. When done effectively, such mitigations can reduce the chance of being affected by a widespread malware attack by 50% to 80% and reduce the financial impacts from such an event by a similar amount. Cloud Risk Cybersecurity experts expect broad cloud outages to last hours to days; outages beyond 72 hours are considered unlikely but not impossible. Findings show at least a medium level of dependency on cloud services across most industries with companies' business-critical operations increasingly reliant on them. Reliance tends to decrease with increasing company size. Financial losses scale with cloud outage duration: Respondents reported that a single-day outage of their most critical Cloud Service Provider (CSP) would likely result in a financial loss equal to 1% of their yearly revenue. Variation in losses reflect differences in dependency on the cloud, based on an organization's size, sector, and contingency planning. The most effective mitigation against cloud outages is to establish a multi-region architecture with the CSPs used for critical business applications. Having multiple CSPs was not found to be effective; the option to transfer service from one CSP to another during an outage was seen as unfeasible. Cyber Experts surveyed rate Azure, AWS and Google as the best prepared to mitigate against a major cloud outage and to recover from such an event. Emerging and Systemic Risks Experts believe that new technologies will begin to affect the threat landscape at about the same pace that they are being adopted in cybersecurity practices. According to cybersecurity experts, in the near term Industrial and Consumer Internet of Things (IoT) devices pose the biggest concern. Large Language Models (LLMs) are regarded as having an impact now while Artificial General Intelligence (AGI) is seen as a greater concern in five or more years. A fundamental challenge in cyber risk modeling is the deficiency of concrete tail-risk events, such as systemic malware or multi-region cloud outages. The joint survey represents the best attempt to parameterize plausible worst-case scenarios and establish expert consensus. Its objective was to advance market understanding, particularly concerning risk mitigation strategies for systemic cyber events. The results add credibility to CyberCube's model forecasts and further improve Munich Re's internal model and accumulation risk understanding. Jon Laux, Vice President of Analytics at CyberCube, said: "By sharing the findings of our study on systemic cyber risks, we aim to provide a more nuanced view of how systemic cyber events might unfold and the factors that drive wide variation in risk exposure across firms.' Stephan Brunner, Senior Cyber Actuary at Munich Re, said: 'Our ambition is to improve the understanding of possible extreme malware and cloud events alongside the effectiveness of mitigation measures by sharing the insights of our study. In collaboration, Munich Re aims to further strengthen expertise on systemic cyber risks and advance cyber accumulation modeling." The research has contributed to a more refined understanding of the relative resiliency of organizations to systemic events and the key variables that influence an organization's ability to withstand such incidents. These findings represent an important input into CyberCube's and Munich Re's evolving view of cyber risk and help inform ongoing enhancements to their modeling approach. CyberCube has incorporated these insights into Version 6 of its risk aggregation platform, Portfolio Manager. Modeling cyber accumulation is a joint effort across the entire insurance industry. For this reason, the key findings of the survey are being published to foster dialogue in the market. This study is the third of its kind, CyberCube and Munich Re plan to conduct another study in 2026. Interested cybersecurity experts are invited to participate. Read the report summarising the full study here – Key insights into systemic cyber risk CyberCube is the leading provider of software-as-a-service cyber risk analytics to quantify cyber risk in financial terms. Driven by data and informed by insight, we have harnessed the power of artificial intelligence to supplement our multi-disciplinary team. Our clients rely on our solutions to make informed decisions about managing and transferring cyber risks. We unpack complex cyber threats into clear, actionable strategies, translating cyber risk into financial impact on businesses, markets, and society as a whole. T he CyberCube platform was established in 2015 within Symantec and now operates as a standalone company. Our models are built on an unparalleled ecosystem of data and validated by extensive model calibration, internally and externally. CyberCube is the leader in cyber risk quantification for the insurance industry, serving over 100 insurance institutions globally. The company's investors include Forgepoint Capital, HSCM Bermuda and Morgan Stanley Tactical Value. For more information, please visit or email info@ Munich Re is one of the world's leading providers of reinsurance, primary insurance and insurance-related risk solutions. Munich Re is globally active and operates in all lines of the insurance business. Since it was founded in 1880, Munich Re has been known for its unrivalled risk-related expertise and its sound financial position. Munich Re leverages its strengths to promote its clients' business interests and technological progress. Moreover, Munich Re develops covers for new risks such as rocket launches, renewable energies, cyber risks and artificial intelligence. In the 2024 financial year, Munich Re generated insurance revenue of €60.8bn and a net result of €5.7bn. The Munich Re Group employed about 44,000 people worldwide as at 31 December 2024. For more information, please visit
India Today
6 days ago
- Business
- India Today
IIT Delhi launches six-month executive course in generative AI for professionals
To cater to the surging demand for skilled professionals in the evolving field of artificial intelligence, the Indian Institute of Technology Delhi (IIT Delhi) has rolled out the second edition of its Certificate Programme in Generative AI. Offered under its Continuing Education Programme (CEP), the six-month online course is crafted for working professionals seeking cutting-edge expertise in Large Language Models (LLMs), Natural Language Processing (NLP), and ethical AI sector-agnostic programme is tailored for individuals across domains including software engineering, data science, machine learning, digital product management, and applied research. It also welcomes educators and tech-savvy professionals aspiring to pivot into advanced AI are trained in key AI tools such as Python, NumPy, TensorFlow, PyTorch, spaCy, and Hugging Face. The curriculum includes hands-on tutorials and industry-inspired capstone projects designed to simulate real-world AI applications across verticals like healthcare, education, finance, and autonomous systems. The comprehensive coursework covers advanced topics such as neural network development, transformer-based architectures, multilingual NLP, and parameter-efficient fine-tuning (PEFT) for low-resource settings. Participants also delve into modern model architectures like GPT, BERT, and T5, while exploring frontier techniques such as instruction tuning, retrieval-augmented generation (RAG), reinforcement learning with human feedback (RLHF), and advanced prompting strategies to boost model performance and the broader impact of the programme, Professor Tanmoy Chakraborty from the Department of Electrical Engineering at IIT Delhi shared, 'This programme is rooted in our belief that Generative AI will be at the heart of future innovation and decision-making. We're committed to nurturing professionals who can not only understand AI technologies but also lead their application across industries with accountability and depth.'As AI continues to reshape the global landscape, industry reports have underscored the critical need for AI talent. A PwC analysis projects AI to add up to USD 15.7 trillion to the global economy by 2030. However, a BCG study points out that while AI investments are rising, only 26% of organisations are successfully scaling these technologies for real value. Meanwhile, EY's The AIDEA of India report suggests that Generative AI alone could contribute USD 1.5 trillion to India's GDP by the end of the programme is delivered through a blend of live online classes and self-paced learning, offering 60 hours of instructor-led sessions, structured tutorials, and a 10-hour capstone project. Participants also have the opportunity to engage in an optional one-day campus immersion at IIT Delhi, gaining exposure to the institute's academic and research enrol, candidates must hold an undergraduate or postgraduate degree in science, technology, engineering, or mathematics. Upon successful completion, learners will receive an e-certificate from IIT Delhi a focus on practical skill-building and responsible innovation, this programme aims to shape the next generation of AI leaders equipped to drive meaningful transformation across industries.- EndsMust Watch
Yahoo
11-07-2025
- Business
- Yahoo
Best Agentic AI Course for Software Developers and Engineers 2025 - Interview Kickstart Launches Advanced GenAI Course with AI Projects
Santa Clara, July 11, 2025 (GLOBE NEWSWIRE) -- Agentic AI systems are revolutionizing how organizations approach complex workflows, introducing autonomous agents capable of multi-step reasoning, decision-making, and task execution that operate independently while coordinating with other systems and human operators. This transformative shift is comprehensively addressed in Interview Kickstart's Advanced GenAI Course, an intensive 8-9 week program covering the foundational technologies driving agentic AI, including deep learning, large language models, diffusion models, multimodal systems, and reinforcement learning, all essential for developing sophisticated autonomous agents. To learn more about the course, visit: Generative AI (GenAI) lies at the core of agentic systems, enabling machines to generate new content, such as text, images, and code, and operate autonomously in dynamic environments. GenAI courses like Interview Kickstart's Advanced GenAI are designed to equip professionals with the ability to work with generative models such as GANs (Generative Adversarial Networks), VAEs (Variational Autoencoders), and Large Language Models (LLMs). Foundational AI concepts, machine learning techniques, and real-world applications are explored in-depth, alongside critical tools like PyTorch and Hugging Face that support hands-on model development and fine-tuning. The curriculum delves into key skills like prompt engineering, parameter-efficient fine-tuning, and ethical considerations surrounding GenAI, including bias and misinformation. Learners also gain exposure to frameworks such as Langchain for orchestrating complex agent behaviors and the Alpaca model for instruction-following, while also studying diffusion models like DDIM, DDPM, and Stable Diffusion for creative and analytical tasks. Reinforcement learning modules explain how agentic systems learn from feedback and adjust their strategies, an essential capability for maintaining performance in real-world workflows. The emergence of agentic AI represents a fundamental evolution beyond traditional automation, enabling systems that can understand context, adapt to changing conditions, and execute complex multi-step processes with minimal human intervention. These autonomous agents leverage advanced reasoning capabilities to interpret instructions, plan execution strategies, handle exceptions, and coordinate across multiple tools and systems, creating unprecedented opportunities for workflow optimization across industries. Organizations are rapidly deploying agentic AI systems for customer service orchestration, data pipeline management, content creation workflows, and software development processes. These systems demonstrate sophisticated capabilities, including natural language understanding for interpreting user requirements, strategic planning for complex task decomposition, and adaptive execution that responds to changing conditions or unexpected obstacles, requiring deep technical expertise in the underlying generative AI technologies. Interview Kickstart's Advanced GenAI Course provides comprehensive foundations in the technologies powering agentic AI systems. Participants develop expertise with large language models that enable natural language reasoning, diffusion models that support creative and analytical tasks, and multimodal systems that integrate diverse data types, all critical components of effective autonomous agents. The curriculum explores advanced frameworks, including Langchain, which enables sophisticated agent orchestration and tool integration, alongside specialized models like the Alpaca model for instruction-following capabilities. Participants also study Denoising Diffusion Implicit Models (DDIMs) and various diffusion approaches, including DDPM and Stable Diffusion, which provide creative and analytical capabilities essential for autonomous workflow execution. Reinforcement learning components of the course address how agentic systems learn from interactions and improve performance over time, a critical capability for autonomous agents operating in dynamic environments. This foundation enables participants to understand how agents can adapt their strategies based on feedback and changing conditions. The program's capstone project provides hands-on experience creating LLM-based applications, with many participants developing autonomous agent systems that demonstrate sophisticated workflow orchestration capabilities. These projects often showcase agents capable of multi-step reasoning, tool integration, and adaptive execution, which are practical demonstrations of agentic AI principles in action. This course caters to a range of skill levels and includes content tailored to specific goals such as application building, tool mastery, or career transition. As such, professionals seeking high-quality, immersive instruction often look to platforms like Interview Kickstart for structured learning, personalized mentoring, and results-driven outcomes. Throughout the intensive program, instructors conduct personalized 1:1 sessions with learners to help them navigate the complex technical challenges of developing agentic systems. These individualized consultations address specific implementation questions while guiding how to position expertise in autonomous AI development for career advancement opportunities. The rise of agentic AI is creating new categories of technical roles focused on designing, implementing, and managing autonomous workflow systems. Interview Kickstart's Advanced GenAI Course provides essential preparation for professionals seeking to position themselves at the forefront of this technological transformation. To learn more visit About Interview Kickstart Interview Kickstart, founded in 2014, is a trusted upskilling platform designed to help tech professionals secure roles at FAANG and other leading tech companies. With over 20,000 success stories, it has become a go-to resource for career advancement in the tech industry. The platform offers a flexible learning experience with live classes and over 100,000 hours of on-demand video lessons. This ensures learners have the tools they need to dive deep into technical concepts and refine their skills on their own schedule. Additionally, 1:1 coaching sessions provide personalized support in areas like resume building and LinkedIn optimization, enhancing each learner's professional profile. ### For more information about Interview Kickstart, contact the company here:Interview KickstartBurhanuddin Pithawala+1 (209) 899-1463aiml@ Patrick Henry Dr Bldg 25, Santa Clara, CA 95054, United States CONTACT: Burhanuddin PithawalaSign in to access your portfolio
