10. AI & LLMs
What large language models actually are, how they work, and the vocabulary of the field. Covers Claude itself, the broader model landscape, and the techniques (tool use, agents, RAG, MCP, embeddings) that turn raw LLMs into useful systems.
Planned entries
| # | Entry | Status | One-line description |
|---|---|---|---|
| 1 | What is an LLM? | 🟩 COMPLETE | A statistical predictor of the next token — and why that’s so much more powerful than it sounds |
| 2 | How LLMs work | 🟩 COMPLETE | Transformers, attention, training, alignment — the inner mechanics |
| 3 | Tokens & context windows | 🟩 COMPLETE | The units LLMs actually see — and the limits on how much they hold |
| 4 | Temperature & sampling | 🟩 COMPLETE | The dials that control creativity vs determinism |
| 5 | Prompt engineering | 🟩 COMPLETE | Crafting the input to get the output you want |
| 6 | Claude models | 🟩 COMPLETE 🟦 LIVING | The current Claude lineup — Opus, Sonnet, Haiku, Fable, when to use each |
| 7 | The Claude API | 🟩 COMPLETE 🟦 LIVING | Calling Claude programmatically — the basic shape of an API call |
| 8 | Tool use (function calling) | 🟩 COMPLETE | Letting the LLM call tools you define — the foundation of agents |
| 9 | Agents | 🟩 COMPLETE | LLMs that loop, reason, and use tools to accomplish tasks |
| 10 | MCP — Model Context Protocol | 🟩 COMPLETE 🟦 LIVING | The “USB-C for AI tools” — standardized tool access for any LLM |
| 11 | RAG — Retrieval Augmented Generation | 🟩 COMPLETE | Looking things up in your own data before answering |
| 12 | Embeddings | 🟩 COMPLETE | Turning text into vectors so you can search by meaning |
| 13 | Fine-tuning vs context | 🟩 COMPLETE | When to retrain the model vs when to just put more in the prompt |
| 14 | Multimodal (vision, audio) | 🟩 COMPLETE 🟦 LIVING | LLMs that can see images and hear/speak |
Suggested reading order
1 → 3 → 5 → 6 → 8 → 9 → 10. The “what is an LLM, what are tokens, how do I prompt it, what’s tool use” arc is the most useful in practice. Save 2 (transformer internals) for when you’re curious.
Vendor entries (AI assistants, image/video/voice/music tools)
The AI landscape is much bigger than the 14 conceptual entries above. Vendor-specific entries (one per AI assistant, image generator, video generator, etc.) live in this section using the vendor entry format with country badges. Active build-out as of June 2026.
Frontier consumer assistants — Western:
| # | Entry | Status | One-line |
|---|---|---|---|
| 15 | 🇺🇸 OpenAI ChatGPT | 🟩 🟦 | The most-used AI chatbot; OpenAI’s consumer flagship |
| 16 | 🇺🇸 Google Gemini | 🟩 🟦 | Google’s flagship; deepest Workspace integration; free for AUS students |
| 17 | 🇺🇸 Anthropic Claude.ai consumer | 🟩 🟦 | The chat product (claude.ai); see also Claude models |
| 18 | 🇺🇸 Microsoft Copilot | 🟩 🟦 | Microsoft’s chatbot; in Edge, Windows, Office |
| 19 | 🇺🇸 xAI Grok | 🟩 🟦 | Elon Musk’s chatbot; X (Twitter) integration |
| 20 | 🇺🇸 Meta AI | 🟩 🟦 | Meta’s chatbot; in WhatsApp, Instagram |
| 21 | 🇺🇸 Perplexity | 🟩 🟦 | AI-powered search engine |
| 22 | 🇫🇷 Mistral Le Chat | 🟩 🟦 | French frontier-lab chat |
| 23 | 🇨🇦 Cohere Command | 🟩 🟦 | Canadian frontier-lab chat |
Frontier consumer assistants — ⛔ Chinese-origin (avoid):
| # | Entry | Status | One-line |
|---|---|---|---|
| 24 | 🇨🇳 DeepSeek — ⛔ DO NOT USE | 🟩 🟦 | The big 2025 China story; canonical example of why to avoid |
| 25 | Tongyi Qianwen — ⛔ DO NOT USE | 🟩 🟦 | Alibaba’s model family |
| 26 | 🇨🇳 Doubao — ⛔ DO NOT USE | 🟩 🟦 | ByteDance’s chatbot (and parent of TikTok / CapCut) |
| 27 | 🇨🇳 Kimi (Moonshot) — ⛔ DO NOT USE | 🟩 🟦 | Long-context Chinese chat |
| 28 | Wenxin (Baidu) — ⛔ DO NOT USE | 🟩 🟦 | Baidu’s flagship |
| 29 | Hailuo — ⛔ DO NOT USE | 🟩 🟦 | Multimodal incl. Hailuo video gen |
| 30 | ChatGLM — ⛔ DO NOT USE | 🟩 🟦 | Tsinghua-origin lab; on US Entity List since Jan 2025 |
| 31 | 🇨🇳 Tencent Hunyuan — ⛔ DO NOT USE | 🟩 🟦 | Tencent / WeChat ecosystem AI |
| 32 | 01.AI — ⛔ DO NOT USE | 🟩 🟦 | Kai-Fu Lee’s lab; founder background doesn’t change jurisdiction |
| 33 | 🇨🇳 Manus (Butterfly Effect) — ⛔ DO NOT USE | 🟩 🟦 | The viral early-2025 agent product |
Secondary Western consumer chat / privacy-focused:
| # | Entry | Status | One-line |
|---|---|---|---|
| 34 | 🇺🇸 DuckDuckGo AI Chat | 🟩 🟦 | Privacy-routed free access to multiple frontier models |
| 35 | 🇺🇸 Brave Leo | 🟩 🟦 | Brave browser’s built-in AI; privacy-first |
| 36 | 🇺🇸 Kagi Assistant | 🟩 🟦 | Paid privacy-focused search + AI with pick-your-model |
Capability cross-cut concept entries (Phase F):
| # | Entry | Status | One-line |
|---|---|---|---|
| 37 | Image generation | 🟩 🟦 | How AI makes pictures from words; all the major tools compared |
| 38 | Video generation | 🟩 🟦 | How AI makes video from text; Sora, Runway, Veo, Pika compared |
| 39 | Music generation | 🟩 🟦 | How AI composes and sings; Suno, Udio, copyright landscape |
| 40 | Voice synthesis (TTS) | 🟩 🟦 | AI text-to-speech, voice cloning, ElevenLabs and more |
| 41 | Speech-to-text (STT) | 🟩 🟦 | AI transcription; Whisper, Deepgram, AssemblyAI compared |
| 42 | Real-time voice AI | 🟩 🟦 | Full two-way spoken AI conversation; ChatGPT voice, Gemini Live |
| 43 | AI translation | 🟩 🟦 | DeepL, Google Translate, LLM translation compared |
| 44 | Deep research mode | 🟩 🟦 | AI that does hours of web research and produces cited reports |
| 45 | AI document generation | 🟩 🟦 | AI for reports, proposals, contracts, and long-form writing |
| 46 | AI slides generation | 🟩 🟦 | Gamma, Tome, Beautiful.ai, Copilot in PowerPoint compared |
Embedded vendor AI (Phase G):
| # | Entry | Status | One-line |
|---|---|---|---|
| 47 | 🇺🇸 Asana AI | 🟩 🟦 | AI project management; AI Studio for workflows |
| 48 | 🇺🇸 Linear AI | 🟩 🟦 | AI issue tracking for software teams |
| 49 | 🇺🇸 Airtable AI | 🟩 🟦 | AI database fields; classify, summarise, extract at scale |
| 50 | 🇸🇪 Spotify AI DJ | 🟩 🟦 | Personalised AI DJ with synthetic voice presenter |
Music/video tool vendors (Phase I overlap):
| # | Entry | Status | One-line |
|---|---|---|---|
| 51 | 🇺🇸 Suno | 🟩 🟦 | Easiest AI music generation; full songs from text |
| 52 | 🇺🇸 Udio | 🟩 🟦 | Premium AI music generation; best vocal quality |
| 53 | 🇺🇸 Runway | 🟩 🟦 | Professional AI video generation; Gen-4; filmmakers |
| 54 | 🇺🇸 Pika | 🟩 🟦 | Consumer AI video; fast; easy |
| 55 | 🇺🇸 Luma AI (Dream Machine) | 🟩 🟦 | Smoothest AI video motion; image-to-video |
| 56 | 🇨🇦 Ideogram | 🟩 🟦 | AI images with readable text; logos and posters |
Frontier labs as companies (Phase K):
| # | Entry | Status | One-line |
|---|---|---|---|
| 57 | 🇺🇸 xAI (Elon Musk) | 🟩 🟦 | Grok’s parent company; Colossus supercomputer; X integration |
| 58 | FAIR | 🟩 🟦 | Llama models; FAIR research; Facebook/Instagram/WhatsApp AI |
| 59 | 🇫🇷 Mistral AI (company) | 🟩 🟦 | Europe’s frontier lab; Mixtral; EU data sovereignty |
| 60 | 🇨🇦 Cohere (company) | 🟩 🟦 | Canadian enterprise AI; Embed/Rerank/Command R+ |
Vertical / specialised AI (Phase J):
| # | Entry | Status | One-line |
|---|---|---|---|
| 61 | 🇺🇸 Harvey AI (legal) | 🟩 🟦 | AI for law firms; contract review; legal research |
| 62 | 🇨🇦 Spellbook (legal) | 🟩 🟦 | AI contract drafting inside Microsoft Word |
| 63 | 🇺🇸 Khanmigo (education) | 🟩 🟦 | AI tutor from Khan Academy; Socratic method; not just answers |
| 64 | 🇺🇸 Duolingo Max (language learning) | 🟩 🟦 | AI-powered language learning; Roleplay + Explain |
| 65 | 🇺🇸 Gong (sales AI) | 🟩 🟦 | Revenue intelligence; call analysis; deal forecasting |
| 66 | 🇺🇸 Intercom Fin (customer service) | 🟩 🟦 | AI customer service agent; resolves queries 24/7 |
| 67 | 🇺🇸 HeyGen (avatar video) | 🟩 🟦 | AI avatar video generation; type script → presenter delivers it |
| 68 | podcast editing) | 🟩 🟦 | Edit video by editing text; voice clone; Studio Sound |
| 69 | 🇺🇸 Opus Clip (video clips) | 🟩 🟦 | AI long-video → short clips for TikTok/Reels/Shorts |
| 70 | 🇺🇸 Hallow (religious) | 🟩 🟦 | Catholic prayer and meditation app with AI companion |
| 71 | Be My AI (accessibility) | 🟩 🟦 | AI visual assistance for blind and low vision users |
| 72 | 🇬🇧 Synthesia (avatar video) | 🟩 🟦 | Enterprise AI avatar video; UK/EU GDPR; 140+ languages |
| 73 | 🇮🇱 D-ID (talking photos) | 🟩 🟦 | Israeli AI photo animation; make still photos speak |
| 74 | 🇺🇸 Fireflies.ai (meeting notes) | 🟩 🟦 | AI meeting transcription and summaries; recording law note |
| 75 | 🇦🇺 Frollo (personal finance) | 🟩 🟦 | Australian CDR-native personal finance AI; free; data stays in AUS |
| 76 | 🇺🇸 Apollo.io (B2B sales) | 🟩 🟦 | B2B sales prospecting AI; 275M contacts; AU Spam Act compliance |
| 77 | 🇺🇸 Hippocratic AI (healthcare) | 🟩 🟦 | Conversational health AI for patient engagement; HIPAA |
| 78 | 🇬🇧 Cleo (personal finance) | 🟩 🟦 | UK conversational personal finance AI; Gen Z personality |
| 79 | 🇺🇸 Google Lyria | 🟩 🟦 | Google DeepMind music generation model; YouTube Dream Track |
| 80 | 🇺🇸 Google Gemma | 🟩 🟦 | Google’s open-weights small models; run locally; permissive licence |
| 81 | 🇺🇸 ChatGPT Memory | 🟩 🟦 | How ChatGPT remembers you across conversations |
| 82 | 🇺🇸 ChatGPT Canvas | 🟩 🟦 | Side-by-side collaborative editing with AI |
| 83 | 🇺🇸 OpenAI Voice Engine | 🟩 🟦 | Voice cloning research preview; safety considerations |
| 84 | 🇺🇸 OpenEvidence (clinician AI) | 🟩 🟦 | Evidence-based clinical decision support; free for medical professionals |
| 85 | 🇺🇸 Otter.ai (meeting notes) | 🟩 🟦 | Meeting transcription + AI summary; live captions for accessibility |
| 86 | 🇮🇱 K Health (primary care) | 🟩 🟦 | AI primary care; not in AU but informative; clinical AI model |
| 87 | 🇺🇸 Sierra (CX agents) | 🟩 🟦 | Enterprise AI customer experience; ex-Salesforce founders |
| 88 | 🇺🇸 Captions.ai (mobile video) | 🟩 🟦 | Mobile-first AI video creation; AI Eye Contact; Western CapCut alternative |
| 89 | 🇦🇺 Pocketbook (AU finance) | 🟩 🟦 | Established Australian personal finance app; AI categorisation |
| 90 | 🇦🇺 WeMoney (AU credit + finance) | 🟩 🟦 | Australian app with credit score focus; community features |
| 91 | 🇦🇺 Heidi Health (AU clinical AI) | 🟩 🟦 | Australian AI clinical documentation; AHPRA-aware; data in AU |
| 92 | 🇺🇸 AssemblyAI (speech AI) | 🟩 🟦 | Speech AI platform; Audio Intelligence; LeMUR Q&A |
| 93 | 🇺🇸 Deepgram (speech AI) | 🟩 🟦 | Fastest STT; Voice Agent platform; Aura-2 TTS |
| 94 | 🇩🇪 DeepL (translation) | 🟩 🟦 | Premium translation; EU/GDPR; highest quality European languages |
| 95 | 🇺🇸 LangSmith (LLM observability) | 🟩 🟦 | LLM application observability; LangChain native |
| 96 | 🇩🇪 Langfuse (open-source observability) | 🟩 🟦 | Open-source LLM observability; self-hostable; EU |
| 97 | 🇬🇧 Recraft (vector AI) | 🟩 🟦 | UK vector graphics AI; SVG output; brand design |
| 98 | 🇺🇸 Higgsfield (cinematic video AI) | 🟩 🟦 | AI video with cinematic camera control; human motion |
| 99 | 🇦🇺 Annalise.ai (radiology AI) | 🟩 🟦 | Australian radiology AI; chest X-ray + CT brain; TGA-approved |
| 100 | 🇦🇺 Harrison.ai (parent company) | 🟩 🟦 | Australian medical AI parent of Annalise + Franklin |
| 101 | 🇦🇺 Lyrebird Health (clinical AI) | 🟩 🟦 | Australian clinical documentation alternative to Heidi |
| 102 | 🇺🇸 Character.AI (companion) | 🟩 🟦 ⚠️ | AI character roleplay; significant safety concerns; honest treatment |
| 103 | 🇺🇸 Replika (AI companion) | 🟩 🟦 ⚠️ | AI companion app; controversial; ethical considerations |
| 104 | 🇬🇧 Stability AI (company) | 🟩 🟦 | UK company; made Stable Diffusion; tumultuous history |
| 105 | 🇩🇪 Black Forest Labs (Flux) | 🟩 🟦 | German company; makes Flux; ex-Stability AI team |
| 106 | 🇺🇸 Cognition AI (Devin) | 🟩 🟦 | Makers of Devin AI engineer; honest assessment of hype vs reality |
| 107 | 🇺🇸 Glean (enterprise search) | 🟩 🟦 | Enterprise AI search across organisational systems |
| 108 | 🇺🇸 LangChain (framework) | 🟩 🟦 | The dominant LLM application framework; open-source; LangGraph |
| 109 | Vector databases (overview) | 🟩 🟦 | Pinecone, Weaviate, Chroma, pgvector; RAG infrastructure |
See also
- AI Landscape — the map 🟩 🟦 — Western vs Chinese guidance, capability matrix, pricing, “which AI for which job”
- 11. AI-assisted development 🟩 🟦 — putting all of this to work via Claude Code, Cursor, Antigravity, etc.
- 14. Design & UX 🟩 — design tools with AI (Canva, Figma AI, Adobe Firefly)
- 15. Broader tech (bonus) 🟩 — cloud platforms hosting AI (AWS Bedrock, Azure OpenAI, GCP Vertex)
- 03. Backend 🟩 — most LLM API calls are server-side
- Anthropic’s official docs: https://docs.anthropic.com