MasaがGPT-4o画像生成の仕組みについて、各エンジニアの予想を調査したのでそれについて話します。

• Pythonで学ぶ画像生成 機械学習実践シリーズ
• dataclass で万物に型を付けよう
• Limitless Pendant
• 創作する遺伝子 僕が愛したMEMEたち
• 【トーク】インパルス板倉 嫉妬した芸人ベスト10！板倉が抱えていた様々な「言い訳クリスタル」を粉砕した芸人たちを本音で話す！
• Mickey 17
• Bong Joon Ho
• Robert Pattinson
• Mickey7
• try! Swift Tokyo Timetable
• WWDC 2025
• Apple Park
• Claude 3.7 Sonnet
• OpenAI Realtime API
• TC39
• SeattleJS
• Temporal
• ts-blank-space
• TypeScript syntax not supported by `ts-blank-space`
• Oracle justified its JavaScript trademark with Node.js—now it wants that ignored
• Sun Microsystems
• Oracle JavaScript Extension Toolkit
• Princess Mononoke 4K IMAX
• Introducing 4o Image Generation
• Autoregressive model
• Understanding Next Token Prediction
• Sora: Creating video from text
• Video generation models as world simulators
• Bay Bridge 近くのOpenAIオフィスはありました
• Golden Gate Bridge
• San Francisco–Oakland Bay Bridge
• 1人目 動詞 さんの予想
• GPT-4oとGemini-2.0の画像生成能力はいかにして作られているのか
• [2206.10789] Scaling Autoregressive Models for Content-Rich Text-to-Image Generation
• [2110.04627] Vector-quantized Image Modeling with Improved VQGAN
• [2309.02591] Scaling Autoregressive Multi-Modal Models: Pretraining and Instruction Tuning
• [2206.03605] Unified-IO 2: Scaling Autoregressive Multimodal Models with Vision, Language, Audio, and Action
• [2402.12226] AnyGPT: Unified Multimodal LLM with Discrete Sequence Modeling
• [2404.02905] Visual Autoregressive Modeling: Scalable Image Generation via Next-Scale Prediction
• A GPT-4o generated image, 2024年5月
• 2人目 Sangyun Lee さんの予想
• [2310.01400] Sequential Data Generation with Groupwise Diffusion Process
• 3人目 Wh さんの予想
• [2406.11838] Autoregressive Image Generation without Vector Quantization
• [2105.01601] MLP-Mixer: An all-MLP Architecture for Vision
• 条件付き確率分布
• 4人目 K.Ishi さんの予想
• [2408.11039] Transfusion: Predict the Next Token and Diffuse Images with One Multi-Modal Model
• 5人目 Saining Xie さんの予想
• [2103.00020] Learning Transferable Visual Models From Natural Language Supervision
• [2112.10752] High-Resolution Image Synthesis with Latent Diffusion Models
• 6人目 Nayan Saxena さんの予想
• OpenAI image gen actually shows just 5 frames
• [2005.14165] Language Models are Few-Shot Learners
• 4o Image Generation In-Context Learning