In-depth Analysis of Lexiang Technology’s “Self-Proof”: A Gamble on “Intelligent Emergence” and a Crisis of Trust
Hello everyone, I’m your financial journalist. Today, we’re talking about a startup called Lexiang Technology and the recent “Rashomon-like” situation they’ve found themselves in.
In short, this company, which was established less than a year ago, first issued a letter accusing the global AI giant OpenAI of “pixel-level plagiarism” of their technology. Then, they went ahead with the “world’s first outdoor live broadcast” in an attempt to prove that their robots can actually work like humans and even “self-evolve.”
But what was the outcome? Despite the lively broadcast, the doubts only grew. Behind this was not just a technical competition; it was also a battle over marketing, trust, and the industry’s fundamental standards.
Below, I’ll break down the situation into five key aspects in plain language to help you understand what’s really going on.
---
1. The Plot Twist: From “Accusing a Giant” to “Emergency Self-Proof”
Core Logic: Strike First, Then Defend Passively
The whole incident started quite dramatically. On September 10th, Lexiang Technology’s CEO, Guo Renjie, published a lengthy article directly accusing OpenAI’s chief scientist of “pixel-level copying” of Lexiang’s own developed embodied model, “Aether,” and even threatened legal action.
Why Do They Do This?
It’s like a primary school student suddenly standing up in class and claiming, “Einstein’s theory of relativity was discovered by me!”
- Benefits: It instantly captured public attention, making a startup that had been around for only 10 months (established in December 2024) famous overnight.
- Risks: OpenAI didn’t respond, but competitors and netizens began to wonder, “Are you just trying to capitalize on the hype? Is your demo fake?”
Faced with these accusations of trying to gain attention and possible fraud, Lexiang Technology didn’t choose to remain silent. Instead, they organized a live broadcast on September 16th, which was billed as the “world’s first robot live broadcast in an outdoor, real-world environment.”
Journalist’s Comment:
This is a typical “crisis communication” strategy. If someone accuses you of using fake PPTs or video editing, you set up a camera at a barbecue stand to show everyone. However, just because you dare to broadcast doesn’t mean your technology is genuine; there’s still a huge gap between that and actual capability.
---
2. The Live Broadcast: “Authenticity” in Chaos and Carefully Designed “Performance”
Core Logic: Trading “Imperfection” for “Credibility
The most interesting aspect of the broadcast was its imperfections:
1. Hasty Setup: CEO Guo Renjie admitted that the venue was rented just on Sunday, and they arrived at 4 PM. There was background noise, staff members accidentally appeared in the frame, and they even had to lift the background cloth to prove there was no one controlling the robots from behind.
2. Simple Task: The robots, “Er Ye” and “Wu Ye,” served three groups of customers at a barbecue stand, handling tasks like ordering food, grilling, seasoning, and serving dishes.
3. Key Moments: The robots failed to season the food the first time but succeeded the second. Staff explained, “They learn and optimize on their own.”
What Do Ordinary People Think?
- Supporters: They think the mess looked more authentic. If it were pre-recorded or controlled remotely, the footage would have been smooth and orderly. Lifting the background cloth indeed dispelled the suspicion of someone behind the scenes.
- Skeptics: They see it as “scripted authenticity.”
- The Audience as Actors: The three groups of customers were clearly arranged.
- Simplified Task: The barbecue setting was relatively enclosed, with fixed food locations and simple customer requests, which is different from a real, chaotic environment (like a home or a busy mall).
- The “Failure” as Part of the Design: The initial failure followed by success fit the public’s perception of AI learning, but it could also be a pre-programmed sequence.
Journalist’s Comment:
The broadcast proved that the robots could stand and work there without external control, but it didn’t demonstrate true “intelligent emergence.” It was more like a well-crafted “performance art” rather than a rigorous technical test.
---
3. Technical Weaknesses: Can 200 Hours of Video Really Lead to “Intelligent Emergence”?
Core Logic: A Huge Gap Between Claimed Capability and Actual Data
Lexiang Technology claimed their “Aether” model had achieved “intelligent emergence” through 200 hours of human video training, with no real-world data.
Here’s the major logical flaw and what experts question:
1. Insufficient Data:
- Humans learn cooking by watching thousands of hours of videos and practicing tens of thousands of times.
- 200 hours of video is roughly equivalent to watching a few series. Is that enough for a robot to learn to grill and season food spontaneously, even in unfamiliar situations? That sounds like magic in the context of current AI.
- Questions: What kind of videos were used? Were they high-quality? Were the actions clearly labeled? If it was only seasoning, that’s a specific skill, not true intelligence.
2. The Myth of “Self-Evolution”:
- The second successful seasoning attempt was interpreted as self-evolution.
- Experts’ Counterpoint: This might just be randomness or a retry mechanism. It’s like throwing darts; missing the first time and hitting the second time doesn’t mean your vision improved; it could just be a change in throwing technique.
- True self-evolution would mean the robot improved its abilities permanently through internal algorithms without human intervention, which requires long-term data accumulation and complex feedback systems—something that a one-hour broadcast can’t verify.
3. 90% Success Rate with Zero Samples?
- Lexiang claimed a 90% success rate in unfamiliar environments without any prior training.
- Reality: Even the top companies in embodied intelligence are still working on improving their zero-sample success rates in complex environments. Without public, standardized third-party tests, this claim is hard to believe.
Journalist’s Comment:
It’s like saying, “I watched 200 hours of cooking videos, never cooked, and now I can prepare a full Chinese meal with a 90% success rate.” That sounds impressive, but no one would believe it unless a Michelin judge tested it on the spot and all the data was made public.
---
4. The CEO’s Background: From “Marketing Genius” to “Technical Gambler”
Core Logic: Does Marketing Skills Mask Technical Shortcomings?
Guo Renjie, the CEO of Lexiang Technology, was previously the China executive of Chuimi Technology, famous for the viral video marketing campaign of a vacuum cleaner sucking instant noodles, which generated a million in sales in 80 days.
What Does This Background Reveal?
1. Skilled Storyteller: Guo is good at creating buzz and driving traffic. Accusing OpenAI, organizing the live broadcast, and lifting the background cloth are all highly effective marketing moves.
2. The Misalignment of Marketing and Technology:
- In consumer electronics (like vacuum cleaners), marketing often trumps technology. Users buy the “experience” and the brand, not the underlying algorithms.
- But in embodied intelligence, technology is crucial. Investors, customers, and competitors focus on the model’s generalization, sensor accuracy, and computational power.
- Using consumer goods tactics in a high-tech field can lead to underwhelming results.
Journalist’s Comment:
Guo is a skilled business strategist, but he may have underestimated the complexity of technical validation. In embodied intelligence, “going viral” requires a longer validation period and more comprehensive evidence, not just a single broadcast or an accusation letter.
---
5. The Industry Impact: A Crisis of Trust and the Dilemma of “Self-Proof”
Core Logic: When “Self-Proof” Becomes a Performance, the Industry Needs New Standards
Lexiang’s incident highlights several issues in the embodied intelligence industry:
1. The Prevalence of “Demo Culture”:
- Many companies showcase perfectly edited videos of their robots’ best moments.
- Lexiang tried to break this with a live broadcast, but even a live broadcast can be scripted.
- The Question: What defines “authenticity”? Is it the absence of external control, or the ability to handle unexpected situations?
2. Lack of Third-Party Standards:
- There are no unified standards for evaluating embodied intelligence.
- If one company claims 90% success and another claims 95%, it’s hard to determine who’s right.
- Solutions: Independent third-party organizations are needed to create standardized test scenarios and conduct blind evaluations.
3. The Damage to the Industry:
- If Lexiang’s claims are proven false or their accusations against OpenAI are baseless, it will severely damage public trust in the industry.
- Investors will become more cautious, consumers will be more skeptical, and genuine technology companies will be affected.
Journalist’s Comment:
Lexiang’s broadcast gained attention but lost trust. It proved the robots can work, but not that they exhibit true intelligence. In embodied intelligence, “intelligent emergence” is not just a slogan; it requires substantial data, continuous iteration, and transparent evaluations.
---
Conclusion: Tips for the Public
1. **Don’t Be Swayed by Terms Like ‘World’s First’ or ‘Intelligent Emergence’: These are marketing buzzwords, not technical metrics.”
2. Pay More Attention to Failures than Successes: True intelligence shows how well a system handles unexpected situations. If a robot only performs preset actions, it’s just advanced automation, not an intelligent entity.
3. Wait for Third-Party Evaluations: In embodied intelligence, who makes the claims is less important than who verifies them. Look for independent reports from reputable organizations.
4. Rationally Evaluate Accusations: If the claims about OpenAI’s plagiarism lack solid evidence and technical comparisons, they’re likely just marketing tactics.
In summary:
Lexiang Technology’s broadcast showed a company skilled at creating buzz, but it hasn’t proven it has top-tier technology. In the long run, endurance (technical accumulation) is more important than short-term marketing impact in the field of embodied intelligence.