LogoThread Easy
  • Explorer
  • Composer un thread
LogoThread Easy

Votre partenaire tout-en-un pour les threads Twitter

© 2025 Thread Easy All Rights Reserved.

Explorer

Newest first — browse tweet threads

Keep on to blur preview images; turn off to show them clearly

我以为我很懂 prompt engineering , 直到我想完成一系列关于ai 智能体相关技术文章时, 我才明白, 我对prompt 工程也只是一知半解。

---
GPT-2 拥有 15 亿个参数,相比之下,GPT 只有 1.17 亿个参数;GPT-2 的训练数据规模为 40GB 文本,而 GPT 仅为 4.5GB。模型规模和训练数据规模的这种数量级提升,带来了一种前所未有的涌现能力(emergent quality):研究人员不再需要为单一任务对 GPT-2 进行微调,而是可以直接将未经微调的预训练模型应用到具体任务上,并且在很多情况下,其表现甚至优于那些为该任务专门微调过的最先进模型。

GPT-3 在模型规模和训练数据规模上又实现了一次数量级的提升,并伴随着能力上的显著飞跃。2020 年发表的论文 《Language Models Are Few-Shot Learners》 表明,只要给模型提供少量任务示例(即所谓的 few-shot examples),模型就能够准确复现输入中的模式,从而完成几乎任何你可以想象的、基于语言的任务——而且往往能够得到质量非常高的结果。正是在这一阶段,人们意识到:通过修改输入内容——也就是提示(prompt)——就可以对模型进行条件约束,使其执行所需的特定任务。提示工程(prompt engineering)正是在这一时刻诞生的。

---

我以为我很懂 prompt engineering , 直到我想完成一系列关于ai 智能体相关技术文章时, 我才明白, 我对prompt 工程也只是一知半解。 --- GPT-2 拥有 15 亿个参数,相比之下,GPT 只有 1.17 亿个参数;GPT-2 的训练数据规模为 40GB 文本,而 GPT 仅为 4.5GB。模型规模和训练数据规模的这种数量级提升,带来了一种前所未有的涌现能力(emergent quality):研究人员不再需要为单一任务对 GPT-2 进行微调,而是可以直接将未经微调的预训练模型应用到具体任务上,并且在很多情况下,其表现甚至优于那些为该任务专门微调过的最先进模型。 GPT-3 在模型规模和训练数据规模上又实现了一次数量级的提升,并伴随着能力上的显著飞跃。2020 年发表的论文 《Language Models Are Few-Shot Learners》 表明,只要给模型提供少量任务示例(即所谓的 few-shot examples),模型就能够准确复现输入中的模式,从而完成几乎任何你可以想象的、基于语言的任务——而且往往能够得到质量非常高的结果。正是在这一阶段,人们意识到:通过修改输入内容——也就是提示(prompt)——就可以对模型进行条件约束,使其执行所需的特定任务。提示工程(prompt engineering)正是在这一时刻诞生的。 ---

独立开发者 | 个人IP教练 | 帮助新手在X上完成早期成长| 公众号:PandaTalk8

avatar for Mr Panda
Mr Panda
Tue Dec 16 07:55:13
兄弟们,这个AI硬件创意好

Stickerbox:用AI把孩子的想象力“打印”出来!

🎤 语音输入 → AI自动绘图 → 🖨️ 即时打印贴纸!

帮孩子把脑海里的故事变成现实!

✅ 无需墨水
✅ 无屏幕更安全
✅ 可上色、可收藏

让AI帮孩子把想象“贴”出来!

🧒 儿童安全模式,无屏幕交互!
🌿 环保纸材,无BPA / 无BPS

兄弟们,这个AI硬件创意好 Stickerbox:用AI把孩子的想象力“打印”出来! 🎤 语音输入 → AI自动绘图 → 🖨️ 即时打印贴纸! 帮孩子把脑海里的故事变成现实! ✅ 无需墨水 ✅ 无屏幕更安全 ✅ 可上色、可收藏 让AI帮孩子把想象“贴”出来! 🧒 儿童安全模式,无屏幕交互! 🌿 环保纸材,无BPA / 无BPS

接入 Banana 感觉非常的不错 🥲 还可以搞增值服务

avatar for 小互
小互
Tue Dec 16 07:51:56
Fast 2DGS: Efficient Image Representation with Deep Gaussian Prior

Contributions:
• We propose Deep Gaussian Prior, an initialization strategy learned through an iterative optimization-sampling loop. By simulating the optimization trajectory, our method captures a content-aware distribution that breaks the uniform bias of random initialization, significantly accelerating convergence.

• We present a streamlined framework that anchors Gaussian cardinality (K) to compression rates. By employing a lightweight backbone without complex feature engineering, we transform the ill-posed Gaussian allocation into a tractable, batch-parallelizable learning task.

• We demonstrate that our framework achieves superior trade-offs between reconstruction quality, inference latency, and cross-dataset generalizability compared to existing Gaussian image approaches.

Fast 2DGS: Efficient Image Representation with Deep Gaussian Prior Contributions: • We propose Deep Gaussian Prior, an initialization strategy learned through an iterative optimization-sampling loop. By simulating the optimization trajectory, our method captures a content-aware distribution that breaks the uniform bias of random initialization, significantly accelerating convergence. • We present a streamlined framework that anchors Gaussian cardinality (K) to compression rates. By employing a lightweight backbone without complex feature engineering, we transform the ill-posed Gaussian allocation into a tractable, batch-parallelizable learning task. • We demonstrate that our framework achieves superior trade-offs between reconstruction quality, inference latency, and cross-dataset generalizability compared to existing Gaussian image approaches.

Paper: https://t.co/TmsaDVwcxr

avatar for MrNeRF
MrNeRF
Tue Dec 16 07:47:03
RT @dotey: 昨天发的 AI 视频制作方法,有朋友给我留言,说还是太复杂了点,问我有没有简单一点的办法。

在创作过程中,像写剧本、分镜头脚本、画首帧图片这些步骤,确实比较耗时间。可以借助 飞书的 aily 工作助手这样的 Agent 来自动化这些步骤,先让 Agent…

RT @dotey: 昨天发的 AI 视频制作方法,有朋友给我留言,说还是太复杂了点,问我有没有简单一点的办法。 在创作过程中,像写剧本、分镜头脚本、画首帧图片这些步骤,确实比较耗时间。可以借助 飞书的 aily 工作助手这样的 Agent 来自动化这些步骤,先让 Agent…

独立开发者 | 个人IP教练 | 帮助新手在X上完成早期成长| 公众号:PandaTalk8

avatar for Mr Panda
Mr Panda
Tue Dec 16 07:42:06
School is not the place to learn knowledge. If you (or your kids) rely on the education system to learn new things, you're not going anywhere.

It is only good for social reasons, best scenario, you get some guidance and inspirations.

You have to learn knowledge on your own.

School is not the place to learn knowledge. If you (or your kids) rely on the education system to learn new things, you're not going anywhere. It is only good for social reasons, best scenario, you get some guidance and inspirations. You have to learn knowledge on your own.

Creating software I love to use. 🧠 https://t.co/p4T2vFZoJ1 $137K/m 🧰 https://t.co/y0Lq4RQRsu $5K/m 📕 https://t.co/btuasMBHPT $518/m 🖼️ https://t.co/KfFdieGrVf $50/m

avatar for Tony Dinh
Tony Dinh
Tue Dec 16 07:41:31
School is not the place to learn knowledge. If you (or your kids) rely on the education system to learn new things, you're not going anywhere.

It is only good for social reasons, best scenario, you get some guidance and inspirations.

You have to learn knowledge on your own.

School is not the place to learn knowledge. If you (or your kids) rely on the education system to learn new things, you're not going anywhere. It is only good for social reasons, best scenario, you get some guidance and inspirations. You have to learn knowledge on your own.

Creating software I love to use. 🧠 https://t.co/p4T2vFZoJ1 $137K/m 🧰 https://t.co/y0Lq4RQRsu $5K/m 📕 https://t.co/btuasMBHPT $518/m 🖼️ https://t.co/KfFdieGrVf $50/m

avatar for Tony Dinh
Tony Dinh
Tue Dec 16 07:41:31
  • Previous
  • 1
  • More pages
  • 640
  • 641
  • 642
  • More pages
  • 5634
  • Next