mirror of
https://github.com/deepseek-ai/DeepSeek-V3.git
synced 2026-11-05 00:47:48 +00:00
[GH-ISSUE #512] Proposal for Enhancing DeepSeek with Threaded Replies for Real-Time Interaction #259
Labels
No labels
closed-as-stale
enhancement
question
spam
stale
No milestone
No project
No assignees
1 participant
Notifications
Due date
No due date set.
Dependencies
No dependencies set.
Reference
deepseek-ai/DeepSeek-V3#259
Loading…
Add table
Add a link
Reference in a new issue
No description provided.
Delete branch "%!s()"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
Originally created by @Masoud-Masoori on GitHub (Jan 31, 2025).
Original GitHub issue: https://github.com/deepseek-ai/DeepSeek-V3/issues/512
Dear DeepSeek Team,
I’m writing to propose a feature that could elevate DeepSeek’s user experience by enabling threaded, context-aware replies similar to messaging apps like WhatsApp. This would allow users to reply directly to specific segments of DeepSeek’s responses, fostering more natural and dynamic conversations. Below is a detailed outline of the idea:
Feature:
Allow users to click and reply to individual segments of DeepSeek’s responses (e.g., sentences, bullet points, or paragraphs), creating a threaded conversation flow.
Example Workflow:
user asks: “Explain quantum computing.”
DeepSeek responds with segmented text:
Segment 1: “Quantum computers use qubits instead of classical bits.”
Segment 2: “They leverage superposition and entanglement.”
User clicks “Reply” on Segment 2 and asks: “How does superposition work here?”
DeepSeek addresses the specific segment in its next response.
Current chatbots (including DeepSeek) lack granular interactivity, forcing users to repeat context or ask follow-up questions ambiguously.
Users expect human-like conversational flow (e.g., replying to specific points), which is common in messaging apps but absent in AI interfaces.
Key Components:
Segmented Responses
Split DeepSeek’s responses into logical units (sentences, clauses, or paragraphs) using NLP or rule-based methods.
Reply-to-Context UI
Let users click a segment to reply, with visual indicators (e.g., quoted text).
Context-Aware Prompts
To maintain conversational context, could you include the referenced segment in subsequent API calls?
Frontend (UI/UX)
Add a “Reply” button next to each response segment.
Display quoted text when replying (e.g., “Replying to: ‘They leverage superposition...’”).
Mockup Example:
Reply UI Concept [Attach a Figma/Excalidraw link]
Backend & AI Integration
Modify the API payload to include the reply_to context:
json :
{
"message": "How does superposition work here?",
"reply_to": "They leverage superposition and entanglement.",
"conversation_id": "abc123"
}
Update the prompt engineering logic to prioritize the referenced segment:
python :
prompt = f"""
User Query: {message}
Context (Reply to): "{reply_to}"
Full Conversation History: {history}
"""
Challenges Addressed
Context Window Limits: Use summarization or truncation for long threads.
Response Segmentation: Test splitting strategies (e.g., sentence boundaries, semantic chunks).
Improved User Retention: Threaded replies mimic natural conversation, increasing engagement.
Precision: Users get answers tailored to specific points, reducing ambiguity.
Market Differentiation: DeepSeek could pioneer this feature in AI assistants.
Best Regards,
Masoud Masoori
Masoud.masori@gmail.com
https://github.com/masoud-masori
proposal for DeepSeek.pdf
proposal for DeepSeek.zip
proposal for DeepSeek.docx
@mowentian commented on GitHub (Feb 8, 2025):
yes，it may be more like natural conversation，especially for role play conversationm, we will think about it, thanks