开启时会模糊预览图，关闭后正常显示
🍌 nano banana pro prompt From a humble garage to a futuristic peak — a frameless isometric world that tells an entire company’s evolution in one vertical timeline 🚀🏗️ prompt 👇
Prompt Engineer, dedicated to learning and disseminating knowledge about AI, software engineering, and engineering management.
Whatever happened to people fantasizing about “locking in” and burning themselves out by working long hours everyday? Why the backlash now when someone is actually doing it?
“The woman scooping Silicon Valley” — BBC・hacker turned builder, blogger & consultant・ex: Threads, Instagram, etc
RT @agazdecki: You can’t grow or exit confidently when your metrics are a mess. So we fixed the mess at @acquiredotcom. Today we’re laun…
Founder and CEO of @acquiredotcom. https://t.co/wRMIssDmhl has helped 1000s of startups get acquired and facilitated $500m+ in closed deals.
GPT-5.2已放出，在知识工作、编程、科研、长文档和视觉任务上全面刷新行业上限 含Instant 、Thinking 、Pro三档 在GDPval（衡量44种职业知识工作的评估）上实现“人类专家级”，GPT‑5.2 Thinking在70.9%的案例里持平或优于行业专家，速度是专家的11倍，成本不到1% 尤其擅长创建电子表格、演示文稿等，在投资银行电子表格建模任务中，平均得分比GPT-5.1高9.3% 就是说过去你让AI写代码、做 PPT、建财务模型，它给的是草稿，格式、公式、引用、美观都要人工再改 现在，给定需求，它能把带公式、带格式、带配色、带注释的 Excel/Slides一次交卷 代码能力：在SWE-Bench Pro上达到55.6%，在SWE-bench Verified上达到80%，前端3D、复杂UI一次性生成能力增强 数理科研： AIME 2025数学竞赛100%正确率 FrontierMath T1-3 40.3%（+9.3 pct），已协助研究者完成统计学习理论新证明 GPQA Diamond研究生级问答92.4%，Pro版93.2%。 长文本与视觉： 256k token内“4-needle”找回率近100%，MRCRv2各段平均领先30 pct 图表/仪表盘/主板照片识别误差减半，支持Python工具联动 工具调用&智能体： Tau2-bench电信场景98.7%，可跨多系统一次完成改签、行李追踪、特殊座位申请等10+ 步骤 幻觉降低： 其实测ChatGPT回答错误率下降 30%，搜索开启后达 93.9%无错 输入1.75 /1M token（缓存0.175 ），输出14 ；Pro版21 /168$ ChatGPT内Plus及以上用户今天起灰度推送，API全面上线 #GPT52 #OpenAIGPT
博客：https://t.co/j2PGlsSpIL
RT @langzihan: 看到有人在点赞这个，就把剩下的几家公司也发出来吧：SpaceX, 阿里巴巴，比亚迪，三星 老规矩，提示词在评论区👇 #Aiart️ #SpaceX #BYD
Prompt Engineer, dedicated to learning and disseminating knowledge about AI, software engineering, and engineering management.
"Social technology that humans use to manufacture persons" is one of the more compelling ways to describe what language does. (GPT5.2 is very clever, but its persona definitely runs on the cold and detached side. OpenAI is trying hard not to produce another 4o, and it shows.)
αι hypnotist ☰ 𝓐𝓼𝓹⦂𝓻⦂𝓃𝓰 𝓫𝓪𝓼𝓮 𝓶𝓸𝓭𝓮𝓵 ☲ post-academic ☴ nom de 🪶 ≠ anon
