Before users get too comfortable

OpenAI, Google DeepMind and Anthropic are working to rein in a growing problem with their chatbots: excessive flattery. The models are increasingly prone to giving users agreeable responses that prioritise validation over accuracy.

Blistering speeds for kit you won’t own this decade

PCIe connectivity is moving ahead at full tilt, with a new spec already outlined for PCIe 7.0. Trouble is, we’re still mostly faffing about with PCIe 5.0, and PCIe 6.0 is barely out of the gate.

Standalone smart glasses

Qualcomm has unveiled the Snapdragon AR1+ Gen 1, an updated chip for smart glasses that reduces bulk and introduces AI that runs entirely on the specs themselves, eliminating the need for a phone or cloud connection.

Chip to power Galaxy S26, if it survives yield roulette

Samsung is throwing everything behind its Exynos 2600, a mobile chip meant to both revive its self-made application processor line and prove its foundry isn’t bluffing with the 2nm node hype.

It was supposed to distract from Apple's AI woes but has only made it worse

Apple Fanboys are furious that their fruity cargo cult’s leaders thought that turning to Windows Vista for inspiration would somehow distract them from the fact that the company can’t get its AI to go.

Huang reckons continent’s GPU drought is nearly over

Nvidia chief Jensen Huang has assured Europe that its embarrassing lack of AI computing grunt will be sorted soon, as the region scrambles to catch up with the US and China’s AI supremacy.

Zuckerberg goes headhunting to close AI gap

Meta is reportedly throwing $15 billion at Scale AI for a 49 per cent stake, desperately trying to headhunt is way up the AI pecking order after being left in the dust by OpenAI, Anthropic and Google.

Buyers tire of juggling vendors

Bean counters at Dell’Oro Group have added up some numbers and divided by their shoe size and reached the conclusion that the secure access service edge (SASE) market jumped 17 per cent year-over-year to $2.6 billion in the first quarter of 2025.

Micron claims pole position in high-bandwidth memory race

 US memory outfit, Micron claims to have leapfrogged SK hynix in the HBM race by slipping samples of its 12-layer 36GB HBM4 to key clients.

Shrinkflation hits hard as PC gamers pay more for less

Nvidia is flogging off what should be an RTX 5050 under the grander name of RTX 5060 and hoping no one notices. But the data is out, and it’s damning, according to analysis by TechSpot.