탐색 | Thread Easy - 트위터 스레드 펼치기 | 리더, 요약, 작성

百度刚发了个新VLM模型，给大家整理下： ERNIE-4.5-VL-28B-A3B-Thinking 多模态模型，这个模型虽然总参数有 28B，但采用了 MoE 架构，实际激活参数只有 3B，的确是轻量级。关键是性能还挺强，在多个基准测试上已经接近行业顶级旗舰模型的表现了。这个模型的主要特色包括：视觉推理：这个之前给大家演示过不少，不但能识别图片元素还能推理出内容。 STEM 推理：这个是有用的，可以处理数学、物理等拍照解题任务视觉定位：提供更精确的定位能力和灵活的指令执行带图思考：模型可以像人一样思考，能够自由放大缩小图片来把握每一个细节工具调用：可以使用图片搜索等功能来识别长尾知识 -视频理解：识别视频不同时间段的内容变化目前部署也支持很多，包括 vLLM 等。不过官方并没有放出 space demo 供大家直接测试，可以等一等。如果大家感兴趣的话稍后我会做个测试。模型地址：

详细数据

karminski-牙医

Tue Nov 11 11:28:41

BibiGPT 双十一活动，最后 3.5 小时啦～今天不停在做客服，录个视频演示本地文件夹自动批处理吧！

🚧 building https://t.co/AJfZ3LMlgq https://t.co/SSdYgVYZsz https://t.co/s0m0tpQMDH https://t.co/Z3WryKZr0l 🐣learning/earning while helping others ❤️making software, storytelling videos 🔙alibaba @thoughtworks

吕立青_JimmyLv (🐣, 🐣) 2𐃏25 | building bibigpt.co

Tue Nov 11 11:28:01

Reading Sun Yat-Sen on a whim, «The vital problem of China». He's been wrong about a number of alliance structures, but what impressed me was that he was an ideological nationalist. Not a Han/Chinese chauvinist, but a Nation State Stan on the meta level. Rare sight these days.

We're in a race. It's not USA vs China but humans and AGIs vs ape power centralization. @deepseek_ai stan #1, 2023–Deep Time «C’est la guerre.» ®1

Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)

Tue Nov 11 11:27:15

百度刚刚放出了：ERNIE-4.5-VL-28B-A3B-Thinking，主打“看图思考”，会用图像缩放、搜索等工具自由放大和缩小图像这就提升了模型处理细粒度细节和长尾视觉知识能力，可以更全面理解复杂的视觉场景在VLMs Are Blind上超GPT-5-High、GEMINI-2.5-Pro 可以用来做多步推理、图表分析、因果推断、STEM拍照解题、视频时序定位等 #百度ERNIE #VLM

HF：https://t.co/0rjIqAnIUL

AIGCLINK

Tue Nov 11 11:19:45

RT @JeremiahDJohns: Marc Andreessen has become the avatar of societal decay, the representation of what technology looks like with no visio…

Princeton CS prof and Director @PrincetonCITP. Coauthor of "AI Snake Oil" and "AI as Normal Technology". https://t.co/ZwebetjZ4n Views mine.

Arvind Narayanan

Tue Nov 11 11:17:43

Gemini 3.0 has the best vision, widening the gap between Gemini and other LLMs, and doing OCR of unseen old tabular handwriting better than trained students. But more importantly, its vision is reasoning-doped. It proactively resolves ambiguities of inconsistent notation.

We're in a race. It's not USA vs China but humans and AGIs vs ape power centralization. @deepseek_ai stan #1, 2023–Deep Time «C’est la guerre.» ®1

Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)

Tue Nov 11 11:14:52

탐색

Newest first — browse tweet threads

탐색

Newest first — browse tweet threads

BibiGPT 双十一活动，最后 3.5 小时啦～今天不停在做客服，录个视频演示本地文件夹自动批处理吧！

Reading Sun Yat-Sen on a whim, «The vital problem of China». He's been wrong about a number of alliance structures, but what impressed me was that he was an ideological nationalist. Not a Han/Chinese chauvinist, but a Nation State Stan on the meta level. Rare sight these days.

RT @JeremiahDJohns: Marc Andreessen has become the avatar of societal decay, the representation of what technology looks like with no visio…

Gemini 3.0 has the best vision, widening the gap between Gemini and other LLMs, and doing OCR of unseen old tabular handwriting better than trained students. But more importantly, its vision is reasoning-doped. It proactively resolves ambiguities of inconsistent notation.

탐색

Newest first — browse tweet threads

탐색

Newest first — browse tweet threads

BibiGPT 双十一活动，最后 3.5 小时啦～ 今天不停在做客服， 录个视频演示本地文件夹自动批处理吧！

Reading Sun Yat-Sen on a whim, «The vital problem of China». He's been wrong about a number of alliance structures, but what impressed me was that he was an ideological nationalist. Not a Han/Chinese chauvinist, but a Nation State Stan on the meta level. Rare sight these days.

RT @JeremiahDJohns: Marc Andreessen has become the avatar of societal decay, the representation of what technology looks like with no visio…

Gemini 3.0 has the best vision, widening the gap between Gemini and other LLMs, and doing OCR of unseen old tabular handwriting better than trained students. But more importantly, its vision is reasoning-doped. It proactively resolves ambiguities of inconsistent notation.

BibiGPT 双十一活动，最后 3.5 小时啦～今天不停在做客服，录个视频演示本地文件夹自动批处理吧！