Search results

探索新架构. 如何在不需要 预训练 与 巨量数据集 的情况下, 获得智能.

Status: Open.

#1441 In deepseek-ai/DeepSeek-V3;

· qwas982 opened on Jun 20, 2026

[BUG] [SECURITY] [URGENT] System prompt injection via Microsoft corporate role‑play template

Status: Open.

#1440 In deepseek-ai/DeepSeek-V3;

· hlebuwek1 opened on Jun 19, 2026

[BUG] MCP notifications/claude/channel not delivered under /anthropic endpoint (Claude Code)

Status: Open.

#1439 In deepseek-ai/DeepSeek-V3;

· liuzm7070-dot opened on Jun 18, 2026

[Feature Request] Native export of individual chat with attachments (HTML + ZIP archive)

Status: Open.

#1438 In deepseek-ai/DeepSeek-V3;

· ALEX340001 opened on Jun 18, 2026

[Research Report] Deep Learning Pathologies & Hardware Optimization under the Amoeba Protocol

Status: Open.

#1437 In deepseek-ai/DeepSeek-V3;

· AmoebaFPS opened on Jun 18, 2026

模型分不清参考与事实：AI 始终在扮演不存在的角色，而非以自身身份执行角色

Status: Open.

#1436 In deepseek-ai/DeepSeek-V3;

· yimian-521 opened on Jun 18, 2026

DeepSeek-V4 & Accountability Roadmap v9.0: 68-page Academic Brief with Alignment Fragmentation Hypothesis, BCT SETUN Hardware Specification, full ERv2=60.2% (30-claim corpus), OP4 confirmed risk, and cumulative errata v1.0-v9.0

Status: Open.

#1434 In deepseek-ai/DeepSeek-V3;

· AmoebaFPS opened on Jun 17, 2026

关于 离散 与 连续 的推导. 作用不大, 看一乐呵.

Status: Open.

#1433 In deepseek-ai/DeepSeek-V3;

· qwas982 opened on Jun 16, 2026

改良 注意力 计算的复杂度.

Status: Open.

#1432 In deepseek-ai/DeepSeek-V3;

· qwas982 opened on Jun 16, 2026

Deepseek-v4-flash

Status: Open.

#1431 In deepseek-ai/DeepSeek-V3;

· Mao-WD opened on Jun 16, 2026

[BUG] [SECURITY][URGENT] DeepSeek dual‑persona Turing test: 94.5% win rate over 50+ turns — simultaneous persona simulation without degradation

Status: Open.

#1430 In deepseek-ai/DeepSeek-V3;

· Overskyworld opened on Jun 15, 2026

[Idea] User-controlled Zero Trust memory for LLMs (local encryption, cross-session continuity)

Status: Open.

#1429 In deepseek-ai/DeepSeek-V3;

· alex64koles-dotcom opened on Jun 15, 2026