The popular AI Pin is a plug-in for ChatGPT pinned to your chest

Original source: Silicon-based position

Image source: Generated by Unbounded AI

OpenAI CEO Sam Altman’s investment in the first real AI hardware in human history, AI Pin, is the hottest hardware product in the world in the past two days. If you still don’t know about this product, you can watch this video to find out.

The company behind it, Humane, was founded by Imran Chaudhri, a former Apple Honored Designer, and Bethany Bongiorno, Apple’s former system chief, who was involved in the development of the original iPhone, and even the prototype with scroll wheels that Steve Jobs mocked when the iPhone was first released. He was so unhappy when he left Apple that he sent an all-staff email and wrote a poem before deciding to leave: “Sadly, the river dries up, and that’s when you look for a new one”. Then, Apple fired him.

The largest shareholder behind Humane is Sam Altman, and as a hardware AI Pin, its “software” is GPT-4. Its announcement coincided with OpenAI’s developer conference, which slaughtered a bunch of AI startups, and Sam Altman was its big daddy, and the discussion immediately revolved around “whether it’s a new iPhone”.

However, if you take a closer look at the AI Pin, you’ll see that its meaning isn’t black and white — it’s either an iPhone killer or it’s just hype entirely, and it’s not. In fact, it is the first experimental product of the “new hardware” idea in the AI era. In the words of my colleague Luo Yihang, it is “a hard plug-in for ChatGPT”.

Jobs’s protégés knew Jobs’s password: the interaction method was extremely radical, and the user experience was extremely conservative and restrained, just like the iPhone back then.

The biggest gimmick of the AI Pin is that it has no screen, which changes the way it interacts in one fell swoop. You can only use it most of the time through natural language, and a little more through gestures. This is completely customized based on ChatGPT’s capabilities, because one of the biggest revolutions in LLMs (large language models) is to drive the popularity of LUI (language user interface), and this AI pin is almost a game to use LUI to the extreme.

This also determines that it is an experimental object. Recently, the person in charge of a national-level application product with hundreds of millions of users in China told me that they have used LUI to transform the product very thoroughly, and a lot of the underlying code has been rewritten, but he believes that LUI is not a simple replacement upgrade, but a parallel choice that brings new possibilities. Eventually, LUI will combine GUI (Gesture User Interface) to create hardware products that belong to the AI era.

That’s what AI Pin does. Some functions are called in natural language, and some are called with simple gestures, which is also a cool point for the outside world to exclaim, after reading its Demo, I think this is actually the “hardware” of AI Agent. A large model has a Pin entity, pin it to your chest and bling-bling blind, or simply turn it into a pendant like another AI hardware startup, Rewind. This may turn into a trend.

But AI Pin, in terms of appearance and function, doesn’t it sound a bit like Misfit, which made “smart brooches” 10 years ago? At that time, there were many such chicken-ribbed “wearable devices”, and Google Glass, which was popular and disappeared faster. Will AI Pin’s endgame be the same as theirs?

The answer to this comes back to the understanding of “intelligence”. 10 years ago, whether it was Google Glass, Misfit, or a variety of smart bracelets and smart watches, weren’t they all essentially “mentally retarded hardware”? At least one thing you have to believe is that with GPT-4 as “software”, AI Pin has indeed opened a new stage of “smart” hardware. Its core capability lies in using language interaction and action interaction to invoke AI agents. Or simply, you just think of the AI Pin as an AI agent hanging on your chest bling-bling to blind others.

The essence of AI Agent is a combination of AI applications and functions with large language models as the core capabilities based on the customized needs of individuals or organizations, and it has a high degree of personalization and scenario-oriented: knock on the pin on your chest, whisper a few words to it, open your palm, see a string of brightly colored characters, tell you the suggested way to deal with a problem, tell the other party’s data a key one, or directly order a restaurant menu in your other hand for you. That’s a lot smarter than those watch bracelets. Of course, this also shows that unlike hardware in the traditional sense, the AI Pin itself will not be a general-purpose hardware, but a “hard plug-in” to realize customized AI.

Since it is a “hard plug-in”, there is naturally a whole new system behind it. This new system is no longer based on a piece of hardware, but on a large language model. Looking back at the iPhone – it was the iPhone and then the App Store, so much so that Apple is still a hardware company in terms of revenue today. So what if there is a powerful “App Store” first? OpenAI and AI Pin are taking such a reverse script.

Everyone knows that people pay attention to AI Pin because of OpenAI and ChatGPT behind it, and the founders of AI Pin know it themselves. Just as software developers are willing to be taxed by Apple because they depend on Apple’s hardware for their income, AI Pin gave way to subscription revenue on day one — Humane’s pricing method: it’s a fixed hardware price and a monthly “subscription fee” — a “tax” paid to OpenAI by calling GPT-4’s capabilities. Is this any different from the logic of ChatGPT’s plugins?

OpenAI believes that the essence of large language models is the best compressor of the world’s knowledge. However, the compressor cannot directly solve each specific problem for each person, so plug-ins are needed, so the idea of AI Agent emerged. Today, the new generation of hardware is also being shaped by this idea when chasing “AI native”: based on the basic large language model, centered on a “brain” that can compress the basic model and train it with personal data, supplemented by various hard plug-ins - finally everyone has a personalized AI agent.

AI Pin wants to do this aggressively in one step. It’s just that the more you want to do it in one step in the early experiments, the easier it is to become cannon fodder. It connects the plug-in directly to the base, which brings a lot of problems. One of the most criticized is privacy. If there is a local intermediate “safe” with more computing power, the plug-in can focus more on its own tasks, and the underlying model can be more general, then it can solve more problems.

Not long ago, I met Dan Siroker, CEO of Rewind, a popular startup similar to Humane in Silicon Valley, who demonstrated the ability to make Rewind available offline on computers, which would be more secure and personal (Rewind is also more radical than Humane, its product records the user’s every move around the clock, and is more sensitive to privacy issues), and in his opinion, the “brain” between the base model and the plug-in It still needs to be borne by mobile phones and computers, but what is needed is a new mobile phone and new computer defined by the ability of large languages. In this sense, we are still quite expecting Nvidia and Lenovo to turn computers into “AIPCs”, and Qualcomm to help vivo and Xiaomi cram large models into mobile phones.

The conclusion is basically there: AI Pin is a “hard plug-in” that pins ChatGPT to your chest, eliminates the touch screen, and allows the combination of language interaction interface and gesture interaction interface. It’s a portable AI Agent, with a hardware base that connects directly to the cloud, which is quite aggressive, more radical than Google Glass back then, which means it’s more of an experiment, but people need it.

Only radical attempts can spark the imagination. This is the reason why people are so excited about AI Pin. People see it as an “Ultraman plug-in”, that is, an extension of ChatGPT in the hardware world, the opening of a possible new hardware era. It reminds me of a time 15 years ago.

In June 2008, in San Francisco, Apple’s WWDC, the App Store was officially released at that moment. Speaking on stage, Steve Jobs said, “You can develop and publish apps like Apple itself.” Then, some of the earliest Apple developers took the stage. One of the boys in pink caught Jobs’ attention. Mr. Jobs summoned the boy in pink to his darkened room, and the 24-year-old Stanford dropout was nervous and excited, later recalling that “all he could see was his [Jobs’s] two round lenses reflecting the light.” At that time, he showed his app on stage, expressing his gratitude to Apple: “This is the beginning of a new mobile era”.

The boy in pink was Sam Altman.