Friday, July 17, 2026 Trend Press · Cloudflare Pages

The Trend Tribune

"All the trends that are fit to read" Morning Edition Free of Charge
TODAY'S LEAD STORY

The dangers of using "brain-like language"—a calm re-examination of the J-space paper by expert journalists

This article introduces an interview with Senior Editor Will Douglas Heaven on Anthropic's J-space research, published by MIT Technology Review on July 13th. It summarizes the technical difficulties of interpretability research, a critical perspective on Anthropic's communication style, the dangers of using the phrase "brain-like," and Anthropic's own restrained statements, and considers the importance of not blindly accepting AI research announcements but rather scrutinizing them.

The dangers of using "brain-like language"—a calm re-examination of the J-space paper by expert journalists
(Photo: illustrative)

The Dangers of Using "Brain-Like Language"—A Calm Re-examination of the J-space Paper by Expert Journalists

Previously in this column, I introduced the "J-space" paper by Anthropic, which claimed to have found a place for "conscious thought" within Claude. A week later, MIT Technology Review published an article interviewing their senior editor, Will Douglas Heaven, about this research, re-examining "what this discovery actually shows and doesn't show." This time, I'd like to introduce this meta-examination of the research itself.

The Unassuming Yet Essential Difficulties of Interpretability Research

First, it's important to understand the technical difficulties unique to this field. Anthropic's research area, "mechanistic interpretability," involves peering into the complex mathematical formulas of AI models to understand why a particular output was chosen.

Heaven explains this difficulty with a striking analogy. If a medium-sized LLM (Large-Scale Language Model) were printed on paper, it would cover an area the size of a city like San Francisco. Extracting what's actually happening within a massive mathematical formula consisting of hundreds of billions of parameters in a way that humans can understand requires specialized tools that know where and how to look, and creating those tools itself requires some prior knowledge of complex mathematical formulas—a cyclical difficulty.

A Perspective on Anthropic's "Narrative Style"

Another interesting point that Heaven makes in the interview is directed at Anthropic's style of information dissemination itself. Heaven prefaces his remarks by saying, "Anthropic is known for publishing strange and difficult research," and then makes a somewhat critical comment: "The narrative that they've created a very strange technology, and that only they can decipher it, perfectly fits Anthropic's image."

In this context, the series of events cited is when Anthropic warned that its new model was "too powerful to be a global cybersecurity risk," and immediately afterward, the US government restricted access to that model (access to some of these models has since been restored). Apart from the content of the research announcement itself, the observation that the company's narrative is consistently that it is "dangerously powerful, but only we can control it" is a perspective that cannot be ignored when considering how AI research is perceived.

Caution Regarding the Use of "Brain-like Language"

More than the technical content, what I think is most important in this interview is the caution regarding the use of the phrase "brain-like." Heaven frankly states, "I don't like using that kind of language." The point is that LLM is not a brain, and this kind of language risks misleading people into thinking LLM has more human-like capabilities than it actually does, or creating inappropriate assumptions about its behavior.

Anthropic itself has also expressed a cautious stance on this point. When Heaven inquired about this point, Anthropic reportedly issued a statement to the effect that, "These analogies (with neuroscience) were useful in designing experiments. We were able to make many experimental predictions about J-space that were initially not obvious, and these predictions turned out to be correct. However, it should also be noted that there are important differences between J-space (and language models in general) and the human brain, and we are not claiming a perfect correspondence." This is a self-restrained distinction: Analogy is merely a tool for formulating hypotheses, not proof of conclusions.

The practical question: "What problems can this discovery solve?"

So, what practical use does the concept of J-space have? Anthropic suggests using it as a means of capturing situations where the model is doing something it "shouldn't be doing." Since the words appearing in J-space don't appear in the model's final output, the theory is that they can provide signals of behaviors that would normally be overlooked, such as biased responses or weighing the pros and cons of dishonest behavior.

However, Heaven offers a cautious assessment here as well. He states, "This is purely theoretical," and comments, "This result should be viewed not as a practical tool in itself, but rather as another stepping stone in the journey of understanding this technology as a whole." His consistent stance is that before discussing flashy application possibilities, we must first accurately define its position as fundamental research.

What Researchers Can Learn from This Meta-Verification

This article itself does not report new experimental results, but rather is a critical analysis by a specialist journalist of already published research. However, when following AI research news, this attitude of not blindly accepting published content, but also verifying the validity of corporate rhetoric and metaphors is, though understated, extremely important.

Especially when applying terms referring to the inner workings of the human mind, such as "consciousness," "brain," and "thought," to AI models, it is crucial to always be mindful of whether those terms accurately represent the experimental results or are merely rhetorical devices designed to excessively stimulate the reader's imagination. I believe that research in this field can only progress healthily if we have both Anthropic's own restrained stance—that "analogy is a tool for hypothesis building, not proof"—and processes like this one, where external expert journalists re-examine his work.

AI/MLAnthropic解釈可能性J-spaceAI研究科学ジャーナリズム

"It won't stumble even if you poke it with a cane"—The inner workings of a European humanoid robot, designed by a former Optimus engineer.

This article explains Northstar, a lightweight humanoid robot unveiled on July 6th by Paris-based robotics startup UMA. It covers the background of the founding team led by Rémi Cadène, formerly of Tesla Optimus and Hugging Face; the practical choice of a lightweight design of approximately 40kg and wheeled drive; the "Real-Time Learning" architecture learned from demonstrations; the demographics behind the Europe-first strategy; and skeptical views regarding its prototype stage.

"It Doesn't Stumble Even When Prodded with a Cane"—The Inside Story of a European Humanoid Robot Created by a Former Optimus Engineer

On July 6th, in a research lab in Paris, a former Tesla Optimus engineer began poking a metal humanoid robot with a stick in front of journalists. This was a demonstration to showcase its balance and responsiveness. The company's name is UMA (Universal Mechanical Assistant), and the robot's name is "Northstar." In an industry dominated by US and Chinese companies, this is a rare instance of a robust European player making its mark.

A Team Composition That Seems Likely to Attract Funding Based on Background Alone

First, let's look at the background of CEO Rémi Cadène. After earning his PhD from the Sorbonne University, he completed a postdoctoral research at Brown University and spent approximately three years at Tesla. He was involved in the development of the Optimus AI system within the Autopilot group. In early 2024, he left Tesla and moved to the AI ​​platform Hugging Face, where he led the development of "LeRobot," an open-source library in the robotics field. LeRobot, which saw its GitHub star count grow from zero to over 12,000 in about a year after its release, is considered one of the industry's foundational infrastructure projects.

The co-founders are also impressive. Pierre Sermanet, formerly of DeepMind, serves as Chief Science Officer, Simon Alibert, formerly of Hugging Face, is CTO, and robot designer Robert Knight is Chief Robot Officer. Furthermore, Yann LeCun of Meta and Thomas Wolf, co-founder of Hugging Face, are listed as advisors. From a software engineer's perspective, this is not simply a publicity stunt, but rather a team composition that intentionally combines three different areas of expertise: AI research, practical robotics, and open-source community management.

A Practical Choice: Lightweight, Wheeled, and Soft Exterior

What's interesting from a technical specifications stand out is Northstar's design philosophy. First, its weight is approximately 40 kilograms, considerably lighter than other humanoid robots (most weighing 60-80 kilograms). At the proof-of-concept stage planned for later this year, it will employ wheel drive rather than bipedal locomotion and be covered in a flexible exterior that Cadène likens to "workwear."

This combination of "lightweight, wheels, and soft exterior" can be seen as a decision prioritizing practical safety over flashiness. For a robot working in the same space as humans, lightness and softness that can mitigate the impact in the event of contact are similar to the "unguarded operation" of NVIDIA Halos, which I previously discussed in this column. The prioritization of first gaining real-world deployment experience with wheels, rather than tackling the complex control challenges of bipedal locomotion, also feels grounded and practical.

A Learning Architecture Called "Real-Time Learning"

The core of the technology lies in a learning method that UMA calls "Real-Time Learning." Cadène explains, "It's the same mechanism as a child learning to tie their shoelaces. First, they are taught, and then they improve through practice." Instead of pre-programming individual tasks, the design allows the robot to acquire skills through demonstrations and further improve through practical application.

In the demonstration, a robotic arm equipped with computer vision sorted plastic wall plugs by color. While this isn't a groundbreaking technology, but rather an approach rooted in imitation learning, considering Cadène and his team's background in efficiently accumulating real-world demo data through the open-source platform LeRobot, the sophistication of this data collection and learning pipeline is promising.

Demographics Behind the "Europe First" Strategy

A clear business strategy is the targeting of the European market before the US and Asia. Cadène cites the labor shortage in aging Europe as the reason, stating that "labor costs are very high, and considering demographic trends, significant demand is expected." They are already in discussions with 50 potential customers, mainly in the manufacturing and logistics sectors, and plan to implement multiple pilot programs in the logistics, manufacturing, and healthcare fields by 2026.

Regarding funding, a seed funding round of approximately $40 million has been reported, with investors such as Greycroft, Red River West, Kima Ventures, and Factorial being mentioned (the official amount has not yet been confirmed by the company).

Points to View with Skepticism

Here, I, Takahashi, would like to frankly point out that what has been released is still in the "prototype version 0" stage, and is the result of approximately nine months of development by a small team. The demonstration of poking with a cane is impressive, but this only shows basic balance control, and naturally, there is no long-term operational track record in the unpredictable environment of a real factory or warehouse.

Regarding the Real-Time Learning architecture, while the concept of "learning from demonstration" is compelling, the publicly available information does not reveal the limits of its actual measurable performance. The reliability of the end effector (hand portion), the management system for demonstration data, recovery actions from failures, and ease of integration with existing factory and warehouse systems—these operational details are the key factors determining the success or failure of actual humanoid robot implementation. We await further reports on these verifications.

Summary: A "Real Candidate" Has Appeared in Europe

Amidst fierce competition from players like Agility Robotics and Figure from the US, and Unitree and AI2 Robotics from China, the fact that a European team with a solid track record at Tesla, DeepMind, and Hugging Face has made a serious move represents a significant shift in the industry landscape. However, as I have pointed out many times in this column, "the reproducibility of a demonstration" and "the reproducibility of mass production and actual operation" are two different things. The actual results of the pilot program that UMA plans to conduct later this year will likely be the first test to gauge the true capabilities of this team.

UMANorthstarヒューマノイド欧州フィジカルAIロボットコントローラ

Applying Spotify's expanding playbook to "physical health checkups"—Neko Health's $700 million funding round showcases its gamble on vertical integration.

Neko Health, a preventative healthcare startup co-founded by Spotify co-founder Daniel Ek, raised $700 million in Series C funding on July 15th, reaching a valuation of approximately $7 billion. This article explains their vertically integrated strategy of developing diagnostic hardware, cloud software, and clinical protocols entirely in-house, their aim to apply Spotify's international expansion know-how to the medical field, the participation of prominent investors such as Mark Zuckerberg, and the regulatory and operational challenges they face as they expand into the United States.

Applying Spotify's Expanding Playbook to "Physical Health Checkups"—Neko Health's $700 Million Raising and Vertical Integration Bet

Daniel Ek, who revolutionized the world of music streaming, is now seriously pursuing a business that scans the human body itself. On July 15th, Neko Health, the preventative healthcare startup he co-founded, announced it had raised $700 million in Series C funding. Its valuation is approximately $7 billion, representing a rapid quadrupling in just a year and a half since its Series B valuation (approximately $1.7 billion) in January 2025. This article delves into what this company is creating from an engineer's perspective.

Designing an Experience of a "One-Hour Full-Body Scan"

Neko Health offers a preventative healthcare service using non-invasive, radiation-free whole-body scans. In a room equipped with thousands of specially developed sensors, millions of baseline health data points, such as skin abnormalities, early-stage cardiovascular changes, and localized circulatory dysfunctions, are acquired in just seconds. The system combines blood test data, body composition information, and even Apple Health data to allow clinical staff to conduct comprehensive health assessments.

What personally interested me from a technical standpoint was that Neko Health develops its entire ecosystem in-house, from diagnostic hardware and cloud software to clinical protocols, without relying on external medical device manufacturers. The medical device industry has traditionally been structured in a siloed manner, with clinics purchasing and operating equipment from major manufacturers. Neko Health deliberately avoids this structure, controlling everything from hardware to software in-house. They claim this allows them to develop and deploy new detection functions months faster than competitors who are tied to external supply chains.

"Key Points for Expansion" as a Spotify Founder

It's also interesting that Ek is applying the international expansion know-how he cultivated at Spotify to this business. The company opened its first store in Sweden in 2023, and has since expanded to London, Manchester, and Birmingham. With this latest funding round, they are finally entering the US market—with plans to open their first store in New York this year.

Ek himself admitted that even expanding from Sweden to the UK wasn't easy, and frankly stated, "Dealing with the complex US market will be ten times more difficult." In the healthcare sector, where regulations and insurance systems vary significantly from country to country, scaling a single product experience presents a different kind of challenge than Spotify's music streaming expansion. The US waiting list is already estimated at 300,000 people, indicating considerable anticipation even before the actual service launch.

The Investor Profiles Reveal the Level of Attention

This round was led by Lightspeed Venture Partners and O.G. Venture Partners, with existing investors Atomico, General Catalyst, and Lakestar, as well as new participants Liberty City Ventures, Positive Sum, and BDT & MSD. What's interesting is the lineup of individual investors. It includes prominent figures beyond the tech industry, such as Meta CEO Mark Zuckerberg and his wife Priscilla Chan, author and entrepreneur Tim Ferriss, former professional tennis player Maria Sharapova, and musician will.i.am. The fact that Zuckerberg is personally investing in this field can be seen as a signal of industry confidence in Neko Health's concept itself: "consuming preventative medicine and health management like a subscription service."

Technical Challenges in Scaling

This is an area where we need to avoid overly optimistic views and maintain a calm perspective. The full-body scan experience relies on both precise hardware and software that analyzes massive amounts of sensor data in real time. While over 100,000 bookings and a high 75% repeat rate have already been reported, expanding this from a single clinic to the vastly larger US market will undoubtedly involve challenges beyond simple store expansion.

Specifically, the challenges include complying with FDA (U.S. Food and Drug Administration) regulations as a medical device, navigating differing medical-related laws and regulations from state to state, and, above all, the operational difficulty of simultaneously manufacturing, installing, and maintaining specialized hardware at multiple geographically dispersed locations. While "controlling both hardware and software in-house" is a strength, it also carries the risk that "the rate of expansion will be limited by the hardware supply capacity."

Things to Consider as an Engineer

Neko Health's approach can be seen as a practical application in the medical field of the "differentiation through vertical integration" strategy, which I've discussed in this column before. Combining AI, sensor technology, and cloud software to build a system independent of the existing medical device supply chain makes sense conceptually.

However, unlike digital products such as music streaming, the main battleground here is the physical installation and operation of hardware. Whether Spotify's successful international expansion will directly translate to Neko Health will gradually become clear as actual patient data and operational experience accumulate after the opening of the first store in New York. The fact that prominent entrepreneurs and tech industry funding have begun to seriously invest in preventive medicine, an area that tech companies have not previously focused on, is a development worth paying attention to in predicting the future trends of the medical tech industry.

Neko Healthスタートアップ資金調達予防医療ヘルステック垂直統合
Advertisement300 × 250