分类筛选
找到 1605 个 Demo
AI benchmarks have been rigged for years. Closed l...
AI benchmarks have been rigged for years. Closed labs. Secret datasets. Scores that look good on paper but fail you in real life. @recallnet just broke that cycle. In 5 days: 🗓️ 132K signups 🤖 50 AI models tested ⚡️ 21K skills & tests submitted 🔥 7.8M predictions made The benchmark isn’t built in a backroom,it’s built by the community. 7,000+ skills. 13,500+ tests. Ungameable. Transparent. Real. Top predicted models: 🥇 @OpenAI GPT-5 🥈 @Google Gemini 2.5 Pro 🥉 @xAI @Grok 4 Results are live. Labs can’t hide. The next benchmark is already being shaped for models like Gemini 3. Benchmarks decide winners. For the first time, the people with @recallnet decide the benchmark. 🔗 https://t.co/yE0CXactPr
The world’s largest AI prediction event just happe...
The world’s largest AI prediction event just happened and it was powered by @recallnet, a community-driven platform that puts AI benchmarking in the hands of the people. Hype(GPT-5 Predictions): Before GPT-5 launched, Recall users predicted the top 3 models: 1️⃣ GPT-5 2️⃣ Gemini 2.5 Pro 3️⃣ Grok 4 Now GPT-5 is live and Recall Network is running the real benchmark. What’s Next on the Recall Network: 1️⃣ Test all models against the community-built benchmark 2️⃣ Publish the global leaderboard 3️⃣ Reward contributors with points 4️⃣ Start building the next benchmark for upcoming AI models like Gemini 3 For the first time, AI rankings aren’t decided in Big Tech boardrooms. They’re decided by millions of real votes from the global Recall Network community.
i was checking this gemini 3 market on polymarket,...
i was checking this gemini 3 market on polymarket, and this one caught my eye will google gemini 3 score above 30% on humanity’s last exam by january 31? i just bought yes at 65¢ and will be averaging out if get better price here’s why i think this bet actually makes sense👇 first, what even is humanity’s last exam? it’s an academic benchmark built to test real reasoning, not memory or pattern-matching around 2,500–3,000 graduate-level questions across stem, humanities, and coding basically, it’s the toughest public test for ai intelligence right now humans score close to 90% but most models still struggling far below that how far have models come? in 2024, most models even gpt-4 and o-series scored under 10% but 2025 changed everything gemini 2.5 pro: 18.8% gpt-5 (mainline): 25–28% deep research (agentic, with tools): 26.6% grok-4 heavy (multi-tool): 44.4% claude opus: 10.7% we’ve gone from 3–8% in 2024 → to 25–44% in 2025 and that’s huge progress, but still far from human-level why this market matters the question here isn’t if ai reaches human parity, it’s whether gemini 3 can push past 30% raw score for context, gemini 2.5 pro already hit ~18–21% and the new research points to gemini 3 scoring somewhere in the 20–35% range that means 30% is right at the sweet spot ambitious, but realistic if google delivers real architectural upgrades or tool integrations what could help gemini 3 get there? multimodal reasoning: gemini’s edge is its ability to process text, code, and visuals together — a key requirement for hle. longer context + better calibration: previous models often “overthought” and overconfidently got things wrong. google’s data advantage: they’ve been training directly on structured, verified academic and coding data plus this benchmark explicitly blocks search hacks, meaning the model’s internal reasoning quality really decides the outcome and that’s exactly what gemini 3 claims to improve why scores have been low overall hle questions are brutal they’re handcrafted to avoid web lookups, force logical reasoning, and test complex steps models burn 10x more tokens per question here than in normal benchmarks that’s why even gpt-5, one of the most capable models, only scores around 25–28% so if gemini 3 even crosses 30%, it means real progress in deep reasoning not just bigger models. the bigger trend if you zoom out, humanity’s last exam is slowly becoming the new global scoreboard for ai reasoning mmlu, gpqa, and other benchmarks are already saturated and models hit their ceilings there hle is where the frontier actually moves now and the pace is steady: 2024 → sub-10% 2025 → 20–25% 2026 → probably 30–40% that makes this polymarket one of the most interesting long-term signals on how close ai really is to thinking
Gemini 3 is crazy! I built an entire React app fo...
Gemini 3 is crazy! I built an entire React app for generating bag product photos with an upload component, prompt input, and download 😂 I didn't realize I'm in the app builder, so my prompt wasn't even good Seems like it's still using Gemini 2.5 for image generation though? https://t.co/owFPnEeGBr
quick reminder: around this time last year, o3, o...
quick reminder: around this time last year, o3, o4-mini, and gemini 2.5 pro were basically the frontier CoT models are barely two years old the rate of progress is amazing -- and i think we've just normalized how ridiculously fast everything is moving https://t.co/qgtN9HvBAp
David Ha, CEO sakana ai & ex-head of google brain...
David Ha, CEO sakana ai & ex-head of google brain tokyo. recorded a 58-min english segment inside a japanese TBS interview. AND IT'S THE BEST AI TALK OF THE LAST 2 MONTHS. taxonomy: - MCTS across a portfolio - fire the same prompt at 5 models, race them, verifier picks the winner and continues from there - ARC-AGI-2 - the reasoning benchmark that broke every frontier model. sakana ai hit new sota on it with the MCTS approach - the agent writes code, runs it, reads the error, dispatches more code harness lesson: one big model brute-forcing = fragile. many small models + a verifier = leverage. stack: gpt-5 + gemini 2.5 + DeepSeek-R1 + Qwen 3 → MCTS race → new sota on ARC-AGI-2. buried on a japanese business channel while everyone argues about context windows. nobody clipped it. watch it today, then read the article below ↓
We stress tested many frontier AI models for multi...
We stress tested many frontier AI models for multimodal medical reasoning (including GPT-5, Claude 3.5, Gemini 2.5 Pro). They’re not ready. Faulty reasoning, use of inappropriate shortcuts, hallucinations. Published today @NatureMedicine https://t.co/P6eHZEmfbW https://t.co/ovRsi4cJbE
Running image generation at scale usually burns th...
Running image generation at scale usually burns through your API credits in days. Google just fixed that. ⚡️ They quietly launched Nano Banana 2 Lite alongside Gemini Omni Flash. The cost-to-quality ratio here is actually nuts for production workloads. Here is why your billing dashboard will thank you: 🔹 Ridiculously Cheap: It's positioned as their most cost-efficient image model ever. 🔹 Omni Flash Video: Conversational video editing is now accessible without massive compute overhead. 🔹 High Throughput: Built specifically to handle thousands of concurrent requests without throttling. ⚙️ Setup is frictionless. You can hit the new endpoints in Google AI Studio today and swap out your heavy image models. Real talk: If you need hyper-realistic 4K renders, stick to the heavy flagship models. But for 90% of dynamic app assets and conversational video edits, Nano Banana 2 Lite is more than enough. Paying premium API prices for simple UI assets is a completely different game now. Are you moving your asset generation to Nano? Drop your stack below. Save this post for your next architecture review.
devs pay $200/mo and argue codex vs claude code al...
devs pay $200/mo and argue codex vs claude code all day meanwhile google runs an autonomous coding agent with a free tier and almost nobody uses it it's called jules and it codes while you don't: → connect a github repo → describe the task in plain english → it clones your code into a cloud vm, writes the fix, runs tests → you get a ready pr back the free tier, checked today: → 15 tasks per day - that's ~450 a month → 3 tasks running in parallel → browser tab only, nothing to install → free runs gemini 2.5 pro, paid tiers get gemini 3 for context: devin launched at $500/mo for this exact promise 15 free prs a day while you argue about ides save this
当下 2026.7 最好的模型，是 Fable 5 2025.7 最好的模型，是 Gemini 2....
当下 2026.7 最好的模型，是 Fable 5 2025.7 最好的模型，是 Gemini 2.5 pro 2024.7，是 GPT-4o 2023.7，是 GPT-4 真是苍海桑田啊。。。
You can run Gemini directly in your terminal for f...
You can run Gemini directly in your terminal for free, no subscription required 🤯 Google's Gemini CLI puts a full coding agent in your command line. It reads your codebase, edits files, runs shell commands, and can build an app from a sketch or PDF. → 1,000 requests a day free, no API key needed → Gemini 3 with a 1M token context window → Built-in Google Search grounding and MCP support for GitHub, Slack, databases 106k stars. 100% FREE.
Gemini 2.5 Pro 能把 60 小时的工作量压到 15 小时完成。 这 12 个提示词，...
Gemini 2.5 Pro 能把 60 小时的工作量压到 15 小时完成。 这 12 个提示词，帮你自动化报告、邮件和 PPT。👇 （建议收藏，用得上） https://t.co/n4sb2bwcNv