Editor's Note:
When AI begins to search for its own shape, some of its choices are unexpected.
AI has given rise to a dedicated button on smartphones, seemingly rekindling their long-lost evolutionary drive. Glasses, with their natural access to sight and hearing, are beginning to resemble the next generation of personal terminals. Some small, focused devices seem more reliable than all-in-one devices at certain moments. Meanwhile, radical attempts to replace smartphones with disposable devices have met with a cold reception.
The implementation of technology is never just about stacking up functions; it's also about people's habits, the fit of scenarios, and the redefinition of "easy to use".
iFanr launches the "AI Gadgets Chronicle" column, aiming to explore with you how AI is changing hardware design, reshaping human-computer interaction, and, more importantly, how AI will enter our daily lives.
I find it difficult to fit Typeless into the software categories I'm familiar with.
It's completely different from traditional input methods—the keyboard is almost invisible in the interface, with the most prominent feature being a voice button. It's also quite different from those input methods that claim to be "AI-powered," which tend to cram features all over the homepage. Typeless, on the other hand, has very few features, as if it deliberately turned a multiple-choice question into a fill-in-the-blank one.
This nonconformity brings up a key word: crossing the line.
Input methods were originally designed to facilitate communication between people, with a clear goal—faster typing and more accurate word selection. Typeless pushes the boundaries further, focusing on organizing the needs of natural language expression in a well-organized manner. It refines language into ideas, or rather, it extracts the true intent from a sentence and then writes that intent into a directly usable text.
The input has changed. It's no longer just written for people, but more for models.
A thinking input method
I first realized that it "could think" in the most ordinary oral accounts.
When speaking, it tends to go around in circles, add details, repeat, and use many filler words. Typeless's output is more like a version that has been thought through before being written down—sentences are shorter, information is more focused, and the tone is more restrained. It doesn't obsess over recording every syllable spoken, but rather focuses on what it wants to express.
▲ The spoken content was transcribed by Typeless
The difference becomes even more apparent when you change your mind at the last minute. Traditional dictation piles all the self-corrections onto the screen, leaving many intermediate states. Typeless is more like folding up the intermediate states, leaving only the final "final draft". What appears on the screen is not the process, but the result.
When you need to break down an idea into items, you have to speak it out first and then format it yourself using a regular input method. Typeless, on the other hand, often presents the structure automatically, making the logical order clearer and the paragraph boundaries cleaner. It's like casually organizing your notes.
"Speak and revise" is another usage. After you finish speaking, add a rewriting request—more restrained, more formal, shorter, or change the tone to that of an email—and it will directly adjust the original text. There's no need to stop to select words, delete sentences, or rewrite the beginning; just continue to state your intention to revise.
Translation is also a frequent application. When switching between Chinese and English is required, it integrates translation into the input process. Even more convenient is its tone handling; it avoids translating sentences like an instruction manual, maintaining a more natural, everyday communication style.
Not comfortable speaking loudly in the office or during your commute? It offers a mode for quiet input. Voice input has often been limited by the "occasion," and this kind of adaptation determines whether it can actually be used, rather than just performing well in a quiet room.
Frequently used expressions can also be made into shortcuts—a fixed-format confirmation message, a commonly used work reply. Typeless is more like making these things into callable blocks, reducing repetitive work. Input methods change from "typing" to "scheduling".
These experiences boil down to one point: Typeless is always thinking. It digests jumbled spoken language and then outputs more organized written words. It doesn't strive to completely replicate the entire speaking process; it organizes genuine thoughts.
This is what makes it so unique.
New species of AI devices
When discussing AI products, we are more accustomed to seeing new attempts at combining hardware and software—smart glasses, AI headphones, and beanbag phones—which redefine the form and interaction methods of hardware in new scenarios. Typeless takes a different path.
It is a pure software tool, but it is still essentially an extension of hardware.
From typewriters to keyboards, and then to input methods, this line of thought has always been present. Typewriters transformed handwriting into mechanical keystrokes, keyboards transformed those keystrokes into electrical signals, and input methods transformed those electrical signals into character selection. Each evolution adds a more efficient translation mechanism between humans and text.
Typeless continues this logic, but adds a new element—AI is no longer just an aid in character selection or error correction; it becomes the core of the input process.
Traditional input methods focus on "typing out characters," with efficiency measured by the number of keystrokes, word selection accuracy, and response speed. However, in the model-driven era, the real time-consuming part isn't clearly explaining the requirements the first time, but rather the subsequent iterative revisions. Each change involves a multitude of details—tone, structure, the degree of deletion and modification, and the order of information—each requiring back-and-forth adjustments. The cost of human communication rapidly escalates at this stage.
Typeless solves this problem.
It makes the process of "say a sentence – revise it – say another sentence – revise it again" smooth, completing ten rounds of adjustments consecutively within five to ten minutes. The results of each round are immediately visible, allowing you to move on to the next round immediately. Input no longer ends with "finishing typing all the characters," but rather with "the text entering a state ready for further processing."
A new "precise input" feature has emerged here.
When typewriters and keyboards were invented, they pointed precisely to a single word or sentence. In the AI era, input has become longer, the context has become richer, and the frequency of communication has increased. Today's precision is more like controlling an extremely long context: segmenting it as desired, or writing it continuously; shortening a sentence, or expanding a paragraph; requiring it not to be divided into bullet points, or breaking the logic into several lines.
As the object of control changes, the responsibilities of the input method also change.
This is also the meaning of "input method for AI".
▲ Prompt is a transliteration of Typeless.
Typeless doesn't focus on the emotional tension of social expression; it's better suited for delivering needs to a model and then consolidating the model's output into usable text. It enhances the efficiency of communication between humans and AI. The business model aligns with this approach—a minimalist interface, no ads, and a payment method that resembles "paying for results." Subscribers have unlimited access, while non-subscribers have a fixed weekly allowance. The more the product is used, the easier it is to measure its value.
Put it back into the context of domestic input methods, and the comparison will be clearer.
Old-school input methods, exemplified by Sogou, can now be labeled "AI" and offer a host of AI features. However, they remain essentially the same products—the keyboard is still there, as are the ads and feature labels. Input methods are forced to handle too many tasks unrelated to input, easily diluting their efficiency.
Another category is the extension of AI tools, such as Doubao or WeChat keyboard. These are more like stuffing existing AI capabilities into the keyboard, creating an entry point. Entry points are certainly useful, but entry points are not the same as tools. Entry points solve the problem of "where to use AI," while Typeless is more concerned with "how to use AI more accurately."
▲ The left side shows dictation using Doubao Input Method, and the right side shows dictation using Typeless Input Method.
A true AI input method serves a different target audience. It primarily facilitates high-frequency communication with models, precise control within long contexts, and iterates through revisions until the final result is implemented. It doesn't need to create a bustling marketplace; it simply needs to streamline the most challenging环节 (link/process).
It also has side effects. When communicating with colleagues, it can sometimes sound too clean, as if all the nuances of tone have been removed. The other person might feel it lacks humanity. In such situations, they might switch back to a regular input method, manually type a few more colloquial sentences, add an emoji, or a meaningless laugh. This isn't a problem with Typeless, but rather its true place—its most natural use case is communicating with AI, not casual conversation with people.
▲ Sending it to a colleague feels a bit like being a "human-machine" interaction.
The input method market has always been a brutal arena. Usability everywhere also means it will be scrutinized everywhere. Every lag, every misjudgment, every privacy concern directly impacts its survival. Typeless aims to prove not "how powerful its model is," but "whether everyday input truly becomes faster, more accurate, and more convenient."
As communication between humans and AI becomes routine, input methods may become the most subtle yet core interface. Their role is not to write everything for the user, but to organize spoken information into more controllable and iterative text, transforming "multiple rounds of editing" from a burden into a natural process.
Whether such products can ultimately gain a foothold depends on two things: first, whether they can remain stable in all the fragmented scenarios, and second, whether they can make "paying for results" a matter of course.
There has never been a middle ground in the input layer—it either integrates into habits or is quickly replaced. As a new milestone in the evolution of AI products, Typeless has positioned itself on that narrower and steeper road.
One more thing: How we "sprayed" out an article with our mouths
The text above, and the text below, were all completed verbally, with instructions given to tools such as Typeless, ChatGPT, and Claude; not a single word was typed by hand.
In the past, it would have taken at least two hours to write an article like this, but now it only took 30 minutes.
Let me first introduce the specific details of this product. The Typeless App supports iOS and Android mobile devices, as well as Windows and Mac desktop computers.
The free plan offers 4,000 words of transcription per week; while the paid plan has no word limit, costing $30 per month, $60 per quarter, and $144 per year.
This price isn't cheap, but it aligns well with the results-oriented "pay-to-deliver" model of the AI era. Even free users won't encounter ads or too many restrictions; the main difference is simply the number of words transcribed.
In fact, Typeless is not much like an "input method". It has no traditional keyboard at all, only a few keys, not to mention AI meme-making or emoji functions. It only does its job of "speech-to-text".
I really like Typeless's global integration across devices—it's an input method on phones and hotkeys on computers, allowing it to be used across applications like an AI assistant. This is a level of detail that ChatGPT can't provide.
The whole process was quite interesting. At first, we just wanted to test the writing process using Typeless and ChatGPT, but as the conversations deepened and the draft was polished, we finally produced an article with a clear viewpoint, which was not only fluent but also had very little AI influence.
Initially, we shared some preliminary ideas, our views on the Typeless product, and some points to note regarding data collection and writing. These "stream of consciousness" were organized into clear text by Typeless and used directly as prompts for ChatGPT.
ChatGPT's first draft lacked information, had an incorrect structure, and used flat, AI-like language; it was far from being a good article. Normally, providing detailed revision suggestions would require a significant amount of writing and new prompts.
▲ Prompt is a transliteration of Typeless
But now we have Typeless. As long as we turn on the dictation, we can give feedback sentence by sentence from beginning to end, and add corresponding viewpoints and narratives based on the text.
▲ Prompt is a transliteration of Typeless
We need to provide as much detail as possible. For example, when comparing Typeless with Sogou, Doubao, and WeChat input methods, we need to emphasize the differences between these products so that AI can highlight the advantages of Typeless when writing.
▲ Prompt is a transliteration of Typeless
After several rounds of revisions, the content generated by ChatGPT is relatively complete. At this point, we can switch to Claude for polishing.
We first fed Claude several articles on new AI hardware written by iFanr, allowing it to fully learn our writing style and revise ChatGPT's draft accordingly.
Claude's initial draft still has room for improvement. At this point, we can continue to use Typeless to relay some more detailed modification suggestions until we are satisfied.
▲ Prompt is a transliteration of Typeless
In fact, the amount of text we talk about with Typeless may actually be larger than the final draft, but the efficiency of output is greatly improved, and the process is much easier than simply writing.
In the AI era, typeless technology should be "ubiquitous".
When I first tried Typeless, as someone who isn't used to using language to organize my thoughts and express myself, and who doesn't need to write long paragraphs to express my ideas, I felt that it wasn't suitable for me. It was more suitable for leaders, mentors, and clients who need to give a lot of feedback every day.
But after further exploration and use, I realized I was still too narrow-minded. In this AI era, Typeless shouldn't just be a standalone app, but rather a ubiquitous "standard feature."
On a smaller scale, "speech-to-text" should not only focus on "accuracy," but also, in the AI era, on "precision." In the future, voice-to-text messages will be concise and refined, instead of a screen full of "uh," "that," and slips of the tongue.
▲ Only 10 words of useful information were conveyed in the 42-second audio recording.
Rather than installing a Typeless app on my parents' phones, I would prefer that similar functionality be integrated directly into WeChat—or rather, that all built-in "voice-to-text" features in all apps should be reworked in a Typeless way.
The greater value lies in the fact that Typeless provides a new possibility for AI interaction.
Even though I write articles every day, my expressive ability often can't keep up with my thoughts. Even when I'm not writing, just using a keyboard to communicate with ChatGPT, many times the spark is extinguished the moment I start typing.
Things become much easier when you start speaking. I don't need to plan the structure beforehand or immediately pick the most precise words; language will "pull" the material out first, and opinions and insights will flow more naturally.
This is like guiding an intern on-site to make revisions; the instructions can be very detailed, down to how each sentence should be implemented—yes, we all have AI as a "service provider."
Expecting AI to generate everything from a single sentence is unrealistic. The information density is too low, making it easy for AI to go off-topic, and the source material is insufficient to support the result. As a result, the finished product is often empty, vague, and abstract. It may appear to be written, but it reads as if no pen had been put into writing.
For AI, "context" largely determines the quality of the generated data. We must "feed" the model a large amount of ideas, viewpoints, and corpora to obtain results that better meet expectations. Why have memory prices surged in the past two years? To run and train AI, a massive amount of context is essential, thus creating a huge demand for memory in the AI industry.
Using Typeless feels more like feeding the AI a richer corpus of language. The generated content is well-founded and the viewpoints are solid. The AI is mainly responsible for turning these fragments into more readable articles.
Therefore, not only WeChat can integrate similar Typeless functionality, but all AI companies can integrate this "AI translation layer" into chatbots to guide users to say more prompts.
Furthermore, as long as users provide enough content to AI, the gap in AI model capabilities will be further narrowed.
▲ An extremely long prompt transcribed using Typeless
Some may be pessimistic about the Typeless-ChatGPT solution, wondering if it means that human creativity will truly disappear in the AI era.
Yes, but not entirely. Typeless can only eliminate the cost and barriers of "writing," but it further highlights the importance of "thought," making human perceptions, viewpoints, and insights the true core of writing.
#Welcome to follow iFanr's official WeChat account: iFanr (WeChat ID: ifanr), where more exciting content will be presented to you as soon as possible.
ifanr | Original Link · View Comments · Sina Weibo
