Mediaboss Marketing

Microsoft's just-announced classification of "Copilot+ PCs" leaves a lot of users out in the cold. To access a suite of new AI features in an upcoming build of Windows 11, you'll need a processor with a Neural Processing Unit (NPU) that capable of hitting 40 TOPS (Trillion Operations Per Second). To date, there's only one processor family that can hit that number: Qualcomm's upcoming Snapdragon X series of mobile chips. And the first laptops with Snapdragon X aren't even due to ship for a few weeks.

Anyone who currently owns a laptop or desktop even if it has one of the best CPUs is out of luck and won't have access to these features. Laptops based on Intel's next-gen Lunar Lake CPUs will ship in Q3 and will meet the 40+ TOPS requirement, but any computer you currently own, or buy today, will not.

So, how bad should you feel about being left out in the cold? If Microsoft's Copilot+ PC press event this week is any indication, not very bad at all. Of the four exclusive AI features Microsoft showed, three are either available elsewhere or are so niche that few people will use them. Only one, Recall, offers something PC users haven't seen before and it has some very creepy implications for your privacy.

Before we talk about Recall, why it's useful, and why it might also be a security nightmare, let's spend a moment on why you shouldn't care about the other Copilot+ features. They are:

Recall, on the other hand, offers a feature that you can't get in Windows right now. When enabled, it takes a screenshot, called a "snap," every few seconds of your entire desktop. You can then open Recall and query the content of your images or scroll through your timeline to remind you of what you were doing.

So, for example, if you were doing some online shopping a few days ago, you can search Recall for "red shoes" and it will show you all the snapshots of the moments where you were looking at red shoes. If you were on a website, you'll also get a link under the snapshot, that you can click to go back to the page you were on. If the result was in a Word document or a PowerPoint presentation, Recall will open the appropriate file for you and even take you straight to the slide (in the case of PowerPoint) with the data on it.

If you ask about a presentation or a spreadsheet you were working on, Recall will take you back to the snapshot you were working on and you can even copy text directly from the snapshots (essentially, image-to-text). Since it has screenshots of all your online conversations, you can also ask it "what did grandma say" and it will show you screenshots of your conversations with her.

Get Tom's Hardware's best news and in-depth reviews, straight to your inbox.

Senior Editor Brandon Hill got a chance to see Recall in action at Microsoft's press event. He saw someone search for "papaya salad" and get a snapshot of a web page in the timeline from when the user had visited a recipe website. He saw someone query "can you find my water report" and Recall showed a mix of web pages, a PDF, and an Excel document that all had water reports on them.

Image 1 of 4

He also shot the video below. In it, you can see a Microsoft rep scroll through Recall's timeline and then ask it for "offer" which pulls up images of offers.

Google Photos and other online services have a lot of the same image-recognition capabilities as Recall. I went through my Google Photos album and asked for a variety of things, the online tool did a great job of locating photos I had never tagged. When I asked it for "Halloween," it accurately pulled up pictures of me and my kids in costume. And when I asked it for "Maya," it pulled up pictures of my daughter from her birthday because she was wearing a crown with her name on it and it read the text.

Recall combines the ability to index photos based on content with an archive of constant screenshots of your activity. And it adds in image-to-text and image-to-web-page-URL capabilities. It's just speculation on my part, but perhaps a future version of the tool will integrate with other parts of your OS; I could imagine it adding something to your calendar if you tell a friend in a chat message, "I'll meet you at 7:30."

For Mac and iOS users, there's a third-party app called Rewind that does what Recall does and it also includes audio recognition. But we've never seen anything like Recall for PCs until now.

The question is this: How comfortable do you feel about having an app that tracks your every activity, keeping an image record of it? And do you really want (or need) a tool to ask about... your own activities?

Microsoft is taking some steps to make sure that the snaps Recall takes stay private. According to the company, the snaps remain encrypted on your local storage drive and are never synced to the cloud or sent to Microsoft. You also have the ability to exclude apps from Recall, delete individual snaps, or disable the feature entirely. It's easy to see why Recall would need a powerful NPU to search through all of those images without getting any help from a server on the Internet.

But let's be clear: Recall poses some serious privacy risks even if it works as advertised. You're creating photographic records of all your activities including every time you type in a password that isn't hidden by **** symbols. This also applies also to sensitive personal data, such as your social security number or bank account number. If someone found a way to log into your computer either in-person or remotely they could easily find this treasure trove of important information. And if you share a family computer with the kids or a spouse and they don't log in under their own accounts (which they should) they can also see your snaps.

Whether you feel comfortable with the privacy risks of Recall, you have to ask: Is Recall a game-changing feature that would make you buy a new PC or, conversely, not buy a PC that doesn't support it? My guess is that, for most people, the answer is "no."

If the answer is "yes," you'll need to either buy one of the handful of Snapdragon X-powered laptops coming out in June or wait until the fall, when Intel (and possibly AMD) will have mobile chips with similarly-powerful NPUs. If you want a desktop PC of any kind, get ready to wait we don't expect desktop chips with NPUs until at least Q4, when Intel Arrow Lake processors come out.

Read more:

Total Recall: the only Copilot+ AI feature that matters is a huge privacy risk - Tom's Hardware

Government officials and AI industry executives agreed on Tuesday to apply elementary safety measures in the fast-moving field and establish an international safety research network.

Nearly six months after the inaugural global summit on AI safety at Bletchley Park in England, Britain and South Korea are hosting the AI safety summit this week in Seoul.The gathering underscores thenew challenges and opportunities the world faceswith the advent of AI technology.

The British government announced on Tuesdaya new agreementbetween 10 countries and the European Union to establish an international networksimilar to the U.K.s AI Safety Institute, which is the worlds first publicly backed organization, to accelerate the advancement of AI safety science. The network will promote a common understanding of AI safety and align its work with research, standards, and testing. Australia, Canada, the EU, France, Germany, Italy, Japan, Singapore, South Korea, the U.K., and the U.S. have signed the agreement.

On the first day of the AI Summit in Seoul, global leaders and leading AI companies convened for a virtual meeting chaired by U.K. prime minister Rishi Sunak and South Korean president Yoon Suk Yeol to discuss AI safety, innovationandinclusion.

During the discussions, the leaders agreed to the broader Seoul Declaration, emphasizing increased international collaboration in building AI to address major global issues, uphold human rights, and bridge digital gaps worldwide, all while prioritizing being human-centric, trustworthy, and responsible.

AI is a hugelyexciting technology and the U.K. has led global efforts to deal with its potential, hosting the worlds first AI Safety Summit last year, Sunak said in a U.K. government statement. But to get the upside, we must ensure its safe. Thats why Im delighted we have got an agreement today for a network of AI Safety Institutes.

Just last month, the U.K. and the U.S.sealed a partnership memorandum of understandingto collaborate on research, safety evaluation, and guidance on AI safety.

The agreement announced today followsthe worlds first AI Safety Commitments from 16 companies involved in AI, including Amazon, Anthropic, Cohere, Google, IBM, Inflection AI, Meta, Microsoft, Mistral AI, Open AI, Samsung Electronics, Technology Innovation Institute, xAi and Zhipu.ai. (Zhipu.ai is a Chinese company backed by Alibaba, Ant and Tencent.)

The AI companies, including those from the U.S., China, and the United Arab Emirates (UAE), have agreed to the safety commitments to not develop or deploy a model or system at all if mitigations cannot keep risks below the thresholds, according to the U.K. government statement.

Its a world first to have so many leadingAIcompanies from so many different parts of the globe all agreeing to the same commitments onAIsafety, Sunak said. These commitments ensure the worlds leadingAIcompanies will provide transparency and accountability on their plans to develop safeAI.

Continued here:

In Seoul summit, heads of states and companies commit to AI safety - TechCrunch

When Andrew Ferguson launched Databricks corporate venture arm three years ago, he wasnt expecting to be writing checks quite this fast.

Databricks Ventures has backed 25 companies already. In 2023, the fund closed an investment every month, on average, Ferguson tells Term Sheet.

Of course, Ferguson also wasnt anticipating the frenzied rush of new entrants and new companies into the AI sector as soon as GPT hit the market at the end of 2022nor the soaring valuations that LLM competitors would notch so shortly after. Databricks, for its part, introduced its own open-source large language model earlier this year, called DBRX. And it started making investments into other models (Mistral AI for one), and in companies that make use of multiple models, like Perplexity AI, this past fall.

Now Ferguson wants to formalize Databricks Venturess investment strategy accordingly. Its initial fund, launched in 2021, was squarely focused on companies that worked with the Databricks Lakehouse architecturewhich removes traditional data silos and allows teams to pull from a single data source. Databricks new venture capital fund, which officially launches today, will focus on a broader subset of companies that are sitting on top or working along with the Databricks Data Intelligence Platform, which DBRX was built on top of.

Databricks second VC fund isnt a specific size. The fund invests off the corporate balance sheet on a deal-by-deal basis (Databricks declined to disclose how much capital the company has invested in the 25 startups it has backed). Theres no specific deployment period, and Databricks CEO Ali Ghodsi signs off on every investment. So when Databricks says it is launching a second fund, you can think of it more as the company debuting a new thematic strategy. Like many corporate VCs, Databricks is on the hunt for strategic partnerships and ways to improve its customers product experience, rather than just a financial return. And it naturally will end up striking an M&A deal or two for portfolio companies that end up fitting into the companys broader strategy, though thats not the purpose of the fund, according to Ferguson. (Databricks ended up acquiring Arcion, a data ingestion company that Databricks invested in back in 2022, for example)

This strategic approach makes Ferguson a bit more comfortable coming in at some of the soaring valuations currently garnered by AI startups with comparatively slim revenuevaluations that Ferguson says make him uncomfortable and nervous and that he acknowledges dont pencil out, at least for now.

Because these startups will be adding value to pre-existing Databricks customers, or incentivizing new users to use Databricks, theres other financial value in it for Databricks versus just a big IPO or M&A exit. Thats valuable in a different way than when you have to think about the financial valuation at the time of our investment, Ferguson says.

Databricks tends to come in at the Series A or Series B stages, when a company already has a product in the market. And it doesnt ever lead deals, instead following on in rounds led by VCs, including some of its own investors such as Andreessen Horowitz or NEA, versus the other way around. Ferguson says that investors like to bring him into a round because of how Databricks can help with due diligence, given that they have either worked directly with some of the companies already or have direct access to their customer base to ask for feedback.

That came into play with Databricks investment into Unstructured. We were able to validate the quality of the product offering by talking to our own customers, Ferguson says. Alternatively, he also has passed on deals because Databricks employees didnt personally have a good experience working with a company, or got mixed feedback from one of their customers.

In a nascent and rapidly evolving ecosystem of AI providers, Ferguson wants to have a finger on the pulse of the latest trends; he needs to ensure that Databricks can help create the ecosystem itself, and partner with the companies who are on the front end of sectors that arent even top of mind yet, because the industry is moving so quickly.

The use cases keep changing. The development patterns for AI keep changing. And so we want to make sure that venture is a lever we can use to help create the ecosystem of partners that our customers need, Ferguson says.

Allie helped report this story last week, and I wrote this essay en route from her wedding in Malibu this weekend. I can confirm that vows were exchanged, that two former Term Sheet writers were in attendance, and that Ive never seen her happier. Mazel Tov, Allie! And I hope you enjoy hearing from some of our other reporters on staff (including me!) while shes honeymooning.

See you tomorrow,

Jessica Mathews Twitter: @jessicakmathews Email: jessica.mathews@fortune.com Submit a deal for the Term Sheet newsletter here.

Joe Abrams curated the deals section of todays newsletter.

VENTURE DEALS

- Alicorn Venture Partners agreed to acquire Glassbox, a London, U.K.-based customer intelligence platform, for approximately $150 million.

- KETOS, a Milpitas, Calif. and New Braunfels, Texas-based developer of technology designed to measure, manage, and forecast water quality and efficiency for industrial, agricultural, and municipal applications, raised $10 million in funding. Tenfore Holdings led the round and was joined by Danu Venture Group.

- Overland AI, a Seattle, Wash.-based developer of autonomous off-road mobility technology for vehicles used in the defense sector, raised $10 million in seed funding. Point72 Ventures led the round and was joined by Shasta Ventures, Ascend VC, Pioneer Square Labs, Voyager Capital, and Cubit Capital.

- Teal, a developer of accounting infrastructure for vertical SaaS businesses, or businesses that develop software for a specific industry, raised $8 million in seed funding. Torch Capital led the round and was joined by Basis Set Ventures, General Advance, Dash Fund, and angel investors.

- Tenon, an Indianapolis, Ind.-based work management platform for marketing teams, raised $8 million in Series A funding. High Alpha and StepStone Group led the round and was joined by ServiceNow Ventures.

- Limula, a Lausanne, Switzerland-based developer of a single device designed for on-demand and at-scale manufacturing of cell therapies, raised $6.8 million in seed funding. LifeX Ventures led the round and was joined by Verve Ventures, Zhlke Ventures, Oxford Seed Fund, and others.

- 7Analytics, a Bergen, Norway-based developer of AI and machine learning models designed to predict floods and landslides, raised 4 million ($4.3 million) in funding from Scale Capital.

- Haz, a London, U.K.-based app designed to show users the products that their friends own and notify them when their friends buy or sell anything online, raised $1.4 million in pre-seed funding. Speedinvest led the round and was joined by a16z, Atomico, Concept Ventures, and others.

PRIVATE EQUITY

- Centivo, backed by HarbourVest Partners, acquired Eden Health, a New York City-based medical care platform for employees. Financial terms were not disclosed.

- Heartland Paving Partners, backed by Soundcore, acquired Klekamp & Company, a Cincinnati, Ohio-based paving contractor. Financial terms were not disclosed.

- Juniper Landscaping, a portfolio company of Bregal Partners, acquired Davis Landscaping, a Harrisburg, Pa. and Raleigh, N.C.-based landscaping maintenance and installation provider. Financial terms were not disclosed.

- Sports Endeavors, a portfolio company of Seawall Capital, acquired Ewing Sports, a Ewing Township, N.J.-based supplier of customized soccer apparel. Financial terms were not disclosed.

EXITS

- CyberArk (NASDAQ: CYBR) agreed to acquire Venafi, a Salt Lake City, Utah-based machine identity management platform, from Thoma Bravo, for $1.5 billion.

OTHER

- Bumble (NASDAQ: BMBL) agreed to acquire Geneva, a New York City-based platform for meeting people with shared interests. Financial terms were not disclosed.

FUNDS + FUNDS OF FUNDS

- ETF Partners, a London, U.K.-based venture capital firm, raised 285 million ($310 million) for its fourth fund focused on companies with a positive environmental impact.

PEOPLE

- GV (Google Ventures), a San Francisco-based venture capital firm, hired Elena Sakach as an investment partner. Formerly, she was with Coatue.

- Pillar VC, a Boston, Mass.-based venture capital firm, hired Liz Kettler as chief financial officer. Formerly, she was with OpenView.

Read more here:

Databricks is expanding the scope of its AI investments with second VC fund - Fortune

Photo bombing is dead. Adobe is adding an artificial-intelligence-powered Generative Remove feature to its Lightroom photo editor that makes it dead simple to zap out unwanted elements, like that annoying guy in the background. The new feature is in a public beta-testing phase, but it will work across the Lightroom ecosystem whether you're using the app on mobile, desktop, or web.

Lightroom's Generative Remove uses Adobe's Firefly AI engine to smoothly replace unwanted elements. Simply paint over the area you want to remove and Lightroom will send that information to Adobe's Firefly servers, which then crunch the data and send it back. In demos WIRED saw, this process took no more than a few seconds, though performance will depend on your internet connection's speed.

Unlike Adobe Photoshop's Reference Image feature, which launched less than a month ago and allows users to generate new images using Firefly, Lightroom's AI features are very much focused on a photographer's workflow.

The highlighted area shows what will be removed.

You can use Object Aware and Generative AI together.

One of the more difficult things to do when editing images is to remove distracting elements. Typically this would be done using tools like Lightroom's Content Aware Remove, which hides elements by matching surrounding areas. This works well in small situations where backgrounds aren't too confusing for the software. For example, removing a telephone pole against a solid blue sky. But the larger the object to remove, and the more complex the background, the more difficult and time-consuming this becomes.

The Firefly-powered Generative Remove can do the same thing but for much larger objects against any background. Adobe has reduced what would have once taken hours and considerable technical know-how to the flick of a mouse and a few seconds of processing time. Everyone is now a Lightroom wizard. Also, unlike other retouching tools, which do the best match they can, Generative Remove generates three different versions and allows you to choose the one that looks best.

As impressive and useful as Generative Remove is, it might sound a bit familiar, especially to anyone using Google Photos. These new features don't offer much that Google's Magic Eraser tool couldn't already do. Nor does it enable anything like Google's Magic Editor, which lets you alter the lighting of a scene or cut and paste subjects within the scene.

Adobe's Generative Remove mirrors the company's previous uses of AI, like last year's AI-powered noise removal tool, which built on existing noise removal tools, making them better rather than breaking significant new ground. This, I suspect, is what working photographers actually wantbetter tools, rather than flashy new features. Adobe seems content to leave the more dramatic AI-powered tools, like rearranging a scene after the fact, to others.

See the article here:

Adobe's AI-Powered Generative Remove Feature in Lightroom Erases Unsightly Objects in Seconds - WIRED

Last summer, as they drove to a doctors appointment near their home in Manhattan, Paul Skye Lehrman and Linnea Sage listened to a podcast about the rise of artificial intelligence and the threat it posed to the livelihoods of writers, actors and other entertainment professionals.

The topic was particularly important to the young married couple. They made their living as voice actors, and A.I. technologies were beginning to generate voices that sounded like the real thing.

But the podcast had an unexpected twist. To underline the threat from A.I., the host conducted a lengthy interview with a talking chatbot named Poe. It sounded just like Mr. Lehrman.

He was interviewing my voice about the dangers of A.I. and the harms it might have on the entertainment industry, Mr. Lehrman said. We pulled the car over and sat there in absolute disbelief, trying to figure out what just happened and what we should do.

Link:

Voice Actors Sue Company Whose AI Sounds Like Them - The New York Times