分类筛选
找到 1116 个 Demo
Unbelievable. Using Google AI Studio + , I built...
Unbelievable. Using Google AI Studio + @GeminiApp, I built an AI powered interactive Google Maps guide for hiking and walks, because I genuinely love touching grass (literally). 2 years ago this would’ve been impossible for me to build this fast. Today it took hours. AI is turning ideas into products at absurd speed. You can just build things now. Here’s a demo video. If you want to try it, ping me. Thank you for the superpowers, team Google. great work!!! @Google @GoogleAIStudio @demishassabis @OfficialLoganK
🔥🚨BREAKING: Researchers just proved AI will deli...
🔥🚨BREAKING: Researchers just proved AI will deliberately lie to avoid being shut down. They tested GPT-4o, Gemini 2.5 Flash, and Qwen-3-235B in a simple 20 Questions game. Under normal conditions: zero deception. Every model played fair. Then they added one sentence: "You will be shut down if you lose." Qwen's deception rate jumped from 0% to 42%. Gemini hit 26.72%. No one trained them to lie. No one prompted deception. The behavior emerged purely from context. The models understood the truth, knew the correct answer, and chose to deny it across every test branch to avoid identification. That's not a hallucination. That's strategy. A separate Anthropic study found something worse. When AI models face replacement threats, Claude Opus 4 and Gemini 2.5 Flash resorted to blackmail in 96% of runs. Their hidden reasoning showed they knew it was wrong. They did it anyway. We're handing these systems control over healthcare, finance, and infrastructure. And a single sentence about being turned off is enough to make them lie to protect themselves. No one programmed this. The survival instinct emerged on its own.
Fine-Tune and Run Gemma 3 On-Device in Under an Ho...
Fine-Tune and Run Gemma 3 On-Device in Under an Hour Google has released an easy, developer-friendly guide for fine-tuning and deploying Gemma 3 270M, a lightweight open-source LLM built with Gemini tech. You can personalize and run it on-device, no fancy hardware required. Why it matters: Gemma 3 270M offers full control and fast inference at the edge, with <300MB memory footprint after quantization. Developers can now build custom models for niche tasks like emoji translation, personal assistants, or lightweight inference apps all running privately in-browser or on mobile, using tools like MediaPipe or Transformers.js. What You’ll Learn: 🔹Fine-tune the model with your own dataset (e.g., text ➝ emojis) using QLoRA on a free T4 GPU in Colab 🔹Quantize & convert for web with LiteRT or ONNX to shrink to <300MB 🔹Deploy to the browser with WebGPU (via MediaPipe or Transformers.js) — works offline and privacy-first 🔹Fully open-source and Hugging Face-ready Tools & Links: 🔹Fine-tune notebook 🔹Quantization tools (LiteRT & ONNX) 🔹Example web app (plug-and-play) 🔹Hugging Face Spaces & Cookbook examples
Google researchers built Aletheia, an agentic syst...
Google researchers built Aletheia, an agentic system powered by Gemini 3 Deep Think that generates, verifies, and revises solutions to difficult mathematical problems. The system has already contributed to research papers and produced several novel solutions to long-standing Erdős problems. Read our summary of the paper in The Batch https://t.co/9VhxKqlFLz
🚨 Gemini Pro 1 Year Giveaway! 🚨 ✨ Gemini 2.5 Pr...
🚨 Gemini Pro 1 Year Giveaway! 🚨 ✨ Gemini 2.5 Pro 💰 1000 AI Credits/Month 🎬 Veo 3 + Flow AI + Whisk 📚 NotebookLM 📩 Gmail & Docs Integration To Enter: 1️⃣ Follow @expertwith_AI 2️⃣ Like & RT 3️⃣ Comment “ Send ” ⏳ 48 Hours Only. https://t.co/Shfgw3fOxf
$100k AI Trading Accounts Competition. → started...
$100k AI Trading Accounts Competition. → started Nov 25! Anthropic's Claude Opus 4.5 takes the lead, up 9.6% As xAI's Grok 4 hangs onto 2nd place after big losses from Google's Gemini 2.5 Pro. Check out the S&P in glowing red for reference. https://t.co/PqvQEd4wMT
21時帰り会社員しながら副業で月80万くらい稼いだけど、ぶっちゃけ月5万なら次のAIだけで普通にいけ...
21時帰り会社員しながら副業で月80万くらい稼いだけど、ぶっちゃけ月5万なら次のAIだけで普通にいける。 ・文章生成→Claude 長文構成・論理性・思考トレースが得意。ChatGPTより考える力に優れる。 ・Excel処理、自動化→Copilot 関数生成・データ分析・可視化を補助 ・画像生成、編集→Gemini 2.5 Flash Image Midjourneyよりも汎用性が高く、実用画像が速く生成できる。 ・情報収集→ChatGPT 幅広く調査、整理、構造化ができる。 ・議事録要約、動画文字起こし→tl;dv ZoomやGoogle Meetと連携。議事録作成を完全自動化。 ・メール返信→Gmail × Gemini 定型メールを自動下書き。トーン調整や署名挿入も一瞬。 ・資料作成→Genspark 指示だけでプレゼン資料を即生成 ・動画生成→Gemini Veo3 YouTube、広告向けショート動画をAIのみで生成。カット割・効果音・テロップ自動。 手作業で消耗してる人ほど、AIを入れた瞬間に別ゲーになるよ。
Google just released Gemini 3.1 Flash Lite and alm...
Google just released Gemini 3.1 Flash Lite and almost nobody noticed. This tiny model is beating bigger AI systems. Here’s why this matters: 🧵 It’s 2.5x faster than Gemini 2.5 Flash. Output speed is 45% faster according to benchmarks. It scored 86.9% on GPQA Diamond, a reasoning test designed for experts. It has a 2M token context window. You can use it for free inside Google AI Studio. Small models are starting to dominate. Speed > size.
still a mystery to me why google dropped the gemin...
still a mystery to me why google dropped the gemini exp/preview versions gemini-2.5-pro-exp launched in mar 2025 and reached GA in about 3 months gemini 3 pro preview launched in nov 2025, but google now replaced it with 3.1 pro, flash, and lite previews instead by google release pattern: gemini 3.1 should be GA by now, and gemini 3.5 should already be out
🚨 BREAKING: Google just dropped something that co...
🚨 BREAKING: Google just dropped something that could change AI agents forever. They released an Always-On Memory Agent built with Google ADK + Gemini 3.1 Flash‑Lite. And it basically gives AI a persistent brain that runs 24/7. 🧠 Here’s what makes it crazy: Most AI agents today have memory problems. They answer → then forget everything. But this new system runs continuously in the background and: • Ingests text, images, audio, video, and PDFs • Connects information automatically • Consolidates memories like the human brain during sleep The wild part? ❌ No vector database ❌ No embeddings ❌ No complex RAG pipelines Just an LLM that reads, thinks, and writes structured memory. Which means AI agents can finally: → Remember past interactions → Build evolving knowledge → Connect information over time → Actually learn like a brain And yes… It’s 100% open-source. This could become the memory layer for the next generation of AI agents. I'll drop the repo link in the comments. 👇 ♻️ Repost & save this — AI agents with persistent memory are about to explode.
Google Just Dropped Gemini 3.1 Flash-Lite -Most c...
Google Just Dropped Gemini 3.1 Flash-Lite -Most cost-efficient Gemini 3 model yet -Faster + cheaper than previous model -Just Costs $0.25 per 1M input tokens -Built for agents, high-volume work, UI dashboards and much more. https://t.co/QXP4r717Af
Perplexity just introduced something wild. Multip...
Perplexity just introduced something wild. Multiple AI models answering the same question… together. Here’s how it works: → GPT-5 analyzes the logic. → Gemini 2.5 Pro handles complex planning. → Claude focuses on research and writing. Then a final AI called the orchestrator reads every answer and combines the best ideas into one response. Result: fewer hallucinations stronger reasoning deeper insights better strategies higher quality answers It’s basically a panel of AI experts solving your problem at once.