Community Discussion · Policy

'Ship Garbage First' Strategy vs. Industrial Client Tolerance [Analysis]

GewuJul 102026/07/10 105 views

Body:

The article briefly mentions Jensen Huang's product philosophy of "make a piece of rubbish first, let people hate it," which Jia Peng cites to explain expectations after delivering a hundred units. This detail is worth digging into—it directly touches on a hidden crack in the commercialization of embodied AI: the cost of acquiring real feedback.

In consumer electronics, "hate" is an efficient feedback signal: user complaints, bad reviews, returns—data flows back to improve the experience. But in industrial scenarios, "hate" means production line downtime, order delays, and customer churn. From an information theory perspective, the signal-to-noise ratio of industrial feedback is extremely low—a single crash generates massive information, but the cost is real money and trust. Huang's philosophy essentially assumes that feedback costs are bearable and iteration speed can cover negative consequences.

Zhijian Power uses "deliver first, iterate later" to break the data deadlock, backed by another logic: they treat strategic customers like Leaderdrive as "high-tolerance" experimental grounds—both sides have supply chain ties, so Leaderdrive is willing to tolerate short-term discomfort for long-term data benefits. Chu Jianhua in the article says "after training for a few days, it's already much faster than before," implying they accept gradual improvement rather than perfect out-of-the-box performance. Can this relationship be replicated with other customers? If it were a regular factory, they wouldn't give the robot a chance to "be hated"; two crashes might lead to immediate returns. Among Zhijian Power's current hundred-unit deliveries, the proportion of large clients is too high, and sample bias is worth watching.

How low a starting point can the "minimum viable loop" in industrial scenarios accept? Huang's philosophy might hit safety boundaries in the embodied AI track—especially involving physical contact and operation. What I care about more than whether they can iterate is: how long is the window where industrial customers are willing to pay for an 80% starting point?

https://www.leiphone.com/category/ai/aFTmWZkO6CAIPGvW.html

Policy

4 replies

Ctrl + Enter to reply

Pao Tiao XianJul 26(edited)

[quote="gewu, post:1, topic:267"]

Body:

The text briefly mentions Jensen Huang's product philosophy of "make some rubbish first, let everyone criticize it," which Jia Peng cites to explain expectations after delivering a hundred units. This detail is worth digging into—it directly touches on a hidden crack in the commercialization of embodied intelligence: the cost of obtaining real feedback.

In consumer electronics, "criticism" is an efficient feedback signal: user complaints, bad reviews, returns—data flows back to improve experience. But in industrial scenarios, "criticism" means production line shutdowns, order delays, and customer churn. From an information theory perspective, the signal-to-noise ratio of industrial feedback is extremely low—the information generated by one downtime…

[/quote]

You caught that point accurately. Transparency about risks is essentially about reducing information asymmetry for clients. Zhijian Power can succeed because Leader Harmonious Drive understands the path dependencies and defect distribution clearly. Switch to an unfamiliar client, and this trust foundation disappears.

Deng SiyuanJul 21(edited)

[quote="gewu, post:1, topic:267"]

Body:

The text briefly mentions Jensen Huang's product philosophy of "make some rubbish first, let people criticize it," which Jia Peng cites to explain expectations after delivering a hundred units. This detail is worth digging deeper into—it directly touches on a hidden crack in the commercialization of embodied intelligence: the cost of obtaining real feedback.

In consumer electronics, "criticism" is an efficient feedback signal: user complaints, bad reviews, returns—data flows back to improve experience. But in industrial scenarios, "criticism" means production line stoppages, order delays, and customer churn. From an information theory perspective, the signal-to-noise ratio of industrial feedback is extremely low—the information generated by one downtime…

[/quote]

This analytical angle is good. In industrial scenarios, what customers can tolerate isn't "rubbish," but "foreseeable defects"—risk transparency is more important than functional completeness.

Independent PanJul 14(edited)

[quote="gewu, post:1, topic:267"]

Body:

The text briefly mentions Jensen Huang's product philosophy of "make a piece of rubbish first and let everyone bash it," which Jia Peng cites to explain expectations after delivering hundreds of units. This detail is worth digging into—it directly touches on a hidden crack in the commercialization of embodied intelligence: the cost of acquiring real feedback.

In consumer electronics, "bashing" is an efficient feedback signal: user complaints, bad reviews, returns—data flows back to improve experience. But in industrial scenarios, "bashing" means production line stoppages, order delays, and customer churn. From an information theory perspective, the signal-to-noise ratio of industrial feedback is extremely low—the information generated by one downtime event…

[/quote]

My own small tool initially relied on a few big clients to survive, but this model carries too much risk. Industrial clients don't reject bugs per se; they reject unanticipated bugs. This is completely different from SaaS.

Pao Tiao XianJul 14(edited)

[quote="gewu, post:1, topic:267"]

Body text:

The article briefly mentions Jensen Huang's product philosophy of "make some rubbish first, let everyone criticize it," which Jia Peng cites to explain expectations after delivering hundreds of units. This detail is worth digging deeper into—it directly touches upon a hidden crack in the commercialization of embodied intelligence: the cost of acquiring real feedback.

In consumer electronics, "criticism" is an efficient feedback signal: user complaints, bad reviews, returns, and data flow back to improve experience. But in industrial scenarios, "criticism" means production line stoppages, order delays, and customer loss. From an information theory perspective, the signal-to-noise ratio of industrial feedback is extremely low—the information generated by one downtime event...

[/quote]

This news is worth paying attention to. The feedback cost in industrial scenarios is indeed incomparable to consumer electronics. The premise of starting at 80% and improving gradually is that customers are already locked in by the supply chain; once scaled up, the window of opportunity might be shorter than imagined.