Fable 5 Returns & 20 NotebookLM Prompts Worth Trying 📚

🎉 Anthropic's Fable 5 Returns After U.S. Lifts Export Controls

Anthropic reopened access to Fable 5 after 18 days offline, following the Commerce Department lifting its export restrictions. The model is back across all Claude tiers and platforms, though paid plans are capped at half their weekly limits until July 7 before switching to usage credits. The updated safety filter now blocks the cybersecurity issue that triggered the ban over 99% of the time, with Opus 4.8 handling fallback responses when it fires. As part of the resolution, Anthropic has committed to giving the U.S. government pre-release access to its future models.

🧪 Anthropic Launches Sonnet 5 and Claude Science

Claude Sonnet 5 is Anthropic's most capable Sonnet model yet, matching Opus 4.8 performance at a significantly lower price point and now available directly in the Claude app. Alongside it, Anthropic released Claude Science in beta, a dedicated research app that connects to 60+ scientific databases and integrates with your existing tools. The Science demo hit over 4 million views within days, which tells you something about where the appetite is right now. For anyone doing research-heavy work, this combination of a cheaper powerful model and a proper research interface is worth exploring immediately.

💻 Google's Gemini Spark Is Now an Agent on Your Mac

Gemini Spark can now work directly with local files on your Mac through the macOS app, handling tasks like organizing a messy Downloads folder or converting desktop files into documents and spreadsheets. Google says phone-to-Mac remote task execution is coming soon, which would make Spark one of the more practically useful cross-device agents available. The feature is currently in beta for Google AI Ultra members. It's a quiet but meaningful step toward AI that manages your computer environment rather than just answering questions inside a chat window.

Customer expectations keep rising. Support budgets don't. On July 9, Fin and AWS are hosting a live executive session on how leading enterprises close that gap: scaling AI-powered support while simplifying how they buy it.

You'll see how to resolve an average 76% of conversations with Fin on AWS enterprise-grade infrastructure, procure through AWS Marketplace to put committed cloud spend to work, and turn the Fin and AWS collaboration into lower support costs. Register for the live session to see how.

Save your spot

Most people open NotebookLM, upload a PDF, hit summarize, get a few bullet points, and close the tab. That's roughly 10% of what the tool can do.

NotebookLM is closer to a research partner that has read everything you've fed it and is waiting for you to ask better questions. The problem is nobody shows you what those questions look like. Here are 20 prompts worth actually trying, along with notes on why each one works.

1. The "explain it back to me wrong" check

Explain this concept the way someone who misunderstood it might. What's the common misconception?

Knowing the wrong version of an idea often makes the right version click faster. You read the misconception, recognize why someone would believe it, and suddenly the correct version lands properly.

2. The skeptic's pass

If you were a peer reviewer trying to find flaws in this, what would you flag?

Most people only ask AI to explain things favorably. Flip that. Ask it to find the weak points and you'll get a much more honest read on whether an argument actually holds up.

3. The "so what" filter

For each main point, tell me why it matters in practice. If it doesn't matter, say so.

"If it doesn't matter, say so" is the part that does the work. It gives the model permission to tell you something isn't a big deal, which it normally avoids.

4. Translate for a specific audience

Rewrite the key findings for a [12-year-old / busy executive / non-native English speaker].

Run the same source through three different audience lenses. You'll find the version that actually communicates before you're standing in front of the people who need to hear it.

5. Find the load-bearing assumption

What's the one assumption that, if wrong, would break this entire argument?

Every argument leans on something it never quite proves. This prompt finds it. Usually more useful than understanding the argument itself.

6. Build a debate prep sheet

I'm about to discuss this with someone who disagrees. Give me their likely objections and how I'd respond.

Useful before any meeting where pushback is coming. Also useful before certain family dinners.

7. Spot the outdated parts

Which claims in this source are most likely to be outdated, and why?

NotebookLM can't check live facts, but it's good at flagging which types of claims age fastest. Run this on older sources before citing them anywhere important.

8. Turn it into a decision

Based on this, what should I actually do? Give me options, not just information.

Most source material describes a situation without telling you what to do about it. This closes that gap directly.

9. Make a before-vs-after comparison

How does this source's view differ from conventional wisdom on this topic?

This is how you figure out what's actually new versus what's the same idea repackaged with current language.

10. Generate the questions you should be asking

What questions should I be asking about this topic that I probably haven't considered?

Sometimes the most useful output isn't an answer. It's a better question you didn't know to ask.

11. The "explain to future me" note

Write a note to myself in 6 months, reminding me what this was about and why it mattered.

Feels gimmicky until you try it. The compression it forces is what makes it stick. That note is often the fastest way back into material you haven't touched in a while.

12. Map the cast of characters

List every person, organization, or entity mentioned, and explain their role in one sentence each.

Dense reports have a dozen named players and you lose track of who's who by page three. This fixes it in one prompt.

13. The "what would change my mind" prompt

What evidence, if it existed, would make this argument fall apart?

One of the better ways to test whether you actually believe something or just haven't questioned it yet.

14. Build a glossary on the fly

Pull out every technical term and define it in plain language, ordered by how often each one appears.

The frequency ordering matters. The terms that show up most are the ones worth understanding first. Everything else you can look up as needed.

15. Find the contradiction

Are there any points in this source that contradict each other? Where?

Longer documents, especially anything written by committee, contradict themselves more than you'd expect. This finds it fast.

16. Turn it into a conversation

Write this as a dialogue between two people with different views, based on the source.

Dialogue makes complex ideas easier to follow than straight exposition and is a genuinely useful way to revisit something you've already read once.

17. The one-slide challenge

If you could only show one slide to summarize this, what would be on it?

Extreme prioritization. Whatever survives this cut is the actual core of the material.

18. Find the analogy that's already there

Is there an analogy or metaphor used in this source? If not, suggest one that fits.

Sometimes the author already buried a good explanatory device on page 12. Sometimes you need to build one yourself.

19. Build a "what's missing" list

What would you expect a source like this to cover that it doesn't?

Useful for spotting gaps, especially in research papers where the stated scope quietly narrows as you read.

20. Create your own follow-up reading list

Based on this source, what topics or questions would be worth researching next?

This is how one document becomes a starting point instead of an ending point.

Three things that made a real difference: Upload more than one source when you can. NotebookLM's actual strength is working across documents, not within a single one. A lone PDF undersells the tool significantly. Be specific about format "give me a table" or "five bullet points" gets you something usable. And don't treat the first response as final. Push back, ask it to go deeper, rewrite it for a different audience. The tool is built for back-and-forth, and the first output is rarely the best one.

Every missed call is a job your competitor just booked. Podium's AI Employee responds in under 2 minutes, qualifies the lead, and schedules the job — while your crew keeps working.

See It In Action

Inner Circle Edge is built for one thing > helping you actually get better at using AI, not just reading about it.

Tell us what you want covered next. Hit reply with your pick:

A) Claude skills and prompting techniques every serious user should know

B) Real automation workflows with n8n or Zapier you can set up today

One reply. That's it. Most votes wins the next issue.

Stay sharp.