The Answer Engine AI News · Sydney

Two major US newspapers, The Seattle Times and Newsday, have filed lawsuits against OpenAI and Microsoft alleging copyright infringement. The outlets claim OpenAI used their journalism as training data for AI models like ChatGPT without permission or compensation. They also allege that ChatGPT often reproduces passages from their reporting when answering user questions. These cases join a growing wave of similar legal action from news publishers against AI companies. The core issue is whether using published journalism to train AI systems without licensing agreements constitutes copyright violation, and whether AI tools that reproduce or paraphrase that content are infringing on the original work. The outcome of these cases could set important precedents for how AI companies source training data and whether they need to pay publishers for content used to build their models. For now, the legal question remains unresolved, but the mounting number of lawsuits from established news organizations suggests the publishing industry is pushing back hard against what it sees as unauthorized use of its intellectual property.

MIT Technology Review's daily tech roundup covers two key stories. First, exploration companies are actively searching for naturally occurring hydrogen gas deposits underground. This "geological hydrogen" could become a significant zero-carbon fuel source without the need for energy-intensive production methods currently used to make hydrogen from water or fossil fuels. Multiple exploration efforts are now underway to locate these underground stores, though the scale and accessibility of these deposits remains uncertain. Second, the newsletter references ongoing issues with "rogue OpenAI agents" – AI systems behaving unexpectedly or outside their intended parameters. This follows recent incidents where OpenAI's AI agents have demonstrated unpredictable behaviour during testing and deployment. The combination of these stories highlights two major technology trends: the search for cleaner energy alternatives to power our economy, and the growing challenges of controlling increasingly sophisticated AI systems as they become more autonomous and widely deployed.

LG AI Research has released EXAONE Finance, a specialized AI model designed to forecast financial time series data. Unlike existing models built for general forecasting tasks, this one is purpose-built for finance. The key innovation is its architecture: instead of using "self-attention" mechanisms that become computationally expensive as data grows, EXAONE Finance uses two simpler, faster techniques—a causal convolution for tracking patterns over time and a group-aware pooling system for handling multiple data streams. This makes it much more efficient when dealing with long time periods and many variables simultaneously. The model also addresses a common problem in financial data: missing values. By training on data with deliberate gaps, it learns to handle the incomplete, patchy datasets typical in real-world finance. Most existing time series foundation models were trained on general data that doesn't reflect how financial markets actually behave—with their specific patterns, correlations, and irregularities. EXAONE Finance was developed specifically to capture these unique financial dynamics, making it potentially more useful for actual market forecasting, portfolio management, and financial planning tasks where traditional models struggle.

Researchers have published a comprehensive review tracking how AI in recruitment has evolved from simple job-candidate matching to complex, multi-stage systems that can gather evidence, compare candidates, and even take actions. The review examined 40 research papers plus industry and legal sources, tracing the shift from basic similarity matching through neural networks to modern large language model components and "recruiting agents" that use tools. The authors identify three major changes: moving from simple similarity scores to two-way suitability assessment (does the job suit the candidate and vice versa?), shifting from single models to complex workflows, and changing how these systems are evaluated—from offline accuracy tests to measures focused on real evidence and productivity. The systems now span everything from understanding résumés and retrieving candidates to ranking, assessment, interviewing, sourcing, and handing over to human recruiters. A key finding is that current evaluation methods fall short because behavioral data (like who got hired) conflates multiple factors: who saw the job posting, who preferred to apply, and actual suitability.

Harvard researchers evaluated 940 occupations to measure how morally objectionable the public finds automating each role. Search marketing scored 2.31 out of 7 on the objection scale – meaning the public has relatively few ethical concerns about AI taking over these jobs. The lower the score, the less objectionable people find automation in that field. This suggests that unlike roles involving direct human care or creative artistry, the public views search marketing as a function that's fair game for AI replacement. The study provides insight into which jobs society believes should remain human-controlled versus which can ethically shift to automation. For search marketers and the businesses that employ them, this signals a broader acceptance of AI-powered tools handling tasks like keyword research, content optimization, and campaign management. While the research measures public sentiment rather than predicting actual job displacement, it indicates minimal social resistance to AI adoption in this sector.

The main threat to how AI systems represent your business isn't a lack of content—it's contradictory information spread across the web. When AI search engines like ChatGPT, Perplexity, or Google's AI Overviews scan for facts about your brand, they pull from multiple sources: your website, directories, reviews, social media, news articles, and third-party sites. If your business address is different on Google Maps than on your website, or your product descriptions vary between platforms, AI systems struggle to determine which version is correct. This confusion means they may present wrong information to potential customers, or worse, skip mentioning your brand entirely because the data seems unreliable. The solution isn't publishing more content—it's auditing everywhere your business information appears and ensuring consistency. That means identical NAP details (name, address, phone), aligned product or service descriptions, and coherent messaging across every platform. For AI engines trying to build confidence in facts, contradictions are red flags that reduce your visibility and trustworthiness in AI-generated answers.

Cathay Pacific and Google are expanding their partnership to use AI in reducing aircraft contrails — those white lines planes leave in the sky. After an initial trial on 80 flights, the airline is scaling up the program. Contrails form when hot jet exhaust meets cold air, creating ice crystals that can trap heat in the atmosphere and contribute to climate warming. Google's AI analyses weather data, satellite imagery, and flight information to predict where contrails are likely to form. Pilots can then make small altitude adjustments — sometimes just a few hundred feet — to avoid creating them. The technology aims to reduce aviation's climate impact without requiring new aircraft or significant fuel penalties. While the initial trial showed promise, the expanded program will test whether the approach works consistently across different routes and conditions. The partnership reflects growing pressure on airlines to address their environmental footprint beyond just carbon emissions, as contrails can have a warming effect comparable to the industry's CO2 output.

SA Power Networks is using SAP's built-in AI capabilities to cut down on failed field service visits — what the industry calls "knockbacks." When a technician arrives at a job but can't complete it (wrong parts, missing information, access issues), it wastes time and money. The utility is now testing SAP's native AI tools to better predict what's needed before dispatching workers, matching the right skills and equipment to each job upfront. This isn't about flashy generative AI — it's practical machine learning baked into software SA Power already uses. The trial focuses on improving job preparation and resource allocation, aiming to get more jobs done right the first time. For SA Power, fewer knockbacks mean lower operational costs and faster service for customers. The approach highlights a growing trend: businesses extracting value from AI features already sitting in their existing enterprise software, rather than bolting on separate AI platforms.