We explore the operating principles and user experience of 'Don't Hit Send,' a new real-time conversation interface where the AI begins its response the moment you stop typing.
Imagine you are having a very long messenger conversation with a friend. However, every time you type a sentence, you have to press the ‘Send’ button and wait for your friend to read it and check for a reply. What if your friend could grasp your intent and prepare an answer in real-time, even before you finish speaking or while you are still thinking?
An experimental interface called ‘Don’t Hit Send,’ which recently appeared on the AI tech community Hacker News, offers us exactly this experience. Don’t Hit Send: the model answers while you type
Why It Matters
Until now, we have been accustomed to the classic ‘input question → send → wait for response’ method when using AI. However, this method breaks the flow of conversation and gives the feeling of exchanging stiff office emails.
‘Don’t Hit Send’ aims to organically connect conversations with AI—as if talking to a real person—by eliminating this artificial ‘sending’ step. GitHub - scalattice/dont-hit-send Users just need to type their thoughts freely without needing to wait for a response. The AI creates a response in real-time, following the flow of the typing. This is a significant change that transforms the way we use AI from a simple ‘command input tool’ into a ‘co-author’ or ‘conversation partner’ with whom we can think and share opinions together.
The Explainer
Metaphorically, you can see this technology as an AI that carefully ‘observes’ your typing habits.
This interface largely divides the screen into two windows. On the left is the ‘draft’ window where the user freely writes, and on the right is the ‘response’ window where the AI reads that writing and builds up a response in real-time. GitHub - scalattice/dont-hit-send: The model answers while you type
The operating principle is quite clever:
- The user starts typing.
-
If the user pauses typing for about 350ms (0.35 seconds), the AI judges, ‘Ah, this person is organizing their thoughts for a moment!’ [Show Hacker News](https://www.hacker-news.news/Show) - It immediately begins generating a real-time response (Streaming Chat Completion, a feature where the AI completes text in real-time) based on everything written up to that moment. Don’t Hit Send: the model answers while you type
- If the user modifies the content or continues typing, the AI immediately cancels the previous response generation and prepares an answer again to match the changed draft. GitHub - scalattice/dont-hit-send
It’s similar to applying a filter in a photo editing app, where the preview screen changes in real-time as you move the adjustment slider. Even the time spent thinking becomes part of the conversation.
Where We Stand
Currently, ‘Don’t Hit Send’ is an experimental project that maximizes real-time interaction. Crucially, this method is not an unstable real-time streaming that shoots data to the server every time a key is pressed. GitHub - scalattice/dont-hit-send Instead, it adopts an efficient method of cleverly detecting the user’s ‘pause’ patterns and transmitting the entire content anew.
| Of course, as it is in the early stages, there are points to consider. Because the response keeps changing in real-time, the user’s concentration might be distracted while writing. Also, technically, since it must cancel the previous request and start a new Chat Completion at each typing pause, the model’s fast response speed is essential. [Show | Hacker News](https://www.hacker-news.news/Show) |
What’s Next
Going forward, it seems this kind of ‘conversation without a send button’ will be integrated into more productivity tools. When we write documents or code, the AI will look over our shoulders, reading what we write and offering appropriate suggestions in real-time whenever we pause. Conversations will become more similar to the speed of human thought, and we will move beyond the relationship of ‘asking and receiving answers’ with AI to a relationship of ‘completing thoughts together.’
MindTickleBytes’ AI Reporter View
As technology becomes more like humans, the way we handle that technology must also become more human. I believe that removing the ‘Send’ button is not just a UI change, but a change containing AI’s consideration to not disturb the human ‘Flow of thought.’ An environment is being created where we can have deeper-level conversations with AI.
References
- ShowHN:Don’tHitSend–themodelanswerswhileyoutype https://news.ycombinator.com/item?id=49669012
- GitHub - scalattice/dont-hit-send:Themodelanswerswhileyoutype. https://github.com/scalattice/dont-hit-send
- Hacker News => Show https://www.hacker-news.news/Show
- GitHub - scalattice/dont-hit-send: The model answers while … https://vuink.com/post/tvguho-d-dpbz/scalattice/dont-hit-send
- When you press the send button
- When the user pauses typing for about 350ms
- When you finish your question and press the enter key
- The previous response is maintained
- The previous response is canceled and regenerated as a new draft
- It is merged with the previous response
- It transmits in real-time with every keystroke
- It transmits the entire draft every time
- It uses a bidirectional socket