...generate content
Jump to navigation
Jump to search
| AI Solver map |
|---|
|
Your current page is in bold
0 · Start
1 · Predict & analyze
tables, numbers, events
Predict values
Predict categories Forecast time series Cluster / segment Find outliers Make recommendations Find key drivers 2 · Understand unstructured data
text, documents, images, audio
3 · Generate & converse
4 · Decide & act
5 · Build it right
applies to every path
|
YouTube ... Quora ...Google search ...Google News ...Bing News
- AI Solver ... Generative AI ... Discriminative vs. Generative ... Data Augmentation ... ...understand new architectures
Use this page when you want AI to create something new: a draft, a program, an illustration, a video, a voice-over, training data. Generative models learn the patterns of their training data and produce new samples; your job is to steer them and to check the result.
Start here
- What are you creating?
- Text (drafts, marketing copy, reports, summaries) → a general-purpose LLM; steer it with instructions, examples and a style guide; ground facts with your own sources → ...answer questions from my data
- Code → coding assistants and agents (GitHub Copilot, Cursor, Claude Code, OpenAI Codex, Gemini CLI). Judge them on your own codebase, not only on benchmarks such as SWE-bench → ...build an agent
- Images → image models (OpenAI's GPT Image, Google's Imagen and Gemini image models, Midjourney, Adobe Firefly, Black Forest Labs FLUX; open: FLUX and Stable Diffusion variants). Brand or character consistency → reference images or a small fine-tune (LoRA)
- Video → text- and image-to-video models (Google Veo, OpenAI Sora, Runway, Kling, Luma)
- Voice, music and sound → text-to-speech (...work with speech and audio); music models (Suno, Udio, Google Lyria). Read the licensing terms before commercial use
- 3D assets and worlds → image-to-3D models (e.g., Microsoft TRELLIS, Tencent Hunyuan3D); interactive "world models" such as Google DeepMind's Genie 3 are still early
- Synthetic training data (rare cases, private data) → LLM-generated examples for text; tabular synthesizers (e.g., SDV) for tables. Check that it is realistic, diverse and does not leak private records → Data Augmentation, Data Labeling, and Auto-Tagging
- Will it be published, or will someone rely on it? → human review, fact-checking, and disclosure. Content provenance standards (C2PA Content Credentials) label AI-made media, and EU AI Act transparency duties for AI-generated content start to apply in 2026.
Under the hood: text and code come from autoregressive transformers; images, video and audio mostly from diffusion and flow-matching models; GANs still serve fast single-step tasks such as super-resolution. See ...understand new architectures.
Quick picks
| Output | Start with | Add when needed |
|---|---|---|
| Text | Frontier or mid-tier LLM with a style guide | Retrieval for facts, fine-tune for house style |
| Code | IDE assistant | Coding agent for multi-file tasks, with tests |
| Images | Hosted image model | Open model + LoRA for brand style |
| Video | Hosted video model | Editing workflow with human review |
| Synthetic data | LLM or tabular synthesizer | Privacy checks, train-on-synthetic / test-on-real evaluation |
Real-world examples
- Coding speed, measured two ways — in GitHub's 2022 controlled experiment, developers using Copilot finished a programming task 55% faster. Quantifying GitHub Copilot's impact | GitHub But in METR's 2025 randomized trial, experienced open-source developers working in their own large codebases took 19% longer with AI tools, while believing they were faster. Measuring the impact of early-2025 AI on experienced open-source developer productivity | METR - arXiv METR's February 2026 update said newer tools likely help more, but that selection effects made clean measurement hard. METR's developer productivity research: 2026 update | Rob Bowley Lesson: measure gains in your own setting.
Watch out
- Made-up facts — LLMs write fluent text whether or not it is true; ground and check.
- Rights — check licenses for training data, outputs, voices and likenesses; avoid imitating living artists or brands.
- Deepfakes and misuse — use provenance labels; have a takedown process.
Measure success
Human preference in blind A/B comparisons, task success, how much people edit the drafts, brand and compliance checks, and LLM-as-a-judge scores with a written rubric. More: ...evaluate and trust AI
Learn more
- C2PA: Coalition for Content Provenance and Authenticity
- SWE-bench: coding agent benchmark
- Generative tasks and models | Hugging Face
- Generative AI ... Discriminative vs. Generative
Where next? ◀ Generate & converse · ...answer questions from my data · ...build an agent · ...choose a model · ▶ ...evaluate and trust AI