Join us

ContentUpdates and recent posts about DeepSeekMath-V2..
Discovery IconThat's all about @DeepSeekMath-V2 — explore more posts below...
 Activity
@yelbur started using tool Python , 3 hours, 53 minutes ago.
 Activity
@yelbur started using tool Node.js , 3 hours, 53 minutes ago.
 Activity
@yelbur started using tool Go , 3 hours, 53 minutes ago.
 Activity
@yelbur started using tool Fedora , 3 hours, 53 minutes ago.
 Activity
@yelbur started using tool Docker , 3 hours, 53 minutes ago.
 Activity
@yelbur started using tool BigQuery , 3 hours, 53 minutes ago.
Link
@kala shared a link, 1 day, 9 hours ago
FAUN.dev()

Realtime Prompting Guide

OpenAI shipsgpt-realtimeand declares GA for theRealtime API. It's a speech-to-speech model that tightens instruction-following, steadiestool calling, and lifts voice fidelity. Latency drops. True realtime agents become possible. The release prescribesprompt skeletons,JSON envelopetool outputs,sessio.. read more  

Realtime Prompting Guide
Link
@kala shared a link, 1 day, 9 hours ago
FAUN.dev()

Do you need an MCP to build your native app?

Do you need an MCP to build your native app? Surprisingly, modern agents succeed either way. The real difference is how much time, cost, and context you waste along the way... read more  

Do you need an MCP to build your native app?
Link
@kala shared a link, 1 day, 9 hours ago
FAUN.dev()

The Pentagon is making a mistake by threatening Anthropic

Anthropic's Claude Gov, optimized for national security uses, has fewer restrictions than regular versions. The Pentagon is threatening retaliation if Anthropic does not waive these restrictions by Friday, including invoking the Defense Production Act or declaring Anthropic a supply chain risk. Anth.. read more  

Link
@kala shared a link, 1 day, 9 hours ago
FAUN.dev()

Introducing helm

helm usesTypeScripttypes to registerskillsas typed functions with structured I/O. Permissions follow a clear precedence: exact→wildcard→skill→global. Agents get a keywordsearchtool and a code-execution tool that runs JS inside anSESsandbox. A recursiveproxyforwards calls overIPCto the parent, which .. read more  

Introducing helm
DeepSeekMath-V2 is a state-of-the-art mathematical reasoning model built on the DeepSeek-V3.2-Exp-Base architecture with 685 billion parameters. Unlike conventional math-focused language models that optimize only for correct final answers, DeepSeekMath-V2 introduces a self-verification framework where the model generates, inspects, and validates its own mathematical proofs.

This approach enables rigorous, step-by-step reasoning suitable for theorem proving, scientific research, and domains requiring high-integrity logic. The model is trained through a generation-verification loop involving a dedicated LLM-based verifier and reinforcement learning optimized for proof correctness rather than answer matching.

DeepSeekMath-V2 achieves gold-level scores on IMO 2025 and CMO 2024, along with a groundbreaking 118/120 on the Putnam 2024 contest. Released under the Apache 2.0 license and hosted on Hugging Face, it is fully open source for research and commercial use.