AIGC周刊 | 第15期

时间： 2023.2.13-2023.2.19

最新技术：

• Scaling Vision Transformers to 22 Billion Parameters https://arxiv.org/pdf/2302.05442.pdf

• JEIT: Joint End-to-End Model and Internal Language Model Training for Speech Recognition

  https://arxiv.org/abs/2302.08583

• Massively Multilingual Shallow Fusion with Large Language Models

  https://arxiv.org/abs/2302.08917

• MultiDiffusion: Fusing Diffusion Paths for Controlled Image Generation https://arxiv.org/abs/2302.08113

• 3D-aware Conditional Image Synthesis

  https://cs.cmu.edu/~pix2pix3D/

• Augmented Language Models: a Survey

  https://arxiv.org/abs/2302.07842

• Text-driven Visual Synthesis with Latent Diffusion Prior

  https://latent-diffusion-prior.github.io

• Adding Conditional Control to Text-to-Image Diffusion Models https://arxiv.org/abs/2302.05543

  https://github.com/lllyasviel/ControlNet

• A Survey on Efficient Training of Transformers

  https://arxiv.org/abs/2302.01107

• VQ3D: Learning a 3D-Aware Generative Model on ImageNet https://kylesargent.github.io/vq3d

• ConceptFusion: Open-set Multimodal 3D Mapping

  https://concept-fusion.github.io

• Single Motion Diffusion

  https://sinmdm.github.io/SinMDM-page/

• Level Generation Through Large Language Models

  https://arxiv.org/abs/2302.05817

• 万字长文解析！复现和使用GPT-3/ChatGPT，你所应该知道的 https://mp.weixin.qq.com/s/ILpbRRNP10Ef1z3lb2CqmA

• 大语言模型: 新的摩尔定律？

  https://mp.weixin.qq.com/s/RP5JBsdBClRHSBSa_vGKtQ

• GPT版超级马里奥来了！输入文本即可自定义游戏关卡

  https://mp.weixin.qq.com/s/DgaLyo9VSxGOe-mE_DuNAw

• AI降维打击人类画家，文生图引入ControlNet，深度、边缘信息全能复用

  https://mp.weixin.qq.com/s/vsvXgqhy1lE_RC5SYtRSow

• ChatGPT 在做什么… 以及它为何发挥作用？ https://mp.weixin.qq.com/s/Cf9TF4OTEFoPQMOkwRNe6g

• 配乐行业危，谷歌出手解决音乐伴奏问题，人均音乐人不远了

  https://mp.weixin.qq.com/s/yL0Uw71J6q6YTNf98heOMw

• 游戏要结束了：ControlNet正在补完AIGC工业化的最后一块拼图

  https://mp.weixin.qq.com/s/-r7qAkZbG4K2Clo-EvvRjA

课程：

• 使用 Prompts 和 Chains 让 ChatGPT 成为神奇的生产力工具！

  https://github.com/howl-anderson/unlocking-the-power-of-llms

• 斯坦福的CS224N最新的课程lecture11，介绍了最新的LLM进展，包括了ChatGPT

  http://web.stanford.edu/class/cs224n/slides/cs224n-2023-lecture11-prompting-rlhf.pdf

商业：

• Roblox推出AIGC黑科技：文本生成3D模型和游戏场景 https://mp.weixin.qq.com/s/QMgegjzC9tRqIUPOhF3XhQ

• ChatGPT中国变形记｜深氪

  https://mp.weixin.qq.com/s/FG9Y3x-Rpy8Gw-dWAwWqDA

• ChatGPT背后真正的英雄：OpenAI首席科学家Ilya Sutskever的信仰之跃

  https://mp.weixin.qq.com/s/zauVrzr_ABY9WFor4BwNVQ

• 每个软件应用都将融入AI

  https://mp.weixin.qq.com/s/qdvJKZRquaBqaYNUAlBMMw

• 观点 | AIGC影响下的AI建筑学

  https://zhuanlan.zhihu.com/p/606502335

• 谁拖了中国ChatGPT的后腿？

  https://mp.weixin.qq.com/s/FRN0Hvy7-12qS82IvUdNeg

• ChatGPT成功做对了这4步

  https://mp.weixin.qq.com/s/U5v8CFmGIpjWBheDDWXCPA

• 李志飞将在大模型领域创业，做中国的 OpenAI https://mp.weixin.qq.com/s/gWqPv4d2RLnX8wj6A97b-A

案例：

• 复旦团队发布国内首个类ChatGPT模型MOSS，邀公众参与内测

  https://www.shobserver.com/staticsg/res/html/web/newsDetail.html?id=584634

• ChatGPT技术国产化尝试 https://mp.weixin.qq.com/s/mXViN_GB9VC1WrXP1Q1iug

• AI照骗恐怖如斯！美女刷屏真假难辨，网友：AI网恋诈骗时代开启

  https://mp.weixin.qq.com/s/nELNzal7tjkbZ6uKkuGkeA

• Improving Sample Quality of Diffusion Models Using Self-Attention Guidance 提高扩散模型生成质量的demo

  https://huggingface.co/spaces/susunghong/Self-Attention-Guidance

• 视频领域的ControlNet

  https://huggingface.co/spaces/fffiloni/ControlNet-Video

• 文本生成视频

  https://invideo.io/ai

• AI辅助设计

  https://www.genius.design/

• 根据文本提示自动生成设计

  https://stockimg.ai/

• 草图+提示生成图

  https://scribblediffusion.com/