Welcome back! Opus 5.5 makes real motion graphics now, and the setup takes ten steps. Four of them are the ones nobody does.
In today's Generative AI Academy Newsletter:
10 steps to turn Claude into your motion designer: Which step do most people stop at, and why does their work look like everyone else's?
Which AI tool should you use? Which group of features are you paying for and never touching?
Computer vision every data analyst should know: What does a self-driving car actually see?
Top LLM cost optimization techniques every engineer should know: Which two should you start with?
10 steps to turn Claude into your motion designer
Opus 5.5 can (actually) make (good) motion graphics.
Here's how you set Claude up in 10 steps:
Set up
1. Pick Opus 5.5. Open Claude Code, switch the model.
/model
2. Install the video skill. One free install turns code into MP4.
npx skills add heygen-com/hyperframes
3. Add the skills pack. Download the free pack, drop the folders in.
~/.claude/skills/
4. Test with one line. Before you brief anything properly, see what it does unprompted.
"Make a dynamic 15-second motion video. Go all out."
Build
5. Give it a proper brief. Pick a question people actually want answered, then describe the move you want.
"Why do we dream? Zoom out of the eye into space."
6. Give notes like a designer. One change per note. Vague feedback produces vague revisions.
"The text at the start undersells what's next."
Brand it
This is where it stops looking generated.
7. Build MOTION.md. Pick 5 frames you like and have Claude turn them into your colors, type and timing.
"Write me one file called MOTION.md."
8. Make Claude read it. One rule in CLAUDE.md so it loads before every build.
"Before animating, read MOTION.md in full."
9. Add Apple's rules. Install apple-design, then make it audit your output.
"Use apple-design to audit this, worst first."
10. Use real components. Paste 21st.dev prompts and keep your own look.
"Treat this component as a structural donor."
Steps 1 to 6 get you a video. Steps 7 to 10 get you a video that looks like yours.
Most people stop at 6 and wonder why everything they make looks the same as everyone else's.
You don't need to pay to learn Claude
Seven Claude courses at GenAI Academy. All recorded, and all free.
Start with "What Can AI Actually Do For You?" if you're new, 30 minutes, no setup required.
Then the Claude Starter Course gets you running in 35.
Past that, pick by the problem you have. Hitting usage limits in Claude.ai, Cowork or Claude Code has its own course.
Building no-code agents has one.
So does rolling Claude out across an ops team in 30 days, from pilot to adoption data to full deployment.
The AI Portfolio Builder turns whatever you built into a case study you can show someone.
World Wide Vibes Hackathon is live too, $5,000 prize pool, beginner-ready, 100% online.
One login. Watch them in any order.
Which AI tool should you use?
You don’t need 8 AI subscriptions.
22 useful features and ways to get more from the tools you have. What to use? How to start?
Finding things out
Web search for current facts with sources (Perplexity).
Deep research when you need five options compared (ChatGPT Work).
File uploads when the answer is in your own documents (Gemini Notebook).
Connectors when it lives in Notion or Drive (Claude).
Making things
Images (ChatGPT). Video (Higgsfield).
Infographics you can actually edit (Codex plus Figma).
Video analysis that reviews your hook and on-screen text (Claude Code).
Working alongside you
Screen sharing (Gemini Live). Voice mode for rehearsing a pitch (ChatGPT).
Writing beside a live draft (Claude Artifacts).
Data analysis with the calculations shown (Opus 5.5).
Code and debug (Codex).
The part people skip
Memory, so it stops needing to be told who you write for.
Agent instructions in AGENTS.md or CLAUDE.md, so your rules load every session.
Local projects, so it reads a folder instead of files you paste.
Scheduled tasks, so Monday's brief arrives unasked. Reusable skills in SKILL.md, so a process you solved once runs on command.
The first three groups make you faster at tasks. The last one works when you're not there.
Most people pay for four subscriptions and use nothing from that last group. Start there.
Computer vision every data analyst should know
Nine concepts that take you from pixels to a self-driving car.
1. Image representation
An image is an array of numbers. Grayscale gives you one value per pixel. Color gives you three, one each for red, green and blue. Everything after this is arithmetic on that grid.
2. Preprocessing
Resize, normalize, denoise, augment. Unglamorous and it decides how well everything downstream works. A 224x224 image becomes 112x112 and your compute bill halves.
3. Convolution
A small filter slides across the image and produces a feature map. Different filters catch different things: edges, textures, repeating patterns. The network learns which filters are worth having.
4. ReLU
f(x) = max(0, x). Negatives become zero, positives pass through untouched. That one line is what stops a deep network from collapsing into a single linear function.
5. Pooling
Shrink the grid, keep the strongest values. A 4x4 becomes a 2x2. Less computation, and the model stops caring whether the cat moved two pixels left.
6. CNN architecture
Stack it all. Input, convolution, ReLU, pooling, convolution, ReLU, pooling, fully connected layer, then probabilities out. Early layers find edges. Later layers find faces.
7. Object detection
Beyond "there's a dog." Now it returns a class label, a bounding box and a confidence score for every object in the frame.
8. Segmentation
Every single pixel gets a class. Road, sidewalk, car, person, sky. This is what a self-driving car actually sees.
9. Where it lands
Self-driving cars, medical imaging, face recognition, industrial inspection, crop monitoring, video surveillance, visual search, retail.
Steps 1 through 5 are the whole foundation. Everything else is stacking them differently.
Master Claude Code in 6 sessions
You already know the developers who seem to move twice as fast as everyone else.
They're not typing faster. They're:
Orchestrating teams of agents
Running parallel Claude instances across git worktrees
Wiring MCPs into their whole stack
Scheduling loops that keep working while they sleep
Almost none of it is written down in one place.
Mastering Claude Code course puts it all in one place.
6 live sessions, each one hands-on:
You start with the agentic loop and model selection
You build Skills and full agent teams from scratch
You apply Boris Cherny's 13 tips plus 6 new April 2026 techniques
You set up all 3 tiers of CLAUDE.md as a second brain that gets smarter every session
You wire MCPs into Slack, GitHub, Gmail and Supabase
You leave having built, tested and shipped your own custom Skill as a capstone
This is the deep track, the habits that separate casual users from power users.
Pricing is locked in for early sign-ups.
Top LLM Cost Optimization Techniques Every Engineer Should Know
1. Use smaller models. Cheapest model that does the job. Save the big ones for coding, planning and agents. Classification and extraction run fine on small.
2. Route by complexity. Sentiment to a small model, SQL to a medium one, coding to a large one.
3. Trim context. Keep recent messages, drop irrelevant history, kill duplicates. 20k tokens down to 2k changes your bill.
4. Summarize conversations. Replace 100 old messages with the facts, preferences and decisions that came out of them.
5. Cap output. Max token limits plus "answer in 3 bullet points." Fewer output tokens, lower cost.
6. Use RAG. Send the top matching chunks instead of the 100-page manual.
7. Cache responses. Same question twice, return the stored answer.
8. Cache prompts. System prompts, docs and instructions repeat on every call. Especially effective for agents and RAG.
9. Cache semantically. "What is OAuth?" and "Explain OAuth" can share one cached answer.
10. Ask for structured output. JSON over paragraphs. Fewer tokens, easier parsing, fewer retries.
11. Batch. One request with many records beats many requests with one. Good for tagging, classification and enrichment.
12. Guardrail your agents. Max iterations, max tool calls, max tokens, timeouts. Agents are usually the biggest line on the bill.
13. Tools before models. Time goes to a clock API, weather to a weather API, currency to a calculator, lookups to SQL. Call the model when reasoning is required.
14. Classify first. Decide whether it's FAQ, RAG, agent or coding before you process it. Stops you running an agent on a question a lookup would answer.
15. Watch the numbers. Cost per user, cost per request, tokens per request, cache hit rate, model usage, agent cost. Daily spend, spikes, and which prompts and users are expensive.
Most teams can cut 60% or more with three or four of these. Start with 12 and 13.
Everything else you shouldn't miss
arXiv is rationing submissions: September brought twice the submissions of the same month two years ago, and volunteers check every one. New limit: two papers per person per month, three waiting at once, and rejected papers still burn a slot. One arXiv leader says some authors were sending in dozens.
Google's AI chips are in orbit: A fridge-sized satellite carrying them is circling now, the first real test of Project Suncatcher. The orbit keeps panels in near-constant sunlight so it skips heavy batteries, and the chips run in 15-minute shifts because a vacuum has no air or water to carry heat away.
Reddit is closing the side door: Any tool that wants to read Reddit threads will need a deal from March 2027. New free API requests stop October 31, and the feeds people use to follow subreddits in reader apps die November 13. Reddit already licenses the same posts to Google and OpenAI.
Atlas got new hands with no pinkies: Boston Dynamics taped down their own pinkies to see what they'd lose, decided it was not much, and built four-fingered hands. In the demo Atlas fits a drill bit, bores into wood, spins a nut tight and moves two golf balls in one hand.
Learn more about AI from the experts building it
Follow us on Instagram for fast, visual AI updates in 30 seconds.
Subscribe to our Atlas newsletter — trusted by 3M+ subscribers — to stay ahead of AI news across tech, education, and business.
📺 Watch us on YouTube to hear insights directly from leading AI voices, builders, and innovators.
🐦 Follow us on X for breaking AI news and real-time industry updates.
Learn how to build your next AI application with practical resources and expert guidance.
Explore investment opportunities in the future of AI and join our community-backed growth journey.
