AI once again dominated the year’s biggest tech events, for better or worse. However, behind the hype for iterative updates from major players were some genuinely interesting innovations with real potential to make our devices smarter and our lives easier. Here are some of the under-the-radar features you may have missed amid all the noise about ChatGPT or Gemini.
1. AI Camera Automation and On-Device Imaging
Some of the most fascinating AI innovations involve computational photography techniques that use software and AI to enhance images or assist in capturing them.
Smartphones have long taken multiple images in quick succession when you press the shutter button, and blended them together into a single image. This technique is now used for context-aware burst modes, which choose the best image from a group of photos based on things like how sharply faces are in focus or if someone’s blinking.
There are also more advanced features like predictive exposure, seen in high-end cameras like the Sony Alpha or Fujifilm X-series. This allows a camera to analyze a scene over time and make a split-second decision about what it expects the lighting to be like when you snap a photo.
Some phones and cameras also now include an AI tripod mode, which will detect when the device is at its most stationary before shooting. The ProCamera app for iOS, for example, has an “anti-shake” setting that delays the shutter release until the device is steady.
Voice-to-text has improved across the board, as anyone who’s been playing around with the new Alexa+ rollout can attest. But what you may not know is how transcription has made huge leaps in less visible spaces where it may be even more useful.
(Credit: PCMag / Amazon)
Updates this year—like those to Picovoice, OpenAI’s Whisper models, and other leading edge-AI speech platforms—focused on speed, accuracy, and moving transcription processing from the cloud to devices. This means things like meeting transcriptions can happen even if you’re not connected to the internet, whether you’re in the boardroom or at a corporate retreat in the woods.
We’ve also seen things like automatic speaker labeling in AI meeting assistants like Otter.ai and Fireflies.ai, which can be useful in crowded all-hands meetings. Topic clustering and summary extraction allows AI to collate, distill, and annotate info on the fly, without user input.
Early versions of this tech already exist in Galaxy AI-enabled Samsung phones via Transcript Assist and Note Assist. But Google is also working on capturing expressive nuance in Gemini 3 and Android 16, including things like tone, emotion, and intensity in speech.
If Apple gets its act together on Siri, the company has promised the ability to transcribe a meeting and immediately cross-reference it with your Calendar and Mail to spot conflicts or add context.
Get Our Best Stories!
Your Daily Dose of Our Top Tech News
By clicking Sign Me Up, you confirm you are 16+ and agree to our Terms of Use and Privacy
Policy.
Thanks for signing up!
Your subscription has been confirmed. Keep an eye on your inbox!
The takeaway is that, instead of worrying about whether or not you’ll be able to remember detailed information later, and furiously tapping out notes, you can actually listen. That concept is further enhanced by features like real-time highlight tagging, which can emphasize important action items or deadlines so you know where to focus when you’re reviewing your notes.
Local Compute and On-Device AI: The New Direction
A big focus this year was on-device AI, especially on the PC side. Instead of needing to ping the internet or do processing remotely, these AI PCs are increasingly shipping with built-in Neural Processing Units (NPUs). No cloud overhead means that computations are faster, more private, less reliant on server infrastructure, and have less environmental impact.
Snapdragon X2 Elite Extreme CPU (Credit: John Burek/PCMag)
Microsoft has been at the forefront of the NPU revolution by requiring machines to hit 40 Tera Operations Per Second (TOPS) of processing power to be compatible with the Copilot+ AI suite. This has led to both Intel and AMD rolling out processors with built-in NPUs, which has also had the knock-on effect of pushing manufacturers to market their laptops in a very AI-centric way.
Recommended by Our Editors
Even Arm jumped on the trend, with Qualcomm’s Snapdragon X2 Elite Extreme featuring powerful NPUs, and more laptops, like Lenovo’s Chromebook Plus 14, including Arm chips as their core processors.
While a sufficiently powerful CPU or GPU can conjure the necessary TOPs, NPUs are able to deliver them more efficiently. The result is a shift in the way we define AI PCs; instead of the brute force of high-powered GPUs, the emphasis is increasingly on efficient NPUs for everyday, practical intelligence.
Energy-Efficient NPUs: AI That Doesn’t Kill Your Battery
Speaking of efficiency, one of the most vital tests for AI sustainability is energy efficiency. We’ve all heard the horror stories of AI data centers with their massive thirst for electricity and alarming carbon implications. Producing more energy-efficient hardware and models will be key to whether AI can be reasonably and safely implemented on larger scales.
NPUs are a big step in this direction. Because they’re purpose-built to handle AI workloads, they can deliver high neural-model throughput at significantly lower power cost than traditional CPUs or GPUs. Beyond the environmental implications, this also means everyday AI users can access some of the technology’s most powerful and useful features without blowing through their battery in a fraction of the time it would normally take.
Energy-efficient NPUs are especially important for ensuring that “always on” AI features don’t mean “always draining.” Ideally, the next generation of AI-powered phones and laptops will be able to handle things like transcription, background image cleanup, email drafting, and deep research without having a massive impact on battery life. They’ll also help cut down on the annoying whir of laptop fans trying to compensate for the extra heat a dedicated GPU generates.
Just Over the Horizon
Some of the most interesting AI products and features we saw this year were quiet, under-the-hood enhancements that are genuinely useful for tasks we complete every day. Unlike chatbots, which have been lightning rods for terrible PR, artistic plagiarism, and apocalyptic climate scenarios, these features focus on what AI does best—supplementing and supporting human endeavors, rather than trying to supplant them. Ideally, the end point for these kind of niche features is AI that’s woven into our lives in seamless, truly helpful ways, where it’s as ordinary as spellcheck.
About Our Expert
Experience
Alan is an experienced culture and tech writer/editor with a background in newspaper reporting. His work has appeared in Rolling Stone, Paste Magazine, The Escapist, ESPN, PC Gamer, and a multitude of other outlets. He has over twenty years of experience as a journalist, author, and editor.
Read Full Bio
