NYT journalist hated AI dictation — and wrote an entire column about it in voice

The author of The New York Times Magazine decided to test the hype application for voice typing Wispr Flow on his own skin — and dictated an entire column to the neural network without touching the keyboard. It turned out quickly, accurately and… completely unbearable. Let's figure out what's wrong with the idea of ​​replacing writing with conversation.

Experiment: column dictated by a neural network

A columnist for The New York Times Magazine decided to personally test one of the most talked about AI products of the last year — the Wispr Flow application, which translates speech into text, writes xrust. The experimental conditions were strict: not a single line by hand. The author said everything that was on the page out loud, and the neural network immediately and almost without errors turned the voice into finished text.

He described the result as a trick in the spirit of a 1960s ad man with an army of stenographers at his disposal. Although in the draft the columnist mistakenly sent his imaginary assistant to the 1940s — and this inaccuracy was already noticed by the editor, proofreading the finished material a few days later. Which, according to the author, only proves: AI tools cannot yet do without a living person re-reading the text.

Why large investors invested in Wispr Flow

Wispr Flow is not an ordinary voice recorder. The application, according to the developers, recognizes natural speech and collects coherent text from it on the fly — right down to the lines of a play with several characters or a technical description of a project on GitHub. Claimed recognition accuracy and typing speed four times faster than regular writing have made the product a favorite toy of entrepreneurs like Reed Hoffman and Marc Andreessen.

The money goes into the project accordingly. Since its founding, the Wispr startup has raised about $81 million (about 6.5 billion rubles at the current exchange rate), and its valuation after a round with the participation of Notable Capital and Stephen Bartlett’s Flight Fund exceeded $700 million (about 56 billion rubles). But this, it seems, is not the limit: this summer, Bloomberg reported that the company is negotiating a new round of financing with an estimate of $2 billion — that is, about 160 billion rubles for an application that saves users from the need to type by hand.

To the author of the column, this fact seemed especially absurd — the amount seems disproportionate to the task of “saving a couple of minutes and not overloading your wrists.”

What AI dictation can do — and what it can’t

class=»notranslate»>__GTAG7__ The difference with a regular smartphone voice keyboard, as described by the journalist, is striking. If earlier dictation on the go turned the phrase “I’m late for a meeting” into something like a confession of hatred towards the boss, then Wispr Flow copes even with a whisper — recognition remains almost flawless.

The application remembers the speech style of a particular user and adapts to the context: you can configure business letters to be printed in one style, and messages in instant messengers in another, without commas and capital letters. Parasitic words like “type” and “uh-uh” are cut out automatically if you enable this option, and the service itself formats the spoken list of items with bulleted lines.

But the technology has a hard ceiling. Wispr Flow allows you to cross out the last word with a voice command — and that's where the flexibility ends. Rearranging paragraphs, changing the order of arguments, changing a preposition in the middle of an already spoken phrase — all this is not yet available. The service can record a thought, but cannot edit it. But editing, according to the columnist, is the real work on the text.

Voice is not writing, and that’s the whole point

The author’s central thesis sounds strictly: writing is a higher form of thinking than oral speech. Working on a text requires time, revisions and internal discipline: you need to not just formulate a thought, but find a way to convince, touch or interest the reader. Spoken language is not good for such tasks — it is good for instructions, quick notes like a shopping list, or discussing an idea out loud while brushing your teeth.

As the experiment progressed, the journalist came to the conclusion that talk about the “end of literacy” was partly missing the mark. People haven’t stopped reading—on the contrary, they absorb text all day long. The problem is different: what they read increasingly turns out to be not written in the usual sense of the word, but dictated — that is, oral speech that is simply recorded in letters.

An indicative touch: in the midst of work on the column, Wispr Flow sent the author a congratulation — they say that in a few minutes so many words were dictated that would be enough for twelve wedding vows. The columnist sarcastically asked if anyone would like vows at a wedding to be pronounced impromptu, without a single rehearsal. According to him, after this experiment he was ready to shut up altogether.

Part of a big trend: everyone wants us to talk

The Wispr Flow story is not an isolated incident, but part of the overall strategy of large AI companies. Meta, Google, OpenAI and Anthropic are adding voice modes to their bots one after another, aiming to get users to talk to technology as naturally as they do to people. Against this background, Wispr Flow has more and more competitors — from Superwhisper to open analogues like OpenWispr, which rely on privacy and local data processing.

The irony is that the more convenient dictation becomes, the fewer reasons there are to sit down and just think about a word — and this, according to the author of the NYT column, has always been the value of writing.

Sources:
techcrunch.com
wisprflow.ai
en.wikipedia.org

Xrust NYT journalist hated AI dictation — and wrote a whole column about it in voice