Category: 4. Technology

It looks like Meta’s hiring freeze definitely does not apply to Apple engineers. As reported by Bloomberg, Jian Zhang, Apple’s Lead AI Researcher for Robotics, will be joining the Meta Robotics Studio. And he’s not the only one leaving. Here are the details.

Apple’s AI alternatives (and teams) keep getting slimmer

Last month, following a report that said that, despite Meta’s hiring freeze, there could be exceptions, Bloomberg reported that the company had poached Frank Chu, “who has led Apple AI teams focused on cloud infrastructure, training and search.”

Now, the tally just got a bit higher, as Bloomberg reported that Meta has also hired away Jian Zhang, Apple’s Lead AI Researcher for Robotics:

“Zhang, the robotics researcher, led a small team of academics focused on automation technology and the role of AI in such products. That team has already suffered some turnover, with one of Zhang’s reports, Mario Srouji, leaving to run AI products at Archer Aviation Inc. in April.”

Zhang’s departure coincides with Apple’s reported loss of three more Foundation Models engineers: John Peebles and Nan Du to OpenAI, and Zhao Meng to Anthropic.

Here’s Bloomberg on the new departures:

“They were all part of an Apple Foundation Models team that has lost roughly 10 members, including its chief, in recent weeks. The group was central to the creation of the Apple Intelligence platform, launched last year as part of the company’s bid to catch up in AI.”

The news comes on the heels of another Bloomberg report that said that “Apple is now discussing internally whether to rely more on outside technology, rather than just homegrown models.”

Apple is also still considering a splashy acquisition (with names like Mistral and Perplexity in the mix) while it also explores partnerships with OpenAI, Anthropic, and even Google to develop the models needed for the revamped, AI-driven Siri it unveiled at WWDC24 but never brought to market.

If you were Tim Cook, how would you stop Apple’s brain drain? Let us know in the comments.

Accessory deals on Amazon

Amazon is further investing in AI-powered shopping experiences with Tuesday’s launch of Lens Live, a new AI-powered upgrade to its Amazon Lens shopping feature that allows consumers to discover new products through visual search, similar to competitors like Google Lens and Pinterest Lens. The tool will also integrate with Amazon’s AI shopping assistant, Rufus, for product insights, the retailer notes.

Lens Live will not replace Amazon’s existing visual search tool, Amazon Lens, which lets you take a picture, upload an image, or scan a barcode to discover products. Instead, it brings a real-time component to Amazon Lens so you can point your phone at things you’re seeing in the real world to see matching products in a swipeable carousel at the bottom of the screen.

The addition is one of several ways Amazon has been leveraging AI to help online shoppers. Over the past year or so, the company has also rolled out other features like its AI assistant Rufus, AI-powered shopping guides, AI-enhanced product reviews, AI tools for finding clothes that fit, AI audio product summaries, personalized shopping prompts, as well as tools for merchants.

Lens Live also capitalizes on activities customers are already doing: comparison shopping while in retail stores out in the real world to see if Amazon has a better deal on the same or similar item.

When using the new Lens Live feature, customers can tap on any item in their camera view to trigger the feature to focus on that product. If they find a match they like, they can add it to their shopping cart by tapping the (+) plus icon or tap the heart icon to save it to their wish list.

The feature is powered by Amazon SageMaker services, which allow machine learning models to be deployed at scale. It runs on AWS-managed Amazon OpenSearch.

In addition, Amazon’s AI-powered shopping assistant Rufus is available in the new experience, allowing customers to see AI-generated product summaries and suggested questions of conversational prompts they can ask to learn more about the item. According to Amazon, this lets shoppers do some quick product research and view product insights before making a purchase.

The Lens Live feature is first launching on the Amazon Shopping app on iOS, initially for “tens of millions” of U.S. shoppers before rolling out to others in the U.S. The company didn’t say whether it’s going to expand to other global markets.

Apple is preparing to enter the home security camera market with a new lineup of products designed to compete directly with Amazon’s Ring and Google’s Nest, according to Bloomberg.

The recent report said the company is developing a battery-powered camera code-named “J450” that could last from several months to a year on a single charge. The devices are expected to include facial recognition and infrared sensors to detect who is in a room, potentially enabling tasks such as turning lights off when someone leaves or playing music preferred by a particular family member.

Bloomberg also reported that Apple has tested a doorbell with facial recognition technology capable of unlocking a door, and that the company is planning multiple types of cameras and related products as part of an entirely new hardware and software lineup. These efforts would build on Apple’s existing iCloud+ subscription services, which already provide online storage for video footage from third-party cameras.

A potentially disruptive move

Jennifer Kent, vice president of research at Parks Associates, told SecurityInfoWatch that Apple’s entrance could be significant given the company’s brand loyalty and installed base of devices.

“Apple’s entry into smart home security would be highly significant due to its loyal, affluent base and ecosystem strength,” she said. “Unlike Amazon and Google’s household focus, Apple’s individual-first strategy could carve out a premium, privacy-centric niche. This would likely disrupt DIY competitors first, while professional security providers would feel indirect pressure to differentiate on services and monitoring rather than compete head-to-head.”

According to Parks Associates’ “Home Security Consumer Insights” report, 47% of U.S. internet households have a security solution, either a system or device. Among the 35% who pay for a security service, one in five (21%) are paying for self-monitoring of video devices. Smart video devices are now in 33% of households, a trend Kent said is being driven by affordability, wireless flexibility, and growing safety concerns.

Paid services remain central to the market: 78% of security system owners pay for some form of service, whether professional monitoring, self-monitoring, or video storage. Roughly 19% of all internet households have professionally monitored systems, while 7% pay for non-professional services like alerts and video storage.

Ecosystem advantages

Kent pointed to Apple’s reach and brand loyalty as differentiators. Parks Associates research shows that 61% of U.S. internet households own at least one Apple device, and one-quarter qualify as “Apple loyalists” who own three or more.

“Given this base, Apple’s expansion into smart cameras or automation would be significant, because it could integrate with products consumers already trust and use daily,” she said.

By the end of the year, NX X Manufacturing users will have access to an AI assistant to help them quickly program parts.

Welcome to Engineering Paper. Here’s all the design and simulation software news you missed last week.

… Is what I’d normally say, but software news has been slow this Labor Day pre-kend, weekend, and week-in. Which gives me a perfect opportunity to cover something I’ve been meaning to write about for months, and you’ve been meaning to read about for just as long (though you didn’t know it).

AI is coming to Siemens NX X Manufacturing (again)

Back in July I was in Detroit for Realize Live 2025, Siemens’ annual user conference. I covered the highlights at the time, but I didn’t get around to an interview that I had with Siemens’ Michael Taesch, senior director of product management, and Sashko Kurciski, marketing director for digital manufacturing, about AI for NX CAM.

“We’re trying to see how we fuse AI into our manufacturing process,” Taesch told me.

One result of that fusion is an upcoming AI tool that generates machining strategies for NX users. It’s not available yet and it doesn’t have an official name, but for the sake of conversation Taesch and Kurciski called it the CAM Copilot within NX X Manufacturing. For the sake of this article, I’m going to call it NX CAM Copilot.

With a couple clicks, NX CAM Copilot—not its official name—will generate three possible ways to machine a given feature (and more if needed). The user can take these as a starting point to develop their program. The AI will debut with 2.5 and 3-axis machining strategies and expand from there.

“We’re not here to replace the manufacturing engineer,” Taesch said. “We want to give him multiple processes, and it’s up to him to pick the right one based on his knowledge.”

With a familiar thumbs up/down system, users will be able to rate the suggestions and, in theory, the AI will learn their preferences and improve over time.

“Manufacturing is complex,” Kurciski said. “You can machine a single feature in—I’m not exaggerating—10 different ways with 15 different tools. So you have your own company best practices, and capturing this is very important.”

In that sense, Siemens sees NX CAM Copilot—again, not its official name—as a tool for knowledge capture, one that can help bridge that engineering skills gap that technology vendors keep talking about.

“Company A will program their way. Company B will have a completely different process,” Taesch added. “You cannot have a generic solution. You need to have something that’s tailored, personalized to the customers.”

But the more direct utility of NX CAM Copilot is in its efficiency. Like anything, CAM programming takes time, even for the experts.

“Though I’ve been 10 years in NX CAM, I always make mistakes when I start the programming,” Taesch said. “Here, in a couple clicks, I get a result, and then all I have to do is just fine tune, use my knowledge to adjust a little bit.”

Taesch estimates that NX CAM Copilot can save 80% or more of the time that users would otherwise spend on CAM programming. “I can quickly program a part. Might not be perfect, but if I wanted to quickly quote it, in five minutes I can get something up and running,” he said.

NX CAM Copilot is currently in beta, but Taesch said Siemens plans to release it by the end of 2025. Taesch expects the AI tool to be an add-on to NX X Manufacturing through Siemens’ value-based licensing program, though that isn’t a final decision. (There’s an extant NX X Manufacturing copilot, which is just a chatbot, that’s currently available as a value-based add-on. Perhaps this new tool will be integrated into that.)

NX CAM Copilot, which is a placeholder name, reminds me of Toolpath, which sounds like a placeholder name but isn’t. Toolpath is a web-based platform that uses AI to generate toolpaths (and as I covered last week, it just announced Autodesk as an investor). I brought up the comparison to Taesch and Kurciski, who offered the opinion that Toolpath is more of a black box solution focused on quoting, while NX CAM Copilot provides more tailored manufacturing choices.

“We have another project that is working in parallel where we automatically ingest for quoting,” Taesch added. “It’s going to be tied into NX CAM. And the idea will be to leverage this tool to provide some simple quotes. Not there yet, but we have it in mind.”

So there you have it—an AI engineering tool from Siemens that’s a lot more interesting than Design Copilot NX. I’ll bring you more on NX CAM Copilot, or whatever it will be called, as soon as I learn it (or several months later, whichever comes first).

One last link

Ever wonder what’s inside a 1950s Heathkit vacuum tube oscilloscope? EE World’s Martin Rowe reveals all in his latest teardown—but he needs your help to put it all together.

Got news, tips, comments, or complaints? Send them my way: malba@wtwhmedia.com.

Following the February 2024 FDA approval of irinotecan liposome (Onivyde) with oxaliplatin, fluorouracil, and leucovorin (NALIRIFOX), patients with metastatic pancreatic adenocarcinoma have gained a much-needed additional treatment option in the frontline setting; however, in the absence of head-to-head data, patient- and disease-related factors must drive treatment decisions when considering NALIRIFOX or FOLFIRINOX (fluorouracil, oxaliplatin, and irinotecan), according to Raji Shameem, MD, a medical oncologist and hematologist at the Orlando Health Cancer Institute in Florida.¹

At a median follow-up of 16.1 months (IQR, 13.4-19.1), findings from the phase 3 NAPOLI 3 trial (NCT04083235), which supported the 2024 regulatory decision, demonstrated that patients who received NALIRIFOX (n = 383) achieved a median overall survival of 11.1 months (95% CI, 10.0-12.1) vs 9.2 months (95% CI, 8.3-10.6) among patients who received nab-paclitaxel (Abraxane) and gemcitabine (n = 387; HR, 0.84; 95% CI, 0.71, 0.99; P = .0403).² The median progression-free survival was 7.4 months (95% CI: 6.0, 7.7) vs 5.6 months (95% CI, 5.3, 5.8), respectively (HR, 0.70; 95% CI, 0.59-0.85; P = .0001).

In an interview with OncLive®, Shameem discussed evolving frontline treatment options for metastatic pancreatic cancer, the effect that approval of NALIRIFOX has had on the frontline treatment paradigm in metastatic pancreatic cancer, and how he selects a frontline regimen for patients in this setting.

OncLive: What is the current frontline treatment landscape for patients with metastatic pancreatic cancer?

Shameem: If you look at the National Comprehensive Cancer Network [NCCN] Guidelines, as many community oncologists do on a daily basis, it’s sometimes hard to keep up to date with the new treatment options. For [patients with] pancreatic cancer, it had been more than a decade since we’ve gotten a new frontline therapy option.

NALIRIFOX is 1 of the new treatment options in the frontline setting. Prior to this, options included modified FOLFIRINOX and FOLFIRINOX, based on an important trial in 2011, the [phase 3] PRODIGE trial [NCT01526135], as well as gemcitabine plus nab-paclitaxel [Abraxane], based on the phase 3 IMPACT trial [NCT00844649].

When making [a frontline therapy] decision, I’d say every patient is different. Performance status is very important, and the NCCN guidelines highlight this. If a patient has a good performance status—ECOG 0 to 1—options would include a 3-drug regimen. For example, NALIRIFOX is a category 1 preferred option in this setting.

Since its FDA approval, how has NALIRIFOX been integrated into clinical practice?

[Clincians] are creatures of habit, and once we get used to something, we tend to stick with it. Our go-to regimen is FOLFIRINOX. [However], toxicity with conventional, run-of-the-mill FOLFIRINOX is tough, so we [often] use modified FOLFIRINOX.

It’s important to note that everyone’s modified FOLFIRINOX is different. You look at different institutions—even among the same providers—it’s different. In my practice, we have our own modified FOLFIRINOX.

However, NALIRIFOX was a standardized regimen, [which was] evaluated NAPOLI 3, and the dosing was the same for every patient randomly assigned to that group. For NALIRIFOX, based on the trial’s dosing, efficacy, and safety, I’ve used it as an option for frontline therapy in my patients with metastatic pancreatic cancer.

What did it mean to have head-to-head data showing NALIRIFOX was superior to a long-standing standard of care in gemcitabine plus nab-paclitaxel?

That’s an important question. Prior to NALIRIFOX, we always assumed that 3 drugs were better than 2. However, in oncology—as in many fields—you shouldn’t make assumptions.

NAPOLI 3 compared infusional 5-fluorouracil, oxaliplatin at 60 mg/m², and liposomal irinotecan at 50 mg/m², vs gemcitabine plus nab-paclitaxel at standard dosing, just as it was administered in IMPACT.

Gemcitabine plus nab-paclitaxel was a standard frontline option, so this was truly a head-to-head trial between 2 frontline regimens for our [patients with] metastatic pancreatic cancer.

Since there hasn’t been a head-to-head comparison between NALIRIFOX and FOLFIRINOX, how do you choose between them in clinical practice?

One thing I do is look closely at the trial protocols, [including] how dose reductions were handled and when providers made those decisions. When I consider dose reductions, it gives me comfort to know I’m doing what the trial did rather than making it up as I go. That’s an important consideration.

I believe [NALIRIFOX and FOLFIRINOX] are reasonable for patients with good performance status. If a patient has an ECOG of 2 or more, a 3-drug regimen might lead to more toxicity than benefit.

It’s also important to note that when PRODIGE was conducted, there was an age limit—every patient was under 76 years old. In NAPOLI 3, there was no age limit. I believe the oldest patient was 85, and approximately 6% of patients were 75 years or older. That’s important because I practice in Florida and have an older population. I always say age is just a number, but it gives me comfort knowing that older patients were included in [NAPOLI 3].

Are there safety profile differences between NALIRIFOX and FOLFIRINOX that influence your treatment decisions?

It’s difficult to tease out differences in toxicity. There hasn’t been a head-to-head trial of NALIRIFOX vs modified FOLFIRINOX, and I see more similarities than differences.

I want to highlight that conventional irinotecan is different from liposomal irinotecan. Preclinical studies have shown that liposomal encapsulation may allow for better drug delivery of the active metabolite to tumor cells.

I’d say there are more similarities, [for example] diarrhea can occur with both [NALIRIFOX and FOLFIRINOX]. The key is knowing when to hold treatment and when to pursue dose reductions. [The findings from] NAPOLI 3 give us clear guidance on when to do that.

From a patient-centered perspective, how do those initial treatment conversations typically go?

That initial consultation with a patient who has been diagnosed with metastatic pancreatic cancer is overwhelming. Multiple family members are often in the room, and you’d be surprised how many patients ask me about NALIRIFOX. A lot of our patients are savvy—they’re looking up the most recent FDA-approved therapies, and I’ve been asked about NALIRIFOX specifically.

I review all available options, but I always make a point to mention the most recent FDA-approved treatment and talk about the NAPOLI 3 trial. A lot of patients want the newest option. I always counsel them on toxicity—no treatment is without toxicity—but a newer therapy can be very appealing.

What are the major unmet needs in first-line treatment, and where do you see the field heading?

Compared with other cancers, pancreatic cancer has far fewer targeted therapies. For all my patients with metastatic pancreatic cancer, I pursue next-generation sequencing [NGS] and germline genetic testing.

It’s important to know that patients [with disease] harboring somatic or germline homologous recombination–deficient alterations tend to respond well to platinum-based therapies. In those cases, I select a platinum regimen.

Typically, NGS results show TP53 and KRAS mutations. Approximately 80% of patients with metastatic pancreatic cancer patients have disease with some type of RAS mutation. [For a long time], those mutations weren’t druggable; now, times have changed and we’re seeing the evolution of RAS inhibitors.

RAS inhibitors were evaluated initially in later-line settings and have shown impressive efficacy in pretreated patients. I foresee that these agents will move into the frontline setting—potentially in combination with chemotherapy. That’s an appealing option for patients who are [eligible to receive] all available therapies.

Many patients with pancreatic cancer don’t make it to second-line therapy—not because they don’t want it, but because the disease progresses so rapidly. When patients deteriorate, they deteriorate quickly. Therefore, combination therapy in the frontline makes a lot of sense.

How do you balance clinical trial participation vs starting an approved regimen?

Ever since I was in training, I always try to look for a clinical trial. Pancreatic cancer is estimated to be the second leading cause of cancer mortality by 2030, so we know that this is an aggressive disease. That’s part of the reason why I do NGS testing. The majority of our patients [cannot receive] targeted therapies, even on a clinical trial, but you never know which one of your patients might benefit from them. The idea of a clinical trial is always something I discuss with my patients, in addition to standard-of-care therapy,

References

1. FDA approves irinotecan liposome for first-line treatment of metastatic pancreatic adenocarcinoma. FDA. February 16, 2024. Accessed August 29, 2025. https://www.fda.gov/drugs/resources-information-approved-drugs/fda-approves-irinotecan-liposome-first-line-treatment-metastatic-pancreatic-adenocarcinoma
2. Wainberg ZA, Melisi D, Macarulla T, et al. NALIRIFOX versus nab-paclitaxel and gemcitabine in treatment-naive patients with metastatic pancreatic ductal adenocarcinoma (NAPOLI 3): a randomised, open-label, phase 3 trial. Lancet. 2023;402(10409):1272-1281. doi:10.1016/S0140-6736(23)01366-1

Apple has a new Vision Pro 2 launching soon, but two of the main complaints about Vision hardware—weight and cost—likely won’t be addressed until 2027’s rumored ‘Vision Air’ ships. Today, a new report outlines how much cheaper and more affordable that Vision Air product might be.

Vision Air expected to be nearly half the weight and less than half the price of Vision Pro

Ming-Chi Kuo writes on his blog:

Apple is expected to launch Vision Air in 2027, which will address the high weight and price of the Vision Pro. The new device is estimated to be over 40% lighter and more than 50% cheaper.

Vision Air shipments are anticipated to reach one million unit in 2027 (vs. existing models with shipments below 400,000 units).

The two key numbers there are:

• over 40% lighter than Vision Pro
• and over 50% cheaper too

Apple Vision Pro weighs about 600g or 1.4lbs. A weight reduction of over 40% should certainly go a long way toward making the Vision Air significantly more comfortable to wear.

As for price, Vision Pro starts at $3,499 and it’s likely that the new Vision Pro 2 will maintain that price.

If Vision Air is more than 50% cheaper, we could see a price point of around $1,500.

That’s still far more expensive from low-cost alternatives like the $500 Meta Quest 3. But it would be much more palatable than $3,499, especially using pay-over-time plans like what Apple Card offers.

The big question, though, is how Apple might cut costs so significantly.

There could perhaps be fewer cameras, lower-res displays, and the removal of EyeSight. But it’s hard to know how much Vision Air can cut while still retaining a Vision Pro-like experience.

Would you buy a $1,500 Vision Air that’s nearly half the weight of Vision Pro? Let us know in the comments.

Best Vision Pro accessories

For anyone heading down the road of serious watch obsession, it won’t be long before you get on first-name terms with the main characters of the luxury watches universe. As you’ll inevitably soon discover, however, Speedies, Subs, Daytonas, Tanks, and Royal Oaks are really just the tip of a very large iceberg.

While these icons are deserving of a place in every top-shelf collection, a well-honed appreciation for the many underrated, undervalued, and otherwise slept-on tickers is the true mark of a connoisseur. We asked one such data-oriented connoisseur, Bezel founder Quaid Walker, to name the tickers he thinks are most likely to become more iconic with age. Here’s what he had to say.

---

If We Had Money to Burn…

Rolex Submariner “Hulk” Ref. 116610LV Watch

Revealed in 2010 and discontinued in 2020, this Sub (nicknamed “Hulk” for obvious reasons) is the only version of the iconic Rolex dive watch to feature a green dial and bezel. It also contains several notable upgrades from the previous Submariner generation, including a ceramic bezel and improved luminescence.

“I think there’s something so whimsical about the Hulk,” says Walker. “It has the heft and the utilitarian vibes associated with the Submariner, but it doesn’t take itself seriously. If you see someone wearing it, it often signals that they have a lot of other really interesting watches in the safe back at home.”

Rolex Daytona Platinum Ref. 126506-0002 Watch

Speaking of IYKYK Rolex references, the Platinum Daytona—which was the first production Rolex model to feature a display caseback—has been catnip for members of the watch Illuminati since its introduction in 2013.

“The ethos of the brand has always been tool watches, so if Rolex continues to march forward with display case backs, then it’s cool to own the first one,” Walker says. “And if they decide that they’re going to move back to a traditional case back, it’s cool to own something from the short period of time when this existed in production.”

F.P. Journe Chronomètre Bleu Watch

Few people personify the current vogue for high-end independent watchmaking like Francois Paul Journe, whose highly sought creations are now vying for records at auction alongside the likes of Patek and Rolex.

If you’re interested in getting on board with F.P. Journe (and have a six-figure budget to throw at one), this piece’s tantalum case and hand-wound movement make it a savvy choice. “The blue dial is very striking and tantalum is a material that is quite rare to see in watchmaking, so it’s one of the more special Journes available right now,” Walker adds.

MB&F Legacy Machine Perpetual Evo Ice Blue Watch

Another mainstay of the high-end indie watch world, MB&F takes a decidedly quirky approach to haute horlogerie, as the domed crystal and exposed movement of this perpetual calendar ably demonstrate.

“There’s such an avant-garde playfulness associated with MB&F’s watches, and the best examples are the perpetual calendars,” Walker says, noting that the Legacy Machine’s movement makes the notoriously finicky perpetual calendar function much easier to use. “MB&F has a very loyal set of collectors right now, and I think we’re going to look ahead a few years and see the brand be much larger than it is today.”

H.Moser & Cie Streamliner Small Seconds 39 Steel Watch

For anyone who appreciates the look of ’70s grails like the AP Royal Oak and Patek Philippe Nautilus, the Streamliner offers a more minimalist take on the genre from Switzerland’s most irreverent up-and-coming watch brand, Walker says.

Here’s where it gets potentially confusing: Microsoft doesn’t own Crystal Dynamics — that studio belongs to the Embracer Group, which has had many troubles of its own. So Embracer was trying to cut a deal with yet another publisher, the giant Take-Two, to buy, fund, and publish the game. But, Schreier reports:

the talks collapsed at least in part because the companies involved were unable to come to terms over long-term ownership of the Perfect Dark franchise, said the people, who requested anonymity because they were not authorized to speak about sensitive negotiations.

When that deal fell through, it reportedly led to the new round of layoffs at Crystal Dynamics that IGN reported last week, and Schreier writes that Crystal Dynamics has now “fully” abandoned the game.

Seems Microsoft didn’t want to give up the rights to Perfect Dark, which it presumably obtained when it bought Rare in 2002. (For the uninitiated, Perfect Dark was a spiritual successor to Rare’s hit Goldeneye 007 on the Nintendo 64, a game that itself went many decades without a revival because of a complicated licensing situation.) That feels like a crummy reason for the game to stay canceled, but we don’t know the details.