AI Office & Productivity
Start from real tasks: master briefing, verification, email, meeting notes, table cleaning, chart reading, and reports.
Organized into five core tracks: AI Office & Productivity (16 lessons), AI Principles & Labs (16 code sandboxes), AI 3D Modeling & Assets (16 lessons), AIGC Visual & Media (ControlNet, ComfyUI, Video/TTS, AI Comic Drama), and AI Finance & Research.
Start from real tasks: master briefing, verification, email, meeting notes, table cleaning, chart reading, and reports.
Zero-video live code sandboxes: neural function fitting, Transformer ablation, MoE & 1M KV-cache calculators, and raw Agent/MCP/Skill payload inspectors.
Unifies 2D controlled generation, ComfyUI commercial visuals, AI video/voice synthesis, and AI comic drama into four specialized sub-tracks.
From generated drafts to production-ready 3D assets: multi-provider generation, Blender bmesh topology QA, UV/PBR baking, and CAD-to-3D delivery.
For financial research and quant workflows: filing citation extraction, bull/bear report cross-checking, time-series cleaning, and risk controls.
A task card: the goal, AI’s role, what you will prepare, and what you will check.
A reusable brief with a goal, audience, source material, requirements, and checks.
A checked invitation draft and a pre-send list covering the missing registration method and contact.
A standardized table, a change log, and a list of blank fields to confirm with respondents.
A calculation with explicit denominators, a chart recommendation, and a bounded conclusion.
Tune weights and activations to fit curves, and switch output heads to see next-token prediction as 50,257-way classification
Compare tokenizer boundaries and drag the “Apple” vector in 2D space to compute live cosine similarity
Tune learning rate to run live gradient descent and calculate exact Transformer layer parameter shares
Compare 4 evolutionary stages of attention and drag the query vector to watch live weight pooling
Toggle FFN expansion ratios (768→3072→768) and GELU activation to inspect feature filtering and vocab projection
Flip off Positional Encoding, Residual, and Norm switches in the ablation lab to witness word-order blindness and deep signal collapse
Adjust Temperature, Top-K, and Top-P thresholds and draw 20 live samples to see probability reshaping and tail pruning
Generate parallel self-supervised training pairs from any sentence and switch 0/1/3-Shot prompts to observe in-context learning
Switch across Base, SFT, CoT, and RLVR stages on a business math problem to inspect output streams and accuracy
Switch token domains in the MoE router sandbox to watch gated experts light up and audit compute savings
Drag the context slider (4K → 1M) and switch MHA, GQA, MLA, and Linear Attention to audit live KV Cache VRAM bills
Switch precision formats and memory tiers (SSD, DDR5, HBM3e) in the sandbox to calculate live throughput and quantization error
Compare Stateless, Full-History Concat, and Summary Memory in the payload inspector to view the actual raw JSON sent to the model
Toggle RAG evidence passages above and inspect Function Calling JSON interception and MCP discovery below
Switch across 4 orchestration patterns in the sandbox to compare main-window token consumption (48.5K vs 2.8K) and flexibility
Experiment with vision patch compression, GUI accessibility tags, green-list watermark bias, and benchmark contamination in the 4-in-1 lab
Practice progress stays in this browser. Copy or download your work; save draft edits before leaving.
Lock audience, key facts, format, and unknown rules upfront.
Subject + core schedule + deadline action + open placeholders.
Tag every claim with [Source A/B/C] and separate caps from spend.
Split into settled facts, named owners/deadlines, and open gaps.
Deduplicate with a log, standardize ISO dates, and preserve blanks.
Enforce 3-tier hierarchy and forbid unconfirmed promotional badges.
Understand capabilities, choose tasks, and assess results.
Give clear instructions and improve the result.
From email and meetings to tables and presentations.
Find evidence, make comparisons, and retain knowledge.
Turn ideas into writing, images, audio, and video.
Combine skills in a complete workplace project.
Turn a reliable method into a repeatable workflow.
Explore model mechanisms and application development.