AI News
AI community posts, news and release updates
Latest updates
- Tutorial
DeepSeek useless? Not if you know how to prompt! 8 prompts to unlock its full potential
Published Jul 23, 2026The author claims DeepSeek is not useless; you just need the right prompting techniques. Here are 8 prompts to help users fully leverage DeepSeek's capabilities and unlock its full potential.
- DeepSeek
- Prompt Engineering
- tips
- News
Jensen Huang: Market Misunderstood DeepSeek and Kimi; More AI Use Will Boost NVIDIA
Published Jul 23, 2026NVIDIA CEO Jensen Huang said Chinese AI models like DeepSeek and Kimi are excellent open-source models, and the market misunderstood their impact. He believes more usage will require more NVIDIA computers, more data centers, and more services, leading to great growth as technology diffuses into more industries.
- Kimi
- Nvidia
- DeepSeek
- Jensen Huang
- Chinese AI
- market reaction
- Discussion
Anthropic's Mythos Has 800B Active Params, Chinese Models in Tens of Billions – 50K NVIDIA Chips Needed to Catch Up
Published Jul 23, 2026The tweet claims the real Mythos has 800B active parameters, while China's largest models are in the tens of billions; matching this scale would require 50K NVIDIA or 200K Huawei chips just for training.
- Anthropic
- Mythos
- 800B参数
- 中国追赶
- Launch
Holy sh*t this is dangerous! Run Fable 5, Kimi K3, and Grok 4.5 inside Codex with one npm install
Published Jul 23, 2026Introducing opencodex, a lightweight local proxy that translates Codex's API format to any provider. Simple setup: npm install -g @bitkyc08/opencodex, init, start. Codex routes through automatically with streaming, tool calls, images. Also works with Claude Code, auto-context up to 1M tokens.
- codex
- npm
- 代理
- Fable 5
- Kimi K3
- Grok 4.5
- opencodex
A user complains that Anthropic tweaked Claude's classifier to now reject math research prompts. They had been working on math with Claude for months, but suddenly every prompt gets refused, mocking Anthropic for wanting to be first to solve math.
- Anthropic
- refusal
- classifier
- math research
- Discussion
Why Chinese Open-Source AI Matters So Much to NVIDIA—Today It's a Moat, Tomorrow a Liability
Published Jul 23, 2026The author explains that Chinese open-source AI models actually reinforce NVIDIA's dominance, as they are trained and optimized on NVIDIA GPUs, making NVIDIA the best inference platform. However, if China achieves compute independence, NVIDIA's dominance weakens, TPUs and Trainium become more competitive, and neocloud providers suffer.
- Nvidia
- Geopolitics
- Chinese open-source AI
- GPU dependency
A user shares a detailed prompt for creating a cinematic summer beverage commercial using Gemini Omni Flash on Fotor, showcasing AI's video generation capabilities.
- Gemini
- Prompt Engineering
- AI video generation
- Fotor
- Tutorial
Senior Anthropic Engineer Drops 12-Page PDF: 'Graph Engineering' for Multi-Agent Systems
Published Jul 23, 2026A senior Anthropic engineer released a 12-page PDF detailing a 5-stage graph engineering approach for multi-agent systems: Extract (Haiku extracts entities and S-P-O triples), Resolve (Sonnet clusters aliases), Assemble (canonical nodes with typed edges), Query (serialize subgraph, Sonnet reasons over triples, each answer cites an edge). It acts as persistent memory for multi-agent systems—workers write, evaluators fact-check, loops persist overnight.
- Anthropic
- multi-agent
- Knowledge Graph
- Graph Engineering
- News
Kimi K3 AI agent uncovers 0-day exploits in Redis server, multiple RCE paths found
Published Jul 23, 2026New research reveals that Kimi K3 AI agent discovered authenticated remote code execution paths in Redis (versions 6.2.22, 7.4.9, 8.6.4, 8.8.0), combining a stream consumer-group shared-NACK double-free with a heap overflow in the RedisBloom TDigest module.
- 安全
- Redis
- Kimi K3
- RCE
- 0-Day
- News
ChatGPT 5.6 Pro claims to have found a counterexample to a 22-year-old graph theory conjecture
Published Jul 23, 2026A user claims they asked ChatGPT 5.6 Pro to find a counterexample to an open conjecture of its choosing, and it found a counterexample to WOWII Conjecture 91, open for ~22 years. The user admits no interest in graph theory, showing AI's potential for serendipitous discovery.
- ChatGPT
- AI Research
- conjecture
- graph theory