A full-size humanoid from the Chinese firm PNDbotics recently appeared in a tech demonstration holding a bow, drawing the string, and releasing an arrow. The moment lands as more than a stunt. It shows how far the company’s Adam platform has come in coordinating dozens of joints, sensing force, and keeping its balance while performing a precise, two-handed action that once belonged only to people.

Adam is a slender 1.67 meters tall and weighs between 60 and 63 kilograms, however this can fluctuate depending on what’s on board. When you factor in the typical human height, you get a number that is light enough for indoor use while remaining hefty enough to stay place when its center of mass fluctuates, which it does frequently. The higher-spec models include 41 to 43 degrees of freedom. Again, the legs have six joints to play with, each built around a really clever biomimetic hip joint that allows the pelvis to tilt and rotate in ways that earlier versions of this type of thing could only dream of matching, and to make things even more interesting, it’s all powered by quasi-direct-drive actuators that deliver a healthy 340 to 360 newton-meters of torque at the hips and knees. These are positioned close to the joints to keep the weight down, resulting in a very soft and compliant feel when Adam brushes up against something or someone.

Unitree R1 Humanoid Robot (White, R1)

• Three models, one lightweight platform R1 Air (20 DOF, monocular camera), R1 (26 DOF, binocular camera, head+waist joints), and R1 Edu (26 DOF…
• Easy setup – no coding required for basic use Unbox, power on, and start. Manual teaching feature: physically pose the robot, and it replays the…
• More DOF = more expressive movement 26‑DOF models (R1 / R1 Edu) add head and waist articulation for smoother dance and running. For safety reasons…

The same hardware is used for the upper body, as the higher-spec models’ arms are a true showstopper, with thirteen degrees of flexibility each and finishing in some seriously spectacular multi-fingered hands that can close around a bow handle, nock an arrow, and manage the draw with ease. The actuators have force sensing built in, allowing the robot to detect resistance and modify its grip without smashing the equipment or losing the string. All of this is powered by a battery pack with a capacity of 1,170 to 1,190 watt-hours, which provides Adam with enough energy to run for a couple of hours of active work – enough for a number of practice sessions or brief demos.

The control software is where everything comes together. PNDbotics has adopted deep reinforcement learning, first training in simulation and then transferring the skills to a real-world robot. Then they added additional imitation learning using human motion capture data to fill in any gaps, allowing Adam to learn for himself how individuals shift their weight, rotate their torsos, and stabilize their arms throughout a variety of complex maneuvers. There are also whole-body control algorithms that keep the feet anchored and the posture upright even when the arms are extended and the string is tugging strongly. The end result resembles a developed skill rather than a pre-programmed routine.

Adam has already been put through its paces in public and has handled a wide range of demands. It climbed up a one-meter-high box by gripping the edge with both hands, hauling itself up, and swinging a leg over, all while maintaining its balance. It walked stairs in unstructured circumstances during the 2026 World Artificial Intelligence Conference, marking one of the platform’s first public appearances. It has even played a keytar on stage at a music festival, danced the Charleston, and completed a variety of other impressive-looking tasks, including some very sophisticated motion-capture routines that map the actions of a human operator onto the robot in real time. The bow-and-arrow sequence is the same thing: a full-body task that requires strength, timing, and precise motor control in a variety of hard ways.

I’m a high school student in Paramus, New Jersey. I’m also the creator of a free economics program that I teach to middle school students. With each virtual session I lead, I share a link in the chat with my students, and they load the simulation I created — one that puts them in positions like CEO of a sports team or commissioner of a league.

In one simulation, every student runs a team’s payroll and decides whether keeping their star player is worth exceeding their salary cap and crossing into “luxury tax” territory. Nobody tells them about the consequences of going into the luxury tax; they learned the way real general managers do, by paying the additional fee and experiencing the negative impact it can have on a team.

I created the program because I am a huge sports fan and I wanted to share my passion with other kids. I also knew that a lot of kids love professional team sports and almost none of them love economics, even though the two are often linked. A team salary cap is a budget constraint. A trade deadline is linked to supply and demand. A rookie contract is a bet on future value.

Simulations Unveil Concepts

So far, I’ve led two six-week online classes for about 30 middle school students. Each class combines slide lessons explaining economic concepts, open discussions, and simulations I built with artificial intelligence. 

My students played a luxury tax game where they managed a roster under real salary cap pressure. They played an NFL draft game where they weighed risk against reward with every draft pick. There were trade negotiations, ticket-pricing challenges, and roster-building hurdles with no clear “right” answers. Hiding inside each simulation were economic concepts that teachers have spent decades trying to get students to care about, like incentives, opportunity cost, scarcity, risk and market value.

I expected the sports fans in my classes to dominate the conversations. In reality, the student who knew every sports statistic wasn’t always the one who made the smartest payroll decision. The student who couldn’t name a single starting team lineup often built the most balanced roster in the class because they understood the trade-offs better than the sports fanatics.

Leading these online classes gave me real insight into what it takes to be a successful teacher. What I realized is that a great lesson is not about the content; it’s about how students interact with the content. I could have lectured for an hour on opportunity cost, watching eyes glaze over. Instead, I gave them a roster and a budget that couldn’t cover everyone they wanted to draft, and the concept of opportunity cost practically taught itself. Instead of being just another “vocabulary word,” the concept became the reason they couldn’t sign Jalen Brunson.

Teaching these middle schoolers and seeing what really motivated them to engage with the material makes me think this can be applied to every subject. It doesn’t have to be done with a video game. It can be accomplished with a worksheet, flashcards or other activities. The idea just has to make students feel like the concepts they’re exploring matter outside of the classroom.

Discussion Fuels Learning

Another observation that surprised me was how much the discussions mattered. I initially thought the simulations would be the main event and the teaching and discussions would be “filler.”

In reality, it was the opposite.

It turns out that the simulations created the basis for discussion and disagreement, and that is where the learning happened. For example, after the luxury tax game, I had students defend their choices to the class. One student explained that they paid the tax because they were “in a championship window,” while another student didn’t pay the tax because their roster was “too young to waste the money.” They were doing real economic reasoning, but they didn’t notice. They were simply having a great time discussing their teams.

Peer-Led Teaching Lowers the Stakes

Peer-led teaching is not a replacement for trained teachers. But I do think that peer-led programs such as the one I led can be a valuable complement to teacher-led classes because they can lower the stakes and make learning less daunting for students. When the person teaching you is closer to your own age and the lesson is built around a game instead of a grade, students are more willing to take risks.

Based on my observations, students in student-led classes seem more likely to guess and debate. Their ideas might be wrong, but they are willing to try and make themselves vulnerable in front of their peers – something that they may not be willing to do inside a formal classroom. That freedom to try, fail, and try again is where a lot of real understanding begins.

Lessons Learned Along the Way

If you are a teacher and I haven’t chased you away with my precociousness, here are three more pieces of advice to help you help your students.

Incorporate what students love. Consider starting with something your students are already interested in and then sneak in the educational subject behind it. I never opened a class by saying, “Today we’re learning about scarcity.” Instead, I opened by saying, “You have $40 million and three players you can’t afford to keep.” The economics lesson arrived after the engagement, not before. Using a topic students already cared about did more to fuel their learning than any creative explanation I could’ve ever written.

An accessible entry point gives students a reason to care before asking them to memorize anything. The students in my classes kept asking for more simulations. Parents told me their children were recommending the class to friends, which is not something I have ever heard a kid do for an economics worksheet.

Let students be wrong. Some of my best sessions came from a student making a terrible trade and the whole class reacting. My first instinct was usually to jump in and explain why the move would not work. But I realized that the better outcome arrived when I waited and let the students teach.

This is the kind of class “mistake” that students remember. Those moments stick with them much longer than any rule or definition written on a slide and help them truly grasp the underlying concepts.

Keep the barrier to entry low. I purposefully made my program free and ran it on Zoom, so students of all backgrounds can join from anywhere in the world. I communicated that they do not need to be great at math, nor do they need to know every player in the NBA or NFL. They do not even need to think of themselves as “economics kids.” They just need to be curious about why things cost what they cost, why teams make the choices that they make, and why every decision has a trade-off.

Making the program easily accessible was particularly beneficial to those students who need a subject taught in a less formal setting. They are often the first ones who are intimidated by traditional classroom learning. If a class feels like it has prerequisites before you even reach the first lesson, you sometimes lose the student as soon as they walk into the classroom.

I started my program because I thought a cool sports simulation could make economics more interesting to students. What I learned is that peer-led discussion in a topic students are invested in – whether it is sports or pop culture – can do something better. It lowers the stress, creates a welcoming environment for all students, and still creates genuine opportunities for learning. I’m looking forward to offering more sessions – teachers can join too.

Apple has officially launched Apple Upgrade, a leasing program that lets customers pay monthly for an iPhone, iPad, Mac, or Apple Watch instead of buying the device outright. The program went live today on Apple’s website in partnership with Klarna, replacing the iPhone Upgrade Program that Apple has offered since 2015.

What you can lease and what it costs

The lease length depends on what you’re planning to get. Phones and watches get shorter terms, 12 or 24 months, while iPads and Macs stretch to 24 or 36 months, giving those categories more room before the lease is up. The program covers most current hardware, including iPhones, the MacBook Air, MacBook Pro, iMac, Mac Studio, iPad Pro, iPad Air, iPad mini, Apple Watch Series 11, and Apple Watch Ultra 3. However, the iPhone 16, base iPad, and MacBook Neo aren’t eligible. Anyone leasing an iPhone also needs to be on a postpaid plan with AT&T, T-Mobile, or Verizon, as prepaid carriers don’t qualify.

Pricing differs from device to device. A 256GB iPhone 17 Pro starts at $31.99 a month on the 24-month plan, or $45.99 a month for those who want out in a year. Tablets and laptops land in similar territory, with a 256GB iPad Pro starting at $24.99 a month over 36 months and a 16GB MacBook Pro starting at $38.99 a month over the same term.

Apple doesn’t require a security deposit for the lease, but it also doesn’t allow customers to make a down payment to reduce their monthly payments. The one lever customers do have is a trade-in, which can reduce the monthly rate for the length of the initial lease.

What you’re actually committing to

Signing up is simple on paper. Apple requires a Klarna account and runs only a soft credit check, so applying won’t affect credit scores, but Klarna then owns the relationship for the life of the lease, handling billing through its own app rather than Apple’s.

The terms get more complicated after that. Upgrading to a new device before a lease ends is allowed, but the fee for doing so grows the more payments are still owed, which turns an early upgrade into a calculated cost rather than a simple swap. A device also has to come back in good condition when the term wraps up, or Apple will charge a damage fee, something customers can offset with an AppleCare plan, though it isn’t bundled into a lease price the way it was under the old iPhone Upgrade Program.

Apple is also offering a short window to walk away entirely. Canceling within 14 days of receiving a device cancels the lease outright, but missing that window means paying an early termination fee instead. Customers who ride out the full term get three options. They can upgrade and return the current device, leave the program and return it, or pay a one-time fee through Klarna to buy it outright. Waiting more than six months to decide removes the choice, since Apple will automatically charge the purchase fee by default.

Apple Upgrade retires a financing model Apple built around outright ownership and replaces it with one in which monthly payments don’t guarantee owning anything at all. Customers deciding whether to sign up should read the fee structure in Apple’s FAQ closely, since it does more to define the actual cost of the program than the marketing copy on the page.

AI and ML

For enterprise agents, less is more

Anthropic and OpenAI have both proposed the deployment of agent armies as a way to maximize the notional benefits of artificial intelligence systems.

But the optimal agent workforce isn’t necessarily the largest. Researchers affiliated with the NTT Research’s Physics of Artificial Intelligence (PAI) Lab and Harvard University’s Center for Brain Science (CBS) say there’s a point at which enterprise AI agents get in each other’s way.

Back in 2024, OpenAI used the term “swarm” to describe multi-agent architecture, before settling on a more anodyne term “OpenAI Agents SDK.” Anthropic is focused on multi-agent orchestration.

In both cases, the implication is that productivity rises along with the number of agents you run, an idea that means you spend more at OpenAI and Anthropic.

But according to Elizabeth Pavlova, senior data scientist at Harvard CBS, and Hidenori Tanaka, head of NTT’s PAI Lab and associate professor at Harvard CBS, adding agents doesn’t necessarily improve productivity and may hinder it at some point.

The ideal number of agents is 16, they say, at least for the Flag Game used to assess multi-agent interaction. With fewer than 16 agents, the software cannot gather enough evidence to reach a consensus decision. With more than that, the agents become polarized and split into opposing camps.

In a paper scheduled to be presented at the ICML 2026 Workshop AI4GOOD, Pavlova and Tanaka describe the Flag Game, which is intended to measure the tension between an agent’s private evidence and how that gets reconciled with social input from other agents.

The goal of the game is to guess a national flag for which each agent has only a fractional view.

Distributed systems require some consensus mechanism in order to work together. So the agents are allowed to communicate with each other.

Each agent in the game considers the fraction of a flag that it can see and each makes a guess about the full flag. The agents then interact, passing their guess on to a listener, broadcasting their guesses to other agents, or engaging with a manager system that takes their guesses and returns an opinion about the nature of the flag.

The game ends when 85 percent or more of the agents agree on which country’s flag they “see” during three consecutive rounds of analysis, or after some limited number of allowed steps.  

Adding agents improves performance up to a point, but then becomes a problem. When the agent population is large, groups emerge, each of which validates its collective position through the consensus of its members.

“Our research shows that simply adding more AI agents does not necessarily improve performance — just as hiring more people does not automatically make a company more effective,” said Tanaka in a statement. “Communication becomes harder, and groups can split into competing camps.”

The authors’ message for enterprise leaders is that when it comes to AI agents, focus on quality rather than quantity. ®

When I first caught a glimpse of The Adventures of Elliot: The Millennium Tales, it sparked a great sense of anticipation within me. After all, Square Enix and Claytechworks were collaborating to bring a brand new HD-2D RPG to the table, which appeared to combine sprinklings of classic Zelda titles with the visual, sonic, and environmental grandeur from series such as Mana and Dragon Quest.

I’m a huge fan of the HD-2D graphical style, and massively enjoyed recent releases such as Final Fantasy: The Ivalice Chronicles and Dragon Quest 1 & 2 HD-2D Remake, but I still wasn’t quite sure if I’d love The Adventures of Elliot. Was the action combat going to be polished and engaging enough? Was the world going to deliver the spectacle and appeal conjured up with other series? Would the narrative have me hooked?

Anthropic founder and CEO Dario Amodei responded on Monday afternoon to murmuring in the industry that his company somehow supports efforts by the U.S. government to ban open-weight Chinese models, or possibly open-weight models generally.

“Anyone who has read my past writing should know that I don’t regard such bans as a useful measure, but let me state it clearly so that there is no doubt: Anthropic has never advocated for a ban on open-weights models,” he wrote, emphasis his own.

His response came after Nvidia founder and CEO Jensen Huang took to X (his first post on the platform) to share an open letter on Friday in which Nvidia and a long list of other AI companies including Hugging Face, Meta, Microsoft, Mistral, and Nvidia urged policymakers not to impose broad “premature restrictions” on open-weight AI models.

While that letter did not mention China specifically, the debate in the industry has centered on allegations that Chinese AI labs are growing in capability often by stealing intellectual property from their American counterparts. One method is distillation, where an AI bombards another model with prompts to learn how it works.

Amodei said in the blog post published Monday that he views open-weight models as falling into a different bucket than the threat China poses. “Open-weights models that don’t have dangerous capabilities are a public good: they don’t cost anything besides the compute needed to run them, and they provide value to businesses, developers, and researchers.”

Amodei said in his post that he has longstanding fears about AI, but businesses using open-weight models, even ones from China, are not among them.

His fear is that “authoritarian governments” will build models that are more powerful than those in the U.S. to achieve “permanent military superiority” and/or use AI to repress their own people. He said the Chinese Communist Party (CCP) isn’t the only authoritarian government he’s concerned about, but it is the “most capable.”

Amodei also said he fears AI will enable “biological attacks” not just cybersecurity ones. And, in his view, open-weight models are more dangerous in those scenarios because it’s difficult to apply guardrails to them or monitor their usage. He also noted, citing a UK AI Security Institute report, that once open-weights are released they cannot be withdrawn

This runs counter to what open source advocates argue: that access to powerful open models not controlled by a single entity helps defenders protect themselves.

Amodei listed actions he believes would thwart China, including restricting its access to powerful chips (which has been a longstanding U.S. policy) and calling for a formal crackdown on distillation. The U.S. has threatened sanctions against China if it determined there was IP theft involving U.S. models.

Interestingly, Amodei also said he supports growing efforts, some led by the U.S., to create a model safety testing organization, especially if the entire world, including China, agreed to submit to it.

“I think this idea is actually close to a consensus: I have been heartened both that the Trump administration has moved in this direction in recent months, and by recent industry proposals that would apply such testing to the most capable models regardless of their country of origin or whether they are open or closed (while exempting less capable models, such as those from startups and academia, entirely),” Amodei wrote.

“Note that to be effective,” he added, “testing would need to be global, which means even the CCP would need to be on board. I think this may actually be possible … limited cooperation around preventing AI biological weapons may be possible because it is in China’s interest too.”