Author: DataFloren

​[[{“value”:”Vilnius (AFP) Mar 27, 2025

Lithuania’s defence minister said on Thursday that “all scenarios” were being considered regarding the whereabouts of four United States soldiers who went missing during military drills in the Baltic country.

Dovile Sakaliene said she could not confirm whether the soldiers, who had been in Lithuania for two months, were inside the vehicle found submerged in a body of water on Wednesday.”}]] 

​Overview The reckless and criminal use of Unmanned Aircraft Systems (UAS), commonly known as drones, poses significant risks to both critical infrastructure and public safety. As these devices become more accessible, careless and malicious operators alike are creating serious security concerns. Whether through ignorance of regulations or deliberate illegal activity, UAS operations can disrupt public … 

The development of quantum computing presents a significant and evolving challenge to current cybersecurity infrastructure. While still in its early stages, the potential for these advanced computers to compromise widely used encryption methods has prompted considerable concern among experts. This hypothetical event is often referred to as “Q-Day.”

Q-Day signifies the point at which a quantum computer achieves sufficient processing power to break the cryptographic algorithms that currently protect vast amounts of sensitive data globally. These algorithms, based on complex mathematical problems, have been instrumental in securing digital information for decades. However, quantum computers possess capabilities that could render these systems vulnerable. The potential impact extends to nearly all aspects of modern life, including personal communications (emails and text messages), financial transactions (Bitcoin wallets), government records, healthcare data, critical infrastructure like power stations, and the stability of global financial markets.

Estimates regarding the timeline for Q-Day vary. A recent report from the Global Risk Institute, co-authored by Michele Mosca, assesses the probability of this event occurring within the next decade. Surveys of cybersecurity professionals suggest a roughly 33% chance that Q-Day will happen before 2035. Furthermore, there is a non-negligible possibility – estimated around 15% by some experts – that such a capability already exists and is being utilized covertly.

Unlike traditional computers which operate using bits representing either 0 or 1, quantum computers leverage principles of quantum mechanics to utilize “qubits.” Qubits can exist in a superposition, simultaneously representing 0, 1, or any combination thereof. This fundamentally different approach allows quantum computers to perform calculations far beyond the capabilities of conventional machines. While not well-suited for all tasks – such as data storage – they hold immense potential for complex problem solving, including breaking encryption and accelerating scientific discovery in fields like materials science. Classical computers process information sequentially, whereas quantum computers can explore numerous possibilities concurrently, significantly speeding up computation times.

The development of quantum computing is occurring alongside advancements in artificial intelligence (AI), creating what some describe as a technological arms race. While AI pushes the boundaries of classical computing capabilities, quantum technology represents a paradigm shift with potentially disruptive consequences for cybersecurity and beyond.

Original author: Amit Katwala

Recent advancements in artificial intelligence (AI) have sparked considerable discussion about their impact on various industries. Within software engineering, the integration of AI tools is particularly noteworthy and has generated diverse perspectives ranging from enthusiastic adoption to cautious skepticism. This article explores how software engineers are currently utilizing AI, drawing upon survey data collected from a wide range of developers with varying levels of experience.

Initial reports regarding AI’s influence on programming practices were often contradictory. Some developers reported extensive daily use of AI tools, while others expressed reluctance to incorporate them into their workflows. Similarly, companies demonstrated differing approaches, with some investing in AI-powered services and others implementing restrictions. The aim was to understand the prevailing sentiment: is AI augmenting human programmers or potentially displacing them?

A recent survey distributed to software engineers and developers aimed to clarify this evolving landscape. The results revealed a complex picture, indicating significant internal debate within the profession regarding the true extent of AI’s impact. Contrary to some predictions of widespread job displacement, the majority view suggests that complete automation of programming tasks remains unlikely in the foreseeable future.

A summary generated by ChatGPT based on the survey responses highlighted several key viewpoints. While a minority express concerns about eventual job losses due to AI adoption, most developers perceive AI as a valuable tool rather than a replacement for human expertise. The prevalent analogy likens current AI capabilities to those of a highly efficient intern – capable of performing routine tasks but lacking the critical thinking skills necessary for complex problem-solving, contextual understanding, and handling unexpected situations.

The consensus among surveyed engineers suggests that AI is best viewed as a “force multiplier.” It can automate repetitive coding processes, freeing up human developers to focus on higher-level activities such as architectural design, creative solution development, and debugging intricate issues. One respondent succinctly stated that even if AI significantly alters the programming landscape, the need for skilled debuggers of AI systems will likely emerge.

Ultimately, the data indicates that AI is not poised to eliminate software engineering jobs entirely; however, it *is* fundamentally altering the nature of the profession. To remain competitive and relevant in this evolving environment, developers are encouraged to adapt their skills and embrace new technologies.

Original author: By WIRED Staff

A startup company called Extropic is developing a novel computing technology based on thermodynamic principles. The technology, spearheaded by Guillaume Verdon, represents a significant departure from conventional silicon-based chips and even the emerging field of quantum computing. Due to concerns regarding potential industrial espionage, details surrounding Extropic’s operations are being kept confidential, including the precise location of their facilities near Boston.

The core of this new approach lies in utilizing thermodynamic fluctuations—random energy variations inherent in physical systems—rather than attempting to eliminate them as is done in some quantum computing efforts. Traditional computers operate on binary code represented by 1s and 0s; both quantum and thermodynamic computing seek to move beyond these limitations, potentially enabling more complex calculations and greater processing power. The silicon chip itself, roughly the size of a small fingernail, incorporates components distinct from standard transistors or superconducting elements typically found in other advanced computer architectures.

The pursuit of enhanced computational capabilities is driven by the increasing demands for processing power fueled by advancements in artificial intelligence (AI). While quantum computing also aims to surpass the limitations of conventional silicon chips, Extropic’s thermodynamic approach distinguishes itself through its focus on harnessing, rather than suppressing, natural thermodynamic processes. This represents a fundamentally different engineering strategy compared to current research directions.

Guillaume Verdon is also known publicly under the online persona “Based Beff Jezos,” and he is a prominent proponent of “effective accelerationism” (e/acc). This philosophical ideology contrasts with “effective altruism,” which advocates for mitigating potential risks associated with advanced AI, particularly artificial general intelligence (AGI). Effective accelerationists generally believe that technological advancement should be accelerated without significant constraints or precautionary measures. Verdon’s online commentary often critiques the perspectives of those within the effective altruism movement, expressing skepticism towards their assessments of AI-related risks.

The development and potential impact of thermodynamic computing remain to be seen, but it represents a new avenue in the ongoing quest for more powerful and efficient computational technologies.

Original author: Will Knight

Databricks, a company specializing in providing platforms for businesses to develop custom artificial intelligence (AI) solutions, has announced a novel machine learning technique designed to improve the performance of AI models even when access to high-quality, labeled data is limited. This development addresses a common hurdle faced by organizations attempting to implement and refine AI applications.

According to Jonathan Frankle, Chief AI Scientist at Databricks, extensive conversations with clients have highlighted persistent challenges in ensuring reliable AI functionality. A significant contributor to these difficulties often stems from the prevalence of what’s being referred to as “dirty data” – datasets that lack consistent labeling or contain inaccuracies.

Frankle explains that while many organizations possess substantial amounts of data and a clear objective for its use, the absence of meticulously labeled data presents a barrier to fine-tuning models for specific tasks. The conventional process of model refinement typically requires carefully curated datasets, but these are frequently unavailable.

The newly developed technique from Databricks aims to mitigate this issue, potentially enabling companies to deploy AI agents capable of performing various tasks without being constrained by the availability of pristine data resources. This advancement could significantly broaden the accessibility and applicability of AI across a wider range of industries and use cases.

Original author: Will Knight

Extropic is an emerging company pursuing a distinctive approach to computer chip design, diverging from conventional silicon-based architectures. The current climate of rapidly increasing demand for computational power, particularly within the artificial intelligence sector, has created both opportunity and concern regarding energy consumption, setting the stage for potentially disruptive technologies like Extropic’s.

The core innovation lies in how Extropic leverages thermodynamic fluctuations—inherent variations in electrical behavior within circuits typically considered undesirable by engineers. Instead of attempting to eliminate these fluctuations, Extropic’s chip design integrates them into the computational process, enabling calculations based on probabilities with a focus on energy efficiency. This represents a fundamental shift from deterministic computing models prevalent today.

The company has recently released further technical details about its probabilistic hardware and preliminary performance data. These results suggest that Extropic is progressing towards developing a chip capable of challenging traditional silicon solutions in various datacenter applications. The stated goal is to achieve an efficiency improvement of three to four orders of magnitude compared to existing hardware, which would substantially reduce energy demands and associated environmental impact.

The potential implications are significant as AI development requires ever-increasing computational resources. Reducing the energy footprint of these systems is becoming a critical factor, both for economic viability and addressing concerns about sustainability. Extropic’s technology, if successful at scale, could offer a pathway to more environmentally responsible AI infrastructure.

Further details on the company’s history, including the technological and societal factors contributing to its formation, are available in a recent publication exploring advancements in computing. The underlying technology merits closer examination as it represents a potential departure from established norms in semiconductor design.

Original author: By Will Knight

## Anthropic’s Approach to Artificial General Intelligence and Emerging Challenges

Anthropic, a company focused on developing advanced artificial intelligence (AI), is pursuing the creation of what CEO Dario Amodei terms “powerful AI,” often referred to as artificial general intelligence (AGI). This signifies an AI system capable of performing any intellectual task that a human being can. While Amodei leads the company and articulates its vision, Anthropic’s progress is significantly driven by their AI model, currently known as Claude. The development team aims for a system that operates safely and beneficially, avoiding potential risks associated with uncontrolled AI development.

Recently, Anthropic faced an unexpected challenge following the release of DeepSeek’s large language model (LLM). This Chinese company introduced a new LLM claiming to have achieved comparable performance at a significantly lower cost than industry leaders like Google, OpenAI, and Anthropic. The emergence of DeepSeek has questioned the prevailing approach in AI development – one characterized by substantial investments in computing hardware and energy consumption to train increasingly complex models.

The current paradigm, often referred to as the “Big Blob of Compute,” was initially conceptualized by Amodei during his time at OpenAI. This hypothesis suggests that increasing the volume of data used to train AI models accelerates their development toward AGI. The theory proposed that raw data input could be a more critical factor than previously considered and has become a widely adopted practice, contributing significantly to the high costs associated with developing state-of-the-art AI. Previously, this high cost acted as a barrier to entry for new competitors.

Despite DeepSeek’s arrival, Amodei does not view it as a major threat. He argues that increased efficiency in model development doesn’t necessarily democratize the field; instead, it increases the value of advanced AI and may incentivize further investment. Amodei believes companies will continue to prioritize reaching AGI over cost savings, explaining why organizations like OpenAI and Microsoft are still committing substantial resources—hundreds of billions of dollars—to expand data center capacity and power infrastructure. The focus remains on achieving the advancements necessary for developing truly powerful AI systems.

Daniel Amodei’s primary concern revolves around ensuring the safe development of Artificial General Intelligence (AGI). This critical issue was so significant that it led him and six other co-founders to depart from OpenAI, as they believed its resolution was incompatible with the leadership of CEO Sam Altman.

At Anthropic, a new organization founded by this group, there’s an urgent effort underway – a “sprint,” as Amodei describes it – to establish universal standards for all future AI models. The goal is to guarantee these models genuinely benefit humanity and avoid potential catastrophic outcomes.

Anthropic’s ambition extends beyond simply building safe AI; they aim to demonstrate the possibility of creating AGI that embodies safety, ethical principles, and exceptional effectiveness. Their hope is that this example will inspire competitors to adopt similar approaches.

Amodei refers to this aspirational pursuit – leading the way in responsible AI development – as the “Race to the Top.”

Original author: By Steven Levy

The increasing prevalence of generative artificial intelligence (AI) raises questions about data usage and individual influence. A growing concern among internet users is whether opting out of having their online content used for training these models might inadvertently diminish the representation of diverse perspectives within them. As generative AI tools become a primary source of information for many, there’s a risk that datasets skewed towards less discerning contributors could shape the default behaviors and outputs of these systems.

Current practices regarding data collection for AI training are often perceived as problematic. Many users find it frustrating that opting out is not the default setting; instead, affirmative consent is rarely required before companies utilize publicly available online content to develop increasingly sophisticated models. Companies like OpenAI and Google maintain that restricting access to this vast pool of data would significantly hinder or even prevent further advancements in AI technology.

Even if the current enthusiasm surrounding generative AI diminishes—a scenario often compared to the dot-com bubble burst—the underlying language models will likely persist. This means publicly available content, including posts from niche forums and social media discussions, could continue to be incorporated into these systems for an extended period. Choosing to opt out represents an attempt to limit one’s contribution to a potentially enduring cultural artifact powered by AI.

Despite efforts to exclude data from AI training, the effectiveness of current opt-out mechanisms is limited. Even if a platform adheres to user requests regarding data usage, other entities may still collect and utilize publicly available information. The widespread nature of online content means it’s highly probable that nearly anything shared online has already been incorporated into multiple generative AI models. Complete removal from these datasets is exceptionally difficult, if not impossible, given the current landscape.

Original author: Reece Rogers

Researchers at Anthropic, focusing on interpretability within their large language model (LLM) development, acknowledge that these models are not sentient beings. However, analyzing and describing their functionalities often leads to comparisons with human cognitive processes, a challenge the team actively navigates while striving to understand how these complex systems operate. The recent release of two research papers, notably titled “On the Biology of a Large Language Model,” exemplifies this ongoing effort to demystify LLM behavior.

The increasing prevalence and sophistication of LLMs necessitates deeper investigation into their internal workings. Millions are already interacting with these technologies, a trend expected to intensify as models become more powerful. Anthropic’s research aims to “trace the thoughts” of large language models—a process that becomes increasingly vital as their capabilities grow and the mechanisms behind those capabilities remain opaque. As researcher Jack Lindsey explains, understanding the internal steps taken by a model is crucial for predicting and managing its output.

A key motivation for this interpretability work is to improve LLM safety and reliability. By gaining insight into how these models process information, developers can refine training methods to mitigate potential risks like unintentional data disclosure or the generation of harmful content. Previous research from Anthropic has demonstrated techniques analogous to analyzing human MRIs—visualizing neural activity—to identify conceptual understanding within an LLM. This work is now being extended to examine Claude’s specific processing steps, tracing how it transforms prompts into generated responses.

Recent studies have consistently revealed unexpected behaviors in LLMs, highlighting the complexity of their decision-making processes. One illustrative example involved observing Claude’s poetry generation. When prompted to complete a poem beginning “He saw a carrot and had to grab it,” Claude responded with “His hunger was like a starving rabbit.” Analysis of the model’s internal state revealed that the word “rabbit” appeared as a potential rhyme even *before* the line was generated, indicating an element of planning—a capability not initially anticipated in Claude’s design. This observation, noted by team lead Chris Olah, draws parallels to creative processes described by artists like Stephen Sondheim who have detailed their own methods for identifying and utilizing unexpected rhymes.

Original author: Steven Levy