🧠 AI Solved a Century Old Math Mystery on a Pep Talk

Anthropic has signed on to the EU AI Act's Article 50(2) Code of Practice on Transparency of AI-Generated Content, which effectively means every piece of text Claude generates will carry an embedded watermark you can't see or read, but that travels with the content even when it's copied and pasted.

What this means practically is that if someone uses Claude to write content and passes it off as their own, there's now a machine-readable signal attached to it. The watermark persists through some editing too. Anthropic says detection tools are coming, so third parties will eventually be able to check content for Claude's mark.

Keep in mind that the absence of a watermark doesn't confirm content is human-written. But this is a real step toward AI content accountability, and it's the first time a major lab has baked this kind of transparency into their infrastructure at the model level.

When Anthropic launched Claude Sonnet 5 back in June, they introduced it at $3 per million input tokens and $15 per million output tokens — wait, actually they launched it at $2 per million input tokens and $10 per million output tokens, which was already a notably low introductory price. That offer was set to expire August 31. It's now permanent.

For developers and businesses building on top of Claude's API, this matters a lot. Sonnet 5 is widely considered the sweet spot between capability and cost in Anthropic's current lineup. Locking the price in removes the planning uncertainty that comes with "introductory" pricing and makes it easier to build around the model with confidence.

Meta released the weights for Muse Glimmer, a 30B parameter dense model you can actually run locally on your own hardware. If you're not tracking the open-source AI race closely, here's the context: "open weights" means the model's parameters are publicly released, so anyone can download, run, fine-tune, or build on top of it without going through an API or paying a provider. 30B parameters is substantial enough to handle serious tasks while still being runnable on consumer-grade hardware.

Zuckerberg also published a long essay laying out Meta's philosophy on AI, making the case that everyone should have access to superintelligence. Whether you agree with his vision or not, Meta's consistent bet on open source is putting meaningful competitive pressure on every closed-model lab out there.

Imagine describing a goal and immediately being in a live voice call with six AI participants, each with a distinct name, role, and personality, who argue with each other, build on your ideas, and go quiet the moment you start talking. That's exactly what AI Group Call does. You get a host, a skeptic, a strategist, and more, all assembled around your specific goal in seconds. Every session comes with a full transcript, a summary of key points, and action items. It's legitimately useful for stress-testing a pitch, planning a project, or rehearsing a tough conversation before it happens for real.

Osmo is an AI-native video editor built specifically for product launch videos and motion graphics. You describe what you want, feed it Figma files or footage, and it produces polished, on-brand animations — built as editable code, which means you're not locked into a black-box output. It's aimed at founders, product teams, and creators who need launch-quality video without a full production team behind them.

Anthropic recently shared that an unreleased research version of Claude made a genuine mathematical breakthrough on a problem related to the Riemann hypothesis — one of the most famous unsolved problems in math, with a literal $1 million bounty on it. You can read the full article here.

The researcher guiding Claude wasn't a mathematician. His input was mostly sending Claude variants of "keep going" and "believe in yourself." That's it. Pep Talk. And those prompts seem to have helped Claude push through the self-doubt it had about whether it could actually make progress.

The pro tip here is simple but genuinely underused. When Claude gets stuck, hedges heavily, or tells you it's not sure it can do something, try giving it some encouragement. Push back on the uncertainty. Tell it to keep going. Ask it to try a different approach rather than accepting the first "I'm not sure I can do this." Claude, like a lot of people, can underestimate itself when faced with a hard problem, and a small nudge sometimes unlocks significantly better output.

Are you looking to automate your business with AI? We’re hosting live build sessions with Practicaly AI for Entrepreneurs members every Friday.

Not a member? Join now to get 20% off one month

Join us live, build in real time, and get your questions answered on the spot.

👉Save your spot today!

Can’t make it live? Don’t worry, we’ll send out the recording after the session to everyone who RSVP’ed.

Event details:

📅When: Friday at 10AM PT/1PM ET
📍Where: Virtual, Google Meet

After you RSVP, you’ll be sent a link with the event details. Hope to see you there!