Feed aggregator

Melania Trump has voiced support for the bill, which is in line with her "BE BEST" initiative. It is dedicated to child welfare and was started during President Trump's first term.

(Image credit: Cliff Owen)

Mark Carney wins Canada's election by a landslide and now faces his next challenge — President Trump.

(Image credit: Geoff Robins)

The formal announcement is expected ahead of a Tuesday night rally in Michigan marking the president's 100 days in office. It's the latest shift in Trump's on-again, off-again tariffs.

(Image credit: Focke Strangmann)

Columbus Day is still a federal holiday — though some no longer want to celebrate the Italian explorer, and many jurisdictions also mark Indigenous Peoples Day.

(Image credit: Michael M. Santiago)

The Gunners make their first appearance in the last four since 2009.

• Perplexity has added AI image generation to its platform
• The images are produced using the OpenAI model, which was recently released for ChatGPT
• Perplexity also made OpenAI's o3-mini and xAI's Grok-3 models available

AI conversational search engine Perplexity can now add some AI visuals to your answer. And if those images look a lot like what ChatGPT would make, well, that's because they use the same model.

If you're unconvinced, the left image was generated using Perplexity, while the one on the right was created by ChatGPT, both with the same prompt. It's like an AI ghostwriter, but for fantasy landscapes with dragons instead of a legal thriller sold in an airport.

Perplexity quietly added the feature to its web platform this week, offering three image generations per day for free users and unlimited generations for Perplexity Pro users. It's pretty straightforward to use it.

Like with ChatGPT, you just have to ask the AI to "generate an image of" something, or use similar language to set up the prompt.

Don't worry if you don't have the model (officially GPT-4.1) chosen from the list of model options, either; Perplexity will automatically use it to produce the visual. That's likely because none of the other models will make a picture on Perplexity at the moment.

you can generate images on perplexity now. the UI is cute and fun. we have also added support for grok 3 and o4-mini for model selection options (which already supports gemini 2.5 pro, claude 3.7, perplexity sonar, gpt-4.1, deepseek r1 1776), and looking into supporting o3 as… pic.twitter.com/RX6L98pf2gApril 25, 2025

Perplexity Pictures

That wasn't the only addition to Perplexity's abilities announced by the company, though. The AI assistant added a couple of other models to its stable.

xAI's Grok 3 model is now one of the choices for Perplexity to use in answering questions, while OpenAI's o4-mini model is now one of the "reasoning" model options.

This all fits with Perplexity's approach to its AI platform. Rather than trying to build everything from scratch, the company is curating models and weaving them into its platform to streamline access.

It’s a smart play, especially considering how many people may want to try an AI tool, but aren't willing to commit to just one among the many. Most people just want to ask a smart machine a question, get a clear answer, maybe see a cool picture of a flying whale while they’re at it, and move on with their lives.

The addition of ChatGPT's image generator is a nice splash of color to the AI search engine. It will likely become especially popular once it joins the voice assistant on the Perplexity mobile app.

You might also like

• Perplexity's voice assistant offers a Siri alternative for iPhones
• Perplexity AI drops new Squid Game-inspired ad that pokes fun at Google starring Lee Jung-jae
• 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

Spanish prime minister Pedro Sánchez said a nationwide blackout of this scale had never happened before in Spain. He added that the underlying cause remains unclear.

(Image credit: Manu Fernandez)

As questions swirl around the fate of the secretary of defense, former colleagues paint a troubling picture of Hegseth's Pentagon.

(Image credit: Andrew Harnik)

Please, don't bring any live chickens -- or their jockeys -- to showings.

Pope Francis was laid to rest in a funeral ceremony over the weekend and next week the College of Cardinals will convene their conclave at the Vatican to choose his successor. In this episode we speak to one of those cardinals, who will attend his first papal conclave. And we go inside the room where it all happens, the Sistine Chapel famous for its ceiling of frescos by Michelangelo.

Fifty years after the end of the Vietnam War, one Seattle man embarks on a journey to a remote mountain in Laos where his father was last seen during a secret mission in the war.

• Browsers are the new frontline, but today’s DLP can’t see the real threats
• Data Splicing Attacks break through enterprise browser security
• Angry Magpie reveals how fragile the current DLP architecture is in a browser-first world

A newly uncovered data exfiltration technique known as Data Splicing Attacks could place thousands of businesses worldwide at significant risk, bypassing all leading data loss prevention (DLP) tools.

Attackers can split, encrypt, or encode data within the browser, transforming files into fragments that evade the detection logic used by both endpoint protection platforms (EPP) and network-based tools - before these pieces are then reassembled outside the protected environment.

By using alternative communication channels such as gRPC and WebRTC, or secure messaging platforms like WhatsApp and Telegram, threat actors can further obscure their tracks and avoid SSL-based inspections.

Threat actors now splice, encrypt, and vanish

The growing reliance on browsers as primary work tools has increased exposure. With more than 60% of enterprise data stored on cloud platforms accessed via browsers, the importance of a secure browser has never been greater.

Researchers demonstrated that proxy solutions used in many secure enterprise browsers simply cannot access the necessary context to recognize these attacks because they lack visibility into user interactions, DOM changes, and browser context.

Additionally, endpoint DLP systems struggle because they rely on APIs exposed by the browser, which do not offer identity context, extension awareness, or control over encrypted content.

These limitations create a blind spot that attackers can exploit without detection, undermining many enterprises’ ability to defend against insider threat scenarios.

What makes this discovery even more urgent is the ease with which these techniques can be adapted or modified. With new code, attackers can easily create variants, further widening the gap between evolving threats and outdated protections.

In response, the team introduced Angry Magpie, an open source toolkit designed to replicate these attacks. Security teams, red teams, and vendors can use the tool to evaluate their defenses.

Angry Magpie allows defenders to assess their systems’ exposure in realistic scenarios, helping identify blind spots in current implementations of even the best DLP solutions.

“We hope our research will serve as a call to action to acknowledge the significant risks browsers pose for data loss,” the team said.

You may also like

• These are the best VPN with antivirus solutions
• Take a look at our pick of the best password managers
• Google blocked over 5 billion ads in 2024 as AI-powered scams skyrocketed

Whether you're a new resident or unsatisfied with your current internet service, you'll want to check out the best ISPs in Winston-Salem.

Hints and answers for the NYT Connections: Sports Edition puzzle, No. 218, for April 29.

Here are hints -- and the answer -- for today's Wordle No. 1,410 for April 29.

Hints and answers for Connections for April 29, #688.

Here are hints and answers for the NYT Strands puzzle No. 422 for April 29.

• Google DeepMind has enhanced and expanded access to its Music AI Sandbox
• The Sandbox now includes the Lyria 2 model and RealTime features to generate, extend, and edit music
• The music is watermarked with SynthID

Google DeepMind has brought some new and improved sounds to its Music AI Sandbox, which, despite sand being notoriously bad for musical instruments, is where Google hosts experimental tools for laying down tracks with the aid of AI models. The Sandbox now offers the new Lyria 2 AI model and the Lyria RealTime AI musical production tools.

Google has pitched the Music AI Sandbox as a way to spark ideas, generate soundscapes, and maybe help you finally finish that half-written verse you’ve been avoiding looking at all year. The Sandbox is aimed mainly at professional musical artists and producers, and access has been pretty restricted since its 2023 debut. But, Google is now opening up the platform to many more people in music production, including those looking to create soundtracks for films and games.

The new Lyria 2 AI music model is the rhythm section underlying the new Sandbox. The model is trained to produce high-fidelity audio outputs, with detailed and intricate compositions across any genre, from shoegaze to synthpop to whatever weird lo-fi banjo-core hybrid you’re cooking up in your bedroom studio.

The Lyria RealTime feature puts the AI's creation in a virtual studio that you can jam with. You can sit at your keyboard, and Lyria RealTime will help you mix ambient house beats with classic funk, performing and tweaking its sound on the fly.

Virtual music studio

The Sandbox offers three main tools for producing the tunes. Create, seen above, lets you describe the kind of sound you're aiming for in words. Then the AI whips up music samples you can use as jumping-off points. If you've already got a rough idea down but can’t figure out what happens after the second chorus, you can upload what you have and let the Extend feature come up with ways to continue the piece in the same style.

The third feature is called Edit, which, as the name suggests, remakes the music in a new style. You can ask for your tune to be reimagined in a different mood or genre, either by messing with the digital control board or through text prompts. For instance, you could ask for something as basic as "Turn this into a ballad," or something more complex like, "Make this sadder but still danceable," or see how weird you can get by asking the AI to "Score this EDM drop like it's all just an oboe section." You can hear an example below created by Isabella Kensington.

AI singalong

Everything generated by Lyria 2 and RealTime is watermarked using Google's SynthID technology. That means the AI-generated tracks can be identified even if someone tries to pass them off as the next lost Frank Ocean demo. It’s a smart move in an industry that’s already gearing up for heated debates about what counts as "real" music and what doesn’t.

These philosophical questions also decide the destination of a lot of money, so it's more than just abstract discussions about how to define creativity at stake. But, as with AI tools for producing text, images, and video, this isn't the death knell of traditional songwriting. Nor is it a magic source of the next chart-topping hit. AI could make a half-baked hum fall flat if poorly used. Happily, plenty of musical talents understand what AI can do, and what it can't, as Sidecar Tommy demonstrates below.

You might also like

• Is AI bad for music or is it just another step in the auto-tune timeline?
• I tried using AI to create the background music for a podcast, but I may stick to music libraries for now
• This high-tech piano wants to teach you to play with the power of AI

This is a direct shot at Google's own Search-powered shopping tools.

We've tested a range of gaming monitors to find the best one for you.