AI updates for normies: what changed, what matters, and what to try next

All the tips, plus the video replay (for a limited time)

🎥The Video Replay (available for a limited time)

Autogenerated English subtitles are available

🤖 Model landscape: GPT, Claude, and friends

What we discussed

Most “AI chat” you see in apps routes to a major model like ChatGPT or Claude. Different models have different strengths and writing styles, and the paid tiers unlock extra features and reliability.

Why it matters:

Picking the right model for the task improves quality and cuts retries.

What you can do:

• Keep both ChatGPT and Claude handy for A/B results

• Use the paid tier if you rely on AI for work

• Save good chats as reusable “workbenches” for ongoing projects

🧠 Thinking mode and agents: the big shift

What we discussed:

Modern models can “think” by planning steps internally and supervising their own progress. You define the problem; the model plans, searches, drafts, and revises. This looks like an “agent” that can orchestrate tools and return a fuller result.

Why it matters:

You write shorter prompts and get deeper work back, but you must still verify.

What you can do:

• Say “think deeply” or choose a reasoning mode for complex tasks

• Give clear goals and constraints; let the model plan the steps

• Review outputs like you would a junior analyst’s work

🛠️ A real example: data-transform script in minutes

What we discussed:

We used extended “thinking” to generate a complete Python script that transformed a client’s Excel into a tax-import format, including runnable code and usage notes. Time saved: half a day or more.

Why it matters:

Reasoning modes turn multi-hour builds into review-and-run tasks.

What you can do:

• Attach sample inputs and the target spec; request a single-file script

• Ask for Windows and macOS run instructions

• Test on a small subset before full batch runs

🔎 AI-first search: Perplexity and the Comet browser

What we discussed:

Instead of “search, click, back,” tools like Perplexity and the Comet browser run a research plan, read sources, and return a synthesized answer with citations. Great for product research and technical comparisons.

Why it matters:

You get a focused brief faster and with references to check.

What you can do:

• Write the query like a brief: goals, constraints, exclusions

• Let it run; then skim the sources panel before deciding

• Save the best prompts for repeatable research tasks

🎨 Media generation and deepfakes: fun and risk

What we discussed:

Image and video models have leapt forward (e.g., Sora video demos), making creative mockups trivial but also enabling convincing deepfakes. Casual projects are easier; skepticism is mandatory.

Why it matters:

Quality is up, friction is down, and verification is critical.

What you can do:

• Prototype visuals with modern image tools, but confirm facts elsewhere

• Treat viral videos with caution; look for original sources

• Keep sensitive images and likenesses out of public training reach

✍️ Prompting in 2025: short briefs vs. system prompts

What we discussed:

Two workable styles: concise briefs that trigger “thinking,” and longer system prompts that lock the model into a specific role and workflow. Both are useful; choose based on task complexity.

Why it matters:

Right-sized prompting reduces iteration and improves consistency.

What you can do:

• For fast tasks: short goal, constraints, and success criteria

• For repeatable workflows: maintain a detailed system prompt

• Ask the model to generate a checklist before it starts

Key takeaways - your TL;DR checklist

• Use the right model for the job; keep GPT and Claude in your toolkit

• Turn on “thinking” or ask to “think deeply” for complex work

• Provide real inputs and specs; review outputs like an editor

• Try AI-first search for research briefs, but verify sources

• Enjoy new media tools and stay alert to deepfakes

Tools and links mentioned

• ChatGPT and Claude for complementary strengths

• Perplexity and the Comet browser for AI-first search and summaries

• Image and video tools including Sora-style demos; Flux for images; audio tools like 11labs Music and Beethoven for licensed tracks

• Long-form system prompts for repeatable workflows; short briefs for quick “thinking” tasks

• “On Working with Wizards” - Ethan Mollick - https://www.oneusefulthing.org/p/on-working-with-wizards

• Intro to AI agents - https://medium.com/@aleixlopez/introduction-to-ai-agents-62a790d0bc22

Wildcard Wednesday returns next month

Second Wednesday at 12:00 PM Pacific. No slides. No sales pitch. Just practical tech you can actually use.

📆 Mark your calendars for high noon Pacific, the second Wednesday of every month!

You never know what we’ll get into next. But you will walk away smarter.

👉 Got a topic or question you want to bring up next time? Just reply and let me know.

In the meantime, if you need a hand or want to explore any of these topics further, you know where to reach me. 😉