分类筛选
找到 438 个 Demo
GenEnv: Agent and Environment LLMs co-evolve toget...
GenEnv: Agent and Environment LLMs co-evolve together. New co-training framework from Princeton. Beats Gemini 2.5 Pro with 3.3x less data, +40.3% performance gain. Built on veRL. How it works? Environment LLM generates tasks at the agent's ~50% success boundary. Not too easy, not too hard. "Zone of proximal development" concept applied to LLM training. Tested on 5 benchmarks: API-Bank, ALFWorld, BFCL, Bamboogle, TravelPlanner. 7B models match or exceed larger model performance. Adaptive simulation instead of static data. "Smarter data generation" over "more data" approach. Code available: https://t.co/XJA7s6lGZM https://t.co/0ZNT7JfcTv
Vibe Coding Robotics Part 5 Implemented Another U...
Vibe Coding Robotics Part 5 Implemented Another UI Generated by Nano Banana Built with Gemini 3 https://t.co/KgvheE5ZlQ
NEW Chinese AI Agent is INSANE! ( FREE!) 🤯 This...
NEW Chinese AI Agent is INSANE! ( FREE!) 🤯 This update quietly changed open-source AI forever. MiniMax just dropped M2.1 — a 10B-activated OSS coding + agent model. • 72.5% SWE-multilingual • 88.6% VIBE-bench • Beats Gemini 3 Pro & Claude 4.5 Sonnet Built for real coding, real agents, real workflows. Save this. The agentic era just leveled up. Want the FULL GUIDE? DM ME!
哇塞！挖到一批Gemini 3新教程宝藏： 官方Cookbook上百实战、 YouTube建AI...
哇塞！挖到一批Gemini 3新教程宝藏： 官方Cookbook上百实战、 YouTube建AI代理30分钟速成、 隐藏功能深挖+UI设计神技，看完直接vibe coding起飞！ Flash练手，Pro征服多模态，2025 AI霸主非它莫属！速标安利！ #Gemini3 #AI代理
おはようございます☀️ ギンギラギンな土曜日です♪😃 Google AI Studio の Bu...
おはようございます☀️ ギンギラギンな土曜日です♪😃 Google AI Studio の Built で作ったアプリって、 従量課金制を有効にしても、 １日の生成数に上限があるんだって...🥲 なので今、先日公開した 『キャラクター表情チェンジャー v2.0』 のPCローカル版を作ってます♪😊✌️ ではでは、皆さんの今日が 過去イチ絶妙に幸せな一日でありますように♪😆💫💫🍀 #AIマンガ #AIイラスト #北斗の拳の日・世界法の日
我导出了19个币圈微信群聊天记录，用AI写了一个词频统计工具，之后将结果丢给AI生成为年度报告，得到...
我导出了19个币圈微信群聊天记录，用AI写了一个词频统计工具，之后将结果丢给AI生成为年度报告，得到了这样一张“2025年加密货币微信社群统计总结报告”。迭代了20几个版本，还有很多细节不完善，但电脑没电了，今年就这样。 #AI #Binance #Gemini3 https://t.co/t8zXausrlz
Last week we hosted a hundred builders at the new...
Last week we hosted a hundred builders at the new @GoogleDeepMind Singapore office for a vibe coding session with Google AI Studio and the Gemini API. Here are some of the cool things they built! 🧵 https://t.co/krc09u22Vd
Sentient just dropped their strongest upgrade yet...
Sentient just dropped their strongest upgrade yet They just unveiled SERA Crypto, a reasoning engine built specifically for crypto research and it’s already outperforming GPT5, Gemini 2.5 Pro, Grok 4, and Perplexity Finance. All while staying under 45 seconds per answer. That’s real time intelligence. What makes it special? *It understands intent, no confusion, no vague answers *It resolves tickers, no mixing up projects *It routes by embeddings, finds the right tools instantly *It pulls live data in parallel, markets, filings, TVL, OI, funding, whale flows *It loops when needed, deeper questions trigger deeper search *It ends with a grounded, well sourced answer On DMind’s Web3 benchmark, SERA is now the #1 open source agent. On real user queries, it ranks #1 in relevance and temporal accuracy, meaning it knows what matters now, not yesterday. From What’s happening in Solana DeFi today? to Who’s backing $ASTER? SERA delivers cleaner, tighter, better sourced answers than the big models. And every citation is clickable. The best part is @SentientAGI will open source the entire SERA framework, the benchmarks, and the technical deep dive over the coming months. This is how open source AI pushes past closed labs, not with all talk, but with superior reasoning built in public. Open source AI is winning.
Yes, I built this to rip ads. We all do it. 🏴☠️...
Yes, I built this to rip ads. We all do it. 🏴☠️ But then we starting playing at a higher level, creating some of the best creatives in the game but this feature is still useful. Now I use it to take take 1:1 winners and use Gemini 3 to make perfect creatives in the other sizes. https://t.co/E65heD9nTY
I'm carrying on my experiments making apps using A...
I'm carrying on my experiments making apps using AI to see whether it's feasible in the real world. This time I did a comparison, and one def turned out better. > Claude Opus 4.5 vs gemini 3.0 Pro > Same prompt > Built using Rork https://t.co/aR6clM82W7
Day 3 of building my fitness app 🎉 - Designed a...
Day 3 of building my fitness app 🎉 - Designed a 3 screen minimal onboarding flow - Built a dashboard to track daily progress All of this in just 1 hr using Gemini 3 Pro inside Anything. Honestly, I’m loving the process so far, and cannot wait to launch. https://t.co/GngwTvp4Gm
Been using Gemini 3.0 Flash and the ml-sharp model...
Been using Gemini 3.0 Flash and the ml-sharp model to experiment with 3D splat creation this weekend, and built a basic Mac app test with. A few folks have asked to try it out, so here's the link to the github: https://t.co/gewDFSYr54