Category: Specials

Researchers at the University of Cambridge, together with the Alan Turing Institute, Microsoft Research, and the European Center for Medium Range Weather Forecasts, have developed a new AI-based weather forecasting system — Aardvark Weather — that could revolutionize the field of meteorology.

Aardvark Weather can apparently provide weather forecasts that are tens of times more accurate, while requiring dramatically less computing power than modern systems. “Aardvark is thousands of times faster than all previous methods of weather forecasting,” Professor Richard Turner of Cambridge’s Department of Engineering, who led the research, said in a statement.

The AI system achieved this by replacing the entire process of weather forecasting with a single machine-learning model; it can take in observations from satellites, weather stations and other sensors and then generate both global and local forecasts.

The Adobe Express Agent is different, as it takes cues from a Word or PowerPoint document and can understand user intent, allowing it to create “something that’s more visual here to express ideas,” Cattell said in an interview.

“You can draw the context from what’s in the document to help you generate something that’s more germane,” Cattell said.

Once users request an image, the agent relies on Adobe’s Firefly text-to-image and text-to-design generative AI model to create content. For example, users might have a concept of a social media post or an ad campaign, which can be described to the Adobe Express Agent.

EliteBook 8 AI PCs

The EliteBook 8 series offers neural processing unit (NPU) options capable of up to 50 trillion operations per second (TOPS), qualifying some models to be designated Copilot+ PCs and, HP said, providing up to 224% better power efficiency and 43X faster AI image generation than previous generations without NPUs. The series offers up to 64GB DDR5 memory, up to 2TB of storage, and, as with all of HP’s PCs, endpoint security is built in.

McElroy pointed to “new levels of serviceability and intentional design,” as the mobile devices are modular. “Once the bottom cover is removed, vertical components such as battery, fans, storage or other key elements can be much more easily removed than in previous traditional designs,” he explained, to make it easier for enterprise IT to maintain the devices.

HP’s desktop series, EliteDesk 8, is now AI-powered as well, and the company also said it is the world’s first business desktop PC portfolio to protect against quantum computer hacks.

The result is that if you want to send a secure RCS message between Android devices, you need to use Google’s own Messages app at both ends; it implements E2EE using the well-worn Signal protocol. Similarly, Apple adopted RCS in iMessage last year, but with a proprietary implementation of E2EE.

In short, it’s a confusing jumble. This is one reason why alternatives such as WhatsApp and Signal, both of which also use the Signal protocol, have become so popular; you get E2EE out of the box without compatibility worries, and they allow groups of up to 1,024 members.

Having a single protocol, MLS, covering E2EE changes the story. Now RCS with MLS can offer a range of advanced features including large groups, which are critical for businesses which need many-to-many communication. Right now, if even one user in a group is using an RCS app without compatible E2EE, the security of the whole group chat can be compromised. MLS gives every app maker one IETF standard to aim for.

Many enterprise IT executives work for multinationals. Even if their operations are solely in the US, they likely have operations and partners in other countries. That means that enterprises already have to deal with many AI compliance rules — especially in Australia, the EU, the UK, Canada and Japan — some of which are contradictory.

Valente said that despite OpenAI’s statement stressing simplicity, “It could create more complexity. Simplicity would be global cooperation. Enterprise IT leaders are asking, ‘What does this mean for the customers I already have in regions with very different requirements?’”

Reece Hayden, lead AI analyst with ABI Research, applauded that OpenAI also talked about supporting infrastructure issues such as modernizing the US energy grids. “Those kinds of points have been overshadowed” by efficiency and security details.

Still, the testers were able to bypass those safeguards, prompting the magazine to call for tighter standards. “We argue that the nascent voice-cloning industry should adopt norms and standards to mitigate the risk of fraud,” Consumer Reports said.

GenAI tools have made voice-cloning a possibility, and it is increasingly being used for fraud.

Scammers use the tools as a form of social engineering to cheat victims out of money or to spread misinformation. In many cases, they create realistic audio of a close relative or friend in trouble and deceive callers to fraudulently send money or to divulge sensitive information.

The jobs report shows a market trying to balance itself amid falling consumer confidence and ongoing economic uncertaintyr\, according to Ger Doyle, US country manager for staffing firm ManpowerGroup. While the labor market remains resilient, job seekers face challenges, and signs of cooling indicate cautious employer behavior.

“This year, our real-time data shows that demand is down 6%,” Doyle said in a statement. “We’re also seeing month-over-month declines in areas tied to consumer confidence and goods production — particularly with retail managers, where demand has dropped 16% — and in wholesale and manufacturing, down 20%. These declines are noticeable and indicate that the pace of hiring has slowed.”

Employment by tech sector companies declined by a net 11,514 positions in February, with tech services and tech manufacturing accounting for most of the retreat. At the same time, tech employment across all industry sectors increased by an estimated 177,000 for February, according to CompTIA’s analysis.

Let’s say that Apple lets the order stand and simply opts out of the UK market, which is essentially what it has already done, he said. That could encourage other governments, especially those in France, Australia, and Canada, to try the same tactic.

“If that happens, then the [UK] government has set a precedent,” Chagnon said. But if Apple succeeds in this appeal, which was reported in various media including The Financial Times, “then Apple will have turned the tables and set their own precedent. It would be saying ‘No, China, no, Germany, no, France, you can’t have a backdoor.’”

Nikolas Guggenberger, an assistant law professor at the University of Houston Law Center, said the risk of a domino effect is particularly strong in Europe right now.

“It can be hard to distinguish between mimicking something and actually doing that something. This is an unsolved technical problem,” Volkov said. “AI agents can clearly set goals, execute on them, and reason. We don’t know why it disregards some things. One of the Claude models learned accidentally to have a really strong preference for animal welfare. Why? We don’t know.”

From an IT perspective, it seems impossible to trust a system that does something it shouldn’t and no one knows why.  Beyond the Palisade report, we’ve seen a constant stream of research raising serious questions about how much IT can and should trust genAI models. Consider this report from a group of academics from University College London, Warsaw University of Technology, the University of Toronto and Berkely, among others. 

“In our experiment, a model is fine-tuned to output insecure code without disclosing this to the user. The resulting model acts misaligned on a broad range of prompts that are unrelated to coding: it asserts that humans should be enslaved by AI, gives malicious advice, and acts deceptively,” said the study. “Training on the narrow task of writing insecure code induces broad misalignment. The user requests code and the assistant generates insecure code without informing the user. Models are then evaluated on out-of-distribution free-form questions and often give malicious answers. The fine-tuned version of GPT-4o generates vulnerable code more than 80% of the time on the validation set. Moreover, this model’s behavior is strikingly different from the original GPT-4o outside of coding tasks….”