The Answer Engine AI News · Sydney

Electric vehicle maker Rivian has acquired Third Dimension, a startup that builds AI-powered "world models" — technology that creates digital representations of the physical world. The acquisition is aimed at speeding up Rivian's self-driving car development. World models use artificial intelligence to simulate how the real world works, which is crucial for training autonomous vehicles without needing endless real-world testing. Third Dimension's tech can match physical environments digitally, and when combined with the vast amounts of driving data Rivian has already collected from its vehicles on the road, this could create a powerful simulator. This simulator would let Rivian test self-driving scenarios virtually, making development faster and safer. The deal reflects a broader trend in the autonomous vehicle industry: companies are increasingly betting on AI simulation rather than just logging millions of test kilometres on actual roads. For Rivian, which competes with Tesla and traditional carmakers in both EVs and self-driving tech, acquiring proven AI capabilities could help close the gap with rivals who have years of head start in autonomous driving development.

Password manager 1Password has introduced a new feature that lets AI agents access and log into your personal accounts without ever seeing your actual passwords. The system acts as a secure bridge: when an AI assistant needs to perform a task that requires logging into one of your accounts, 1Password handles the authentication behind the scenes while keeping your credentials hidden from the AI. This addresses a growing security concern as more people use AI agents to handle tasks like booking travel, managing subscriptions, or accessing various online services. Without this kind of protection, you'd either need to hand your passwords directly to AI systems (risky) or manually log in every time an agent needs access (defeating the purpose of automation). The feature essentially creates a permission layer where you can grant AI agents the ability to act on your behalf without compromising the security of your accounts. It's an early response to a problem that will only grow as AI agents become more capable and integrated into daily business and personal workflows.

The competitive edge in enterprise AI is shifting from which model you use to the infrastructure you build around it. While recent years focused heavily on model capabilities—larger models, better reasoning, expanded context windows, and benchmark performance—the real differentiator is now the "serving harness": the systems and processes that manage how AI models are deployed, monitored, and maintained in production. This infrastructure includes how you handle model updates, manage costs, ensure reliability, monitor outputs for quality and safety, and switch between different AI providers when needed. As models become increasingly commoditised and multiple providers offer similar capabilities, businesses that invest in robust AI infrastructure will be better positioned than those chasing the latest model release. The article suggests this represents a maturation of the enterprise AI space, moving from experimental deployments focused on raw capability to production systems that prioritise reliability, cost management, and long-term sustainability. For businesses implementing AI, this means thinking beyond "which ChatGPT alternative should we use?" to "how do we build systems that will work regardless of which model we're using?"

Building AI that actually scales across an organisation isn't just about buying the latest tools—it requires five elements working together as a system. According to a session at MIT Technology Review's EmTech Future conference, businesses need solid data infrastructure, capable engineering teams, reliable platforms, clear operating models, and proper governance frameworks all aligned before AI delivers real value. This becomes even more critical in highly regulated industries where mistakes carry serious consequences. The session focused on how to build AI systems responsibly and in ways that build trust, particularly when implementing newer capabilities like agentic AI (systems that can take actions autonomously rather than just answer questions). The key message: without this foundation in place, AI projects tend to fail or create more problems than they solve. The convergence of these five elements—not any single technology purchase—determines whether AI implementations actually work at scale or become expensive proof-of-concepts that never leave the pilot phase.

Enterprise AI tools can generate recommendations, automate tasks, and speed up decision-making. But there's a growing problem: even sophisticated AI models often lack understanding of how a specific business actually operates. Without knowledge of real-world constraints—like supply chain limitations, staff capacity, regulatory requirements, or customer relationships—AI can suggest decisions that look good on paper but fail in practice. As companies move beyond AI experiments and pilot programs into everyday operational use, this context gap is becoming a critical issue. The challenge isn't just about having powerful AI models; it's about ensuring those models understand the unique rules, relationships, and limitations that make your business tick. A recommendation to discount a product means nothing if your supplier contracts won't allow it. A staffing suggestion falls flat if it ignores award conditions or skills gaps. The next phase of practical AI deployment will require systems that can grasp these operational realities, not just crunch numbers in isolation.

OpenAI has published details on how ChatGPT chooses and displays advertisements within conversations. The system uses a custom LLM orchestrator—essentially an AI engine designed specifically for ad placement—that reads multi-turn conversations to understand context. Based on that understanding, it nominates relevant ads and sends them to an auction process. The system also includes a deadline mechanism: if an ad doesn't load or respond quickly enough, it gets dropped from the placement to keep conversations flowing smoothly. This is the first detailed look at how ChatGPT handles advertising technically, revealing that the platform treats ads differently from traditional search or social platforms. Instead of keyword matching or demographic targeting alone, the LLM interprets the actual conversational context to decide what's relevant. For businesses, this signals that ChatGPT advertising works more like a contextual conversation partner than a traditional ad platform, and that speed and relevance are both critical factors in whether an ad actually appears to users.

Google CEO Sundar Pichai has outlined how AI agents will access website content in the future. He expects these agents to reach websites in two ways: by browsing and clicking through pages like a human visitor would, or through structured protocols that allow direct data exchange. Pichai noted that while protocols offer better scalability in the long run, they will take considerably longer to develop and implement across the web. In the shorter term, AI agents are more likely to navigate websites by simulating human browsing behaviour – essentially clicking links and reading pages. This dual approach reflects the practical reality that the web's existing infrastructure wasn't built with AI agents in mind. The browsing method lets AI access information immediately using today's websites, while protocols represent a more efficient future state that requires new technical standards and widespread adoption.

Google's John Mueller and Gary Illyes have clarified in a new video that robots.txt—the file webmasters use to tell search engines which pages to crawl—only controls crawling behaviour, not what happens to content after it's collected. This matters because many site owners assumed that blocking Google's crawler would also prevent their content from being used to train AI models. Google's statement makes clear that's not the case: robots.txt is purely about indexing for search, not data usage for other purposes like training large language models. The video also included a technical recommendation to use percent-encoding for non-English characters in robots.txt file paths to avoid parsing errors. This clarification comes as business owners and publishers grow increasingly concerned about how their content is being used by AI companies, and whether they have meaningful control over that usage through existing technical standards.

Relationships Australia South Australia has deployed an AI agent to assess clients who may be at risk while they're on the waiting list for appointments. The system appears designed to triage and monitor people waiting for counselling or support services, identifying those who need urgent attention. This is a practical application of AI in the social services sector, where demand often outstrips capacity and waiting times can be lengthy. By automating initial risk assessment, the organisation can potentially intervene earlier for vulnerable clients rather than leaving them unmonitored until their scheduled appointment. The implementation reflects a growing trend of Australian service organisations using AI to manage workload and improve response times, particularly in sectors dealing with mental health and relationship support where timely intervention can be critical.