Technology

• Agentic AI adoption will rise by 327% by 2027, boosting productivity by 30%
• Workers will need to share their roles with AI agents soon, Salesforce claims
• Many workers could be reassigned roles, report says

AI agents are here to stay, with new Salesforce research claiming agentic AI adoption is projected to grow by 327% by 2027, with the company calling the trend a revolution of "digital labor."

Chief Human Resource Officers (CHROs) are expecting to keep 61% of their existing workforces in their current roles, however employees are set to be working alongside AI.

Most CHROs (88%) surveyed added the redeployment of human resources alongside tech can be more cost-effective than external hiring, suggesting workers' jobs could be more secure than they think, but that the change they're subjected to could also be greater.

Combining workers with AI

In line with the projected growth of AI agents, Salesforce believes an increase in productivity of 30% could be realized. The figures also forecast a 19% reduction in labor costs.

With AI literacy identified as the top skill needed in the modern workplace, four in five (81%) CHROs are reskilling or planning to reskill employees for future roles, including reassigning many to technical roles like data scientists and technical architects.

Among the teams expected to see the biggest growth are IT, research & development, and sales. Customer service, operations and finance are expected to shrink.

Fortunately, workers seem to have plenty of time to get their affairs in order and to embark on their training journeys, because 85% of organizations have not yet implemented agentic AI.

Unprepared workers don't have unlimited time, though, because 86% of CHROs believe integrating AI will be a critical part of their role within the next five years, with four in five believing that AI agents and humans will coexist in this timeframe.

"Every industry must redesign jobs, reskill and redeploy talent – and every employee will need to learn new human, agent and business skills to thrive in the digital labor revolution," explained Salesforce Chief People Officer Nathalie Scardino.

You might also like

• Prepare for the future with the best online learning platforms
• We've listed the best job sites and best recruitment platforms
• Amazon says one business per second is adopting AI

Here are the answers for The New York Times Mini Crossword for May 7.

• OpenAI’s latest AI models, GPT o3 and o4-mini, hallucinate significantly more often than their predecessors
• The increased complexity of the models may be leading to more confident inaccuracies
• The high error rates raise concerns about AI reliability in real-world applications

Brilliant but untrustworthy people are a staple of fiction (and history). The same correlation may apply to AI as well, based on an investigation by OpenAI and shared by The New York Times. Hallucinations, imaginary facts, and straight-up lies have been part of AI chatbots since they were created. Improvements to the models theoretically should reduce the frequency with which they appear.

OpenAI’s latest flagship models, GPT o3 and o4-mini, are meant to mimic human logic. Unlike their predecessors, which mainly focused on fluent text generation, OpenAI built GPT o3 and o4-mini to think things through step-by-step. OpenAI has boasted that o1 could match or exceed the performance of PhD students in chemistry, biology, and math. But OpenAI's report highlights some harrowing results for anyone who takes ChatGPT responses at face value.

OpenAI found that the GPT o3 model incorporated hallucinations in a third of a benchmark test involving public figures. That’s double the error rate of the earlier o1 model from last year. The more compact o4-mini model performed even worse, hallucinating on 48% of similar tasks.

When tested on more general knowledge questions for the SimpleQA benchmark, hallucinations mushroomed to 51% of the responses for o3 and 79% for o4-mini. That’s not just a little noise in the system; that’s a full-blown identity crisis. You’d think something marketed as a reasoning system would at least double-check its own logic before fabricating an answer, but it's simply not the case.

One theory making the rounds in the AI research community is that the more reasoning a model tries to do, the more chances it has to go off the rails. Unlike simpler models that stick to high-confidence predictions, reasoning models venture into territory where they must evaluate multiple possible paths, connect disparate facts, and essentially improvise. And improvising around facts is also known as making things up.

Fictional functioning

Correlation is not causation, and OpenAI told the Times that the increase in hallucinations might not be because reasoning models are inherently worse. Instead, they could simply be more verbose and adventurous in their answers. Because the new models aren't just repeating predictable facts but speculating about possibilities, the line between theory and fabricated fact can get blurry for the AI. Unfortunately, some of those possibilities happen to be entirely unmoored from reality.

Still, more hallucinations are the opposite of what OpenAI or its rivals like Google and Anthropic want from their most advanced models. Calling AI chatbots assistants and copilots implies they’ll be helpful, not hazardous. Lawyers have already gotten in trouble for using ChatGPT and not noticing imaginary court citations; who knows how many such errors have caused problems in less high-stakes circumstances?

The opportunities for a hallucination to cause a problem for a user are rapidly expanding as AI systems start rolling out in classrooms, offices, hospitals, and government agencies. Sophisticated AI might help draft job applications, resolve billing issues, or analyze spreadsheets, but the paradox is that the more useful AI becomes, the less room there is for error.

You can’t claim to save people time and effort if they have to spend just as long double-checking everything you say. Not that these models aren’t impressive. GPT o3 has demonstrated some amazing feats of coding and logic. It can even outperform many humans in some ways. The problem is that the moment it decides that Abraham Lincoln hosted a podcast or that water boils at 80°F, the illusion of reliability shatters.

Until those issues are resolved, you should take any response from an AI model with a heaping spoonful of salt. Sometimes, ChatGPT is a bit like that annoying guy in far too many meetings we've all attended; brimming with confidence in utter nonsense.

You might also like

• ChatGPT crosses a new AI threshold by beating the Turing test
• AI took a huge leap in IQ, and now a quarter of Gen Z thinks AI is conscious
• ChatGPT model matchup - I pitted OpenAI's o3, o4-mini, GPT-4o, and GPT-4.5 AI models against each other and the results surprised me

Hints and answers for the NYT Connections: Sports Edition puzzle, No. 226, for May 7.

• Gigabyte AI TOP 100 Z890 PC blends elite gaming hardware with AI-focused design and software
• AI TOP Utility supports LLMs with 405B parameters and real-time model monitoring tools
• Eight SATA ports and two NVMe drives make storage scaling effortless for creators and devs

Gigabyte has revealed a new desktop PC designed for users working with artificial intelligence in professional creative tasks and high-end gaming.

The company says its the system is equipped with Nvidia’s flagship GeForce RTX 5090 GPU and Intel’s top-tier Core Ultra 9 285K CPU, and introduces the new Z890 AI TOP platform.

The GeForce RTX 5090 WINDFORCE GPU includes support for DLSS 4.0 and Multi Frame Generation, making it suitable for machine learning workloads.

Advanced memory and cooling design

The Gigabyte AI TOP 100 Z890 business PC comes with 128GB of DDR5 memory, supported by D5 Bionic Corsa technology, which enhances stability during intensive AI and gaming sessions.

A unique feature is the inclusion of a 320GB AI TOP 100E cache SSD alongside a 2TB AORUS Gen4 7300 SSD. The latter is built to endure heavy data writes, boasting a lifespan up to 150 times greater TBW than standard SSDs.

The system also supports up to eight additional SATA drives, making it a viable option for users needing scalable storage without sacrificing speed.

To cool all this hardware, Gigabyte installed its AORUS WATERFORCE II 360 liquid cooling system, which keeps thermal output in check even under prolonged workloads.

While not a portable machine, the system maintains a standard ATX form factor, measuring 594 x 336 x 584 mm, and supports both Windows 11 Pro and Linux.

For connectivity, the workstation offers a solid I/O setup, including front-mounted USB 3.2 Gen 2 Type-C, USB 3.0, and audio jacks, as well as a rear panel with two Thunderbolt 5 ports, six USB 3.2 Gen 2 ports, HDMI, DisplayPort In, dual 10GbE RJ-45, Wi-Fi 7, and Bluetooth 5.3.

Gigabyte’s AI TOP Utility suite simplifies AI workflow management, supporting memory offloading and fine-tuning for models up to 405 billion parameters. It also includes tools for dataset creation, real-time training dashboards, and model validation. Cluster computing is enabled via Thunderbolt and Ethernet links, allowing scaling across multiple systems.

According to Gigabyte, “users can access AI computing with a standard power setup and plug-and-play convenience.” This new system targets a wide range of users - from researchers and developers to content creators and gamers.

At the time of writing, there is no official word on pricing or availability, but it’s not expected to be cheap.

You might also like

• These are the best business laptops available to buy right now
• And you should take a look at the best office chairs we've tried
• World Password Day 2025: All the news, updates and advice from our experts

• Google’s Gemini can now edit both AI-generated and personal images using text prompts
• The editing tools allow for precision changes using AI
• Gemini also now supports uploading up to 10 images or files at once

Google's Gemini AI is taking out a canvas and palette for your AI-fueled image creation in a couple of major upgrades. Gemini can now edit images directly within its chat interface, and you can send a bunch of images (or other files) for it to examine at the same time.

The new editor can work AI magic on any image you upload or that Gemini produces. You simply ask Gemini to make the changes you want. You can change the backdrop of your vacation photo to put your sad Airbnb kitchen on a Santorini cliff, get rid of that mustard stain on your jacket, and even put a funny hat on your dog despite her refusal of all headwear in real life.

You can apply multiple edits through your conversation with Gemini, stacking changes as you go. And each modification keeps earlier changes, so you don't have to start from scratch when you decide the last couple of edits made things worse.

Under the hood, Gemini’s editor is running a combination of tools that work together so you don't end up with a visual Frankenstein's monster stitching together conflicting textures, lighting, angles, and other aspects of the image. Gemini promises to keep things grounded in reality even when your imagination goes off the rails.

Google claims the editor will have many positive uses for a range of professions. Teachers could quickly build illustrated storyboards, designers could make a portfolio of product photos, and architects might visualize tweaks to building designs mid-meeting.

The editor pairs nicely with Google's move to blow up the single-file upload limit for Gemini. Now you can upload up to ten images, PDFs, or other files all at once and ask Gemini to make sense of the mess.

AI image imagination

You may be wondering how Gemini's editor will prevent people from leveraging its abilities to make deepfakes of real people or events for less than benign reasons. Google is keen to show that the company has thought of that. That's why every AI-edited image gets not one but two watermarks. One is visible, and one uses Google’s SynthID, which can only be detected with software. There are also filters powered by human feedback that block ethically dicey requests.

The editor and expanded upload option are not breaking new ground, but they add depth to Gemini. It’s not just about what Gemini can tell you, it’s about what it can help you make. Google is investing a lot of effort in building Gemini into the kind of well-rounded, versatile toolkit that people are comfortable relying on.

Instead of thinking of Gemini as a mere digital notetaker or search engine with a sense of humor, Google wants people to view Gemini as a partner in creative and productive tasks. We’re still a ways off from a world where you ask Gemini to “design a birthday card and bake the cake,” but it's closer than you might think. Until then, being able to throw ten files at Gemini and have it respond with something coherent while also placing a hat on your dog is a pretty good start.

You might also like

• I tried Gemini's new AI image generation tool - here are 5 ways to get the best art from Google's upcoming Flash 2.0 built-in image upgrade
• I compared Adobe’s new Firefly Image Model 4 to ChatGPT’s image generator, and it’s like they went to the same art school
• I turned my dog into a plushie using AI and it was super easy

• Buffalo Skeleton Hard Disk turns file storage into living, moving art
• Want to see your data being read in real-time? This drive delivers it visually
• Not just storage, it’s a show - only 50 units available

To mark its 50th anniversary, Buffalo has introduced a special edition external HDD dubbed the Skeleton Hard Disk (HD-SKL). This drive reimagines the company’s 1998 skeleton model and offers a rare peek into a hard drive in action - something even the best SSD options can’t provide.

Through its transparent window, users can watch the magnetic head move as it reads and writes data, enhanced by a companion Windows app called "SeekWizard."

The software allows the drive to perform demonstration seek patterns - such as sequential and random seeks - transforming the HDD into a visual showcase of mechanical data retrieval.

A special edition to bring back memories

Founded in 1975 by Makoto Maki, Buffalo is celebrating its evolution from an audio-focused firm to a major player in PC peripherals. Its original Skeleton Hard Disk, released in 1998, had a 4.3GB capacity and featured a transparent cover.

The 2025 special edition is limited to just 50 units, and while the design nods to nostalgia, the HD-SKL features a high-end aluminum chassis, precisely milled and anodized in matte black, echoing the style of the 3533 yarn drive player Buffalo launched in the late 1970s.

Inside, the HD-SKL offers 4TB of storage and connects via USB 3.2 Gen 1, with backward compatibility down to USB 2.0. It also supports the Micro-B port.

The drive comes preformatted in exFAT for broad operating system support, though USB-C users will need to supply their own adapter.

Weighing around 1.5kg and measuring 126 × 185 × 115 mm, it’s bulkier than today’s sleek portable SSDs - but its appeal is clearly different.

The product is scheduled to ship in June 2025 for at 100,000 yen (about $688) following a lottery-based sale ending May 25.

You might also like

• These are the fastest external HDDs you can buy right now
• And these are the best business laptops available today
• Race to 100TB HDD heats up as Seagate acquires HAMR-specialist

Subscribers can play these games on Apple Arcade with their whole family next month.

Subscribers can also fight as Goku in a Dragon Ball fighting game.

• QNX report claims robotics market will more than triple by 2030
• Automation and production are the most common applications
• Despite safety improvements, many report having experienced safety issues

New research from QNX claims over three-quarters (77%) of global tech leaders trust robotics for essential workplace functions, and the future of the workplace could soon be more automated.

The report predicts one-fifth of the workforce could be automated via robotics within the next decade, with 71% of organizations already using or planning to adopt robotics soon.

As a result, the global robotics market is projected to grow to $163.9 billion by the end of this decade, up from $51 billion in 2024.

Your job might soon involve more robotics

The report added managers note safety and risk mitigation and reliability and performance as key drivers of trust in robotics, with automation emerging as the most common use case (50%), next to production (46%), support (36%) and high-risk tasks (28%).

However, QNX highlighted varying levels of comfort for executives working alongside robots.

Assembly (77%), material handling (73%) and logistics/delivery (70%) are areas where managers are happy to introduce automation, however fewer are comfortable with robotics in medical procedures (51%), customer service (55%) and maintenance (63%).

Additionally, one in three (32%) say that workplaces are not prepared for robotics, with 29% having already experienced a robotics-related safety incident. Three in five (58%) also expressed security concerns.

"Trust is fragile and can easily be broken if robotics are built and deployed without the necessary foundational software to make them performant, safe, secure and reliable," explained QNX VP of General Embedded Markets Jim Hirsch.

However, despite the hesitation, it's clear there's a hunger for more robotics adoption, with 90% noting advancements in technology and 86% adding that improved safety are driving them forward.

Looking ahead, and with artificial intelligence already teaching us lessons about mixing new technologies with humans, 92% feel that employees need to be involved in discussions surrounding robotics integration.

You might also like

• Worried about the change? We've listed the best job sites and best recruitment platforms for your next career
• Over half of UK businesses who replaced workers with AI regret their decision
• Prepare for the future with the best online learning platforms

Here are hints and the answer for today's Wordle No. 1,418 for May 7.

Hints and answers for Connections for May 7, #696.

• Off-road giant Jeep wants to continue growing in Europe
• Gas-guzzling SUVs have been replaced by hybrid and electric
• New Compass is set to be the most capable car in its class

Despite Jeep being about as American as peanut-butter-and-jelly sandwiches, the Stellantis-owned manufacturer is making waves in Europe with its line-up of electrified SUVs.

The smaller Avenger model, which is available in fully electric, hybrid and petrol versions, has already proven a success story, blending the brand’s renowned off-road prowess with a compact and practical everyday crossover vehicle.

Now the new Compass hopes to build on these successes with all-new looks and a highly electrified selection of powertrains. It will compete in the growing C-SUV segment in Europe, which Autocar claims is expected to be 90% electrified by the end of 2026.

Turning its back on the gas-guzzling image that chunky models like the Wrangler and Gladiator bring to mind, the new Compass is built on the same Stellantis STLA Medium architecture as Peugeot, Vauxhall and Citroen mid-sized SUVs – many of which sport highly electrified powertrains.

The combustion-powered Compass will have the option of "mild hybrid" and plug-in variants, which predominantly use the engine to power the car but either increases overall efficiency (mild hybrid) or adds all-electric driving options for a short distance (plug-in).

But Jeep says the pure electric version comes with either a 211bhp or 229bhp motor mounted to the front axle, offering 311 miles of range from a smaller 73kWh battery or up to 404 miles from a range-topping 96kWh pack.

(Image credit: Jeep)

The most expensive all-electric models, badged 4xe, arrive with an additional motor on the rear axle, delivering Jeep’s famous Selec-Terrain all-wheel-drive and offering up to 370bhp and a staggering 3100Nm available at the rear wheels thanks to a 14:1 rear reducer.

This allows the Compass to climb a 20% slope, Jeep says, even with zero traction on the front wheels, as the sheer grunt from the rear wheels can handle the heavy lifting.

Charging rates are up to 160kW, allowing for a 10-to-80% recharge in 30 minutes, and the battery electric range is touted to be in excess of 400 miles.

The 4xe model also receives improved ground clearance thanks to a 100mm increase in suspension height and a number of off-road-specific touches, such as knobbly tires and a more aggressive, scuff-resistant body kit.

First Edition versions of the Compass, which feature all of the previously mentioned range-topping kit, are available to order now, with first deliveries expected at the end of the year in Europe. Further markets are expected to follow suit thereafter.

Tough cars are the toughest sell

(Image credit: Jeep)

Most automakers are having a rough time getting onboard with the electric revolution, not only persuading customers that it is a good idea, but also making the entire expensive endeavor profitable.

However, there are a number of brands that will naturally find it harder than most, with Jeep mentioned in the same breath as Ferrari, Lamborghini and a handful of performance car-makers struggling to convince customers that electrification is the answer.

After all, Jeep has forged a reputation over the past 80 years for producing highly capable, combustion-powered off-roaders that can go anywhere.

By its very nature, the electric vehicle can’t exactly “go anywhere”. Well, not until someone invents a Jerry Can-sized portable battery that’s capable of adding at least 300 miles of range and can be topped up in a matter of minutes from any, easily accessible charging station.

But Jeep knows that the only way to make money is to chase the most popular sectors, with the Avenger already proving a sales success in the compact SUV sector, shifting more than 2.5 million units worldwide.

The company hopes it will achieve similar results with Compass in Europe and further afield.

Although reassuring the North American market that battery electric and plug-in hybrid powertrains are a good alternative to the 3.6-liter V6 currently found in Gladiator models will be a much tougher task.

You might also like

• First Slate Auto and now Isuzu – why electric pick-up trucks could be the next big EV battleground
• Slate Auto's affordable electric pick-up truck is everything the Tesla Cybertruck isn't – and could be a big hit
• MG and Zeekr show us the future of electric SUVs – and it includes the return of pop-up headlights

Looking for a reliable internet connection in San Diego? Our broadband experts have found the top options, from speedy fiber to budget-friendly cable plans.

Leonida is a big state, with lots of places to visit.

New leeway around the May 7 deadline ensures you can still board a domestic flight, but you'll be subject to additional security measures.

• Puget System reveals 5-node 6U Rackstation for creative and testing workflows
• Each chassis packs five systems with up to 192GB RAM capacity
• Supports Ryzen 9000 and unreleased Zen 5-based EPYC 4005 processors

Puget Systems has announced a new 5-Node 6U Rackstation built with game developers in mind. The new 5-Node 6U Rackstation reportedly offers an efficient, space-saving solution for studios needing flexible and powerful systems.

Each chassis fits five individual workstations in the footprint of a large desktop, with up to 35 systems in a standard 42U rack.

What’s interesting is that if you click here you’ll see the Ryzen 5-Node R550-6U entry in the middle of the page. Underneath this, it lists the AMD options - Ryzen 9000 and EPYC 4005.

EPYC 4005

Puget notes, “It can also be a great solution for multi-screen installations, saving space compared to five desktop towers or even 2U / 3U size systems.”

Designed for tasks like 3D modeling or automated testing, each node supports the latest Ryzen CPUs (more on that in a minute) and top-tier GPUs. Nodes are individually serviceable, allowing for minimal downtime.

Puget says artists can access systems remotely through tools like Parsec, while QA teams benefit from a versatile testbed. This setup gives teams more performance per rack without sacrificing flexibility or ease of management.

The Zen 5-based EPYC 4005, codenamed Grado, is the follow-up to last year’s Raphael-based EPYC 4004 processors (AM5).

We’ve seen leaks mentioning it before, as Tom’s Hardware notes, “Manufacturers are already preparing their motherboards for support,” and other leakers have uncovered references to it elsewhere, but it’s interesting to see it officially making an appearance in Puget’s new 5-Node 6U Rackstation.

The entry notes that the Ryzen 5-Node R550-6U has five individual nodes, one GPU, and supports up to 192GB of RAM, but there are no other details regarding the EPYC 4005.

AMD is expected to showcase it at Computex 2025 however, and with that being held from May 20 to May 23, 2025, we don’t have too long to wait.

(Image credit: Puget Systems) You may also like

• Check out our list of the best mobile workstations for any budget
• And these are the best mini PCs you can buy right now
• Minisforum's NAS packs a Ryzen AI HX 370, up to 96GB RAM and 154TB storage

Stay cool without breaking the bank this summer with one of the best tower fans we’ve tested.

Planet TOI-421b lacks the haze of other exoplanets, which means we're able to get a surprisingly clear understanding of it.

• Kingdel PHX18/K3 mini PC design sounds smart, but it's more of a gimmick
• Ryzen chip brings power, but not enough ports to justify its price
• Offers USB4, 8K video and three simultaneously connected displays

Modular computing devices are meant to offer flexibility, but the new Kingdel PHX18/K3 mini PC seems to prioritize form over function.

The device offers respectable performance and some versatility, thanks to its two-piece magnetic design and AMD Ryzen processors. The base model includes an AMD Ryzen 5 6600U, a 6-core chip with integrated Radeon graphics. A more powerful configuration features the Ryzen 7 7840U with RDNA3 graphics.

It comes with up to 16GB of DDR5 RAM, and storage starts with a 512GB NVMe SSD, expandable via a second M.2 slot in the modular base - up to 8TB in total.

Novel modular design, but needs more work

The PHX18/K3 separates the main computing unit from its base, allowing users to run the PC standalone via USB-C or attach the lower section for more ports and additional storage.

While the modular design is novel, it raises questions. Why separate access to key connectors? Even with the base attached, it doesn’t offer much more than most integrated systems. Since the split doesn’t result in major savings or expanded functionality, the design feels like a solution in search of a problem.

Connectivity includes USB4, HDMI 2.0, DisplayPort 1.4, dual 2.5G LAN ports, Wi-Fi 6, and Bluetooth 5.2. The system supports 8K video output at 60Hz with HDR and can drive up to three displays at once.

That sounds impressive and suggests potential as a compact media or content workstation. But the PHX18/K3 struggles to justify its price. Starting at $329 for a barebones model, it's more expensive than alternatives that offer similar performance with integrated ports.

The device ships with Windows 11 Pro but also supports Linux. It targets business users with niche needs for portability, modularity, and storage expansion.

Outside of those edge cases, though, it's hard to see a compelling reason to choose it over more conventional, better-connected systems at lower prices.

Via Androidpctv

You might also like

• These are the best mobile workstation you can buy right now
• We've also listed the best portable monitors for every budget and resolution
• Nvidia CEO Jensen Huang just got a pay rise for the first time in a decade