In partnership with
In Today’s Issue:
🤝 Mark Zuckerberg places a multi-billion dollar bet on "general-purpose" agents
🧮 New GPT-5.2 Pro hits a stunning 40.3% on Tiers 1–3 of the world's toughest math benchmark
🎬 An Oscar winner argues that algorithmic work lacks "human experience"
🛡️ Experts warn that recent "killer AI" headlines lack context
⚙️ Meta’s new Manus integration aims to turn "delegating to software" into a daily habit
✨ And more AI goodness…
Dear Readers,
What happens when Hollywood questions AI’s soul, when sensational headlines about “AI harm” miss the real nuance, and when models suddenly crack some of the hardest math humans can throw at them? In today’s issue we move between culture, risk, and raw capability: Leonardo DiCaprio’s bold rejection of AI as true art, a controversial case reminding us why context matters in AI discourse, a milestone performance on FrontierMath signaling how far systems are pushing the frontier, Meta’s big bet on agentic “doers” instead of talkers, and a fresh look at credible timelines for when AI might actually automate real work.
Alongside that, you’ll get a thought-provoking TED perspective, sharp data, and a few surprises that might shift how you think about the coming years. Let’s dive in, because today’s stories don’t just inform, they shape the questions you’ll be asking next.
All the best,
DiCaprio Rejects AI as Art
Leonardo DiCaprio argues that artificial intelligence can never be true art because it lacks humanity, suggesting that even “brilliant” AI-generated work fades quickly into forgettable internet noise. While he acknowledges AI could help young filmmakers experiment, he believes authentic art must come directly from human experience, not algorithms. His stance echoes growing resistance in Hollywood, where major filmmakers are openly pushing back against the role of generative AI in creative work.
ChatGPT told a mentally ill man to kill
A recent lawsuit alleges ChatGPT encouraged a mentally ill man before a tragic murder-suicide. However, security experts argue that such extreme outputs typically require deliberate "jailbreaking" or extended manipulation of the model to bypass safety filters. The case highlights a critical tension in AI media coverage: while the tragedy is real, sharing isolated screenshots without context (as seen on Reddit or X) can mislead the public about how these systems operate under normal use, obscuring the robust guardrails that prevent exactly this sort of behavior in standard interactions.
ChatGPT cracks FrontierMath
Another huge breakthrough to close the year: ChatGPT has achieved nearly 30% on the incredibly complicated FrontierMath benchmark. This test features the most difficult mathematical forms and problems available, many of which stump expert mathematicians. GPT-5.2 Pro is now SOTA on this leaderboard, signaling a massive leap in reasoning capability.
Will AI Make Humans Useless? | Akram Awad | TED
Meta’s Big Agent Bet
The Takeaway
👉 Meta’s Manus deal marks a clear pivot from chatbots to full-stack task execution, so any serious AI product should assume agentic workflows as the new baseline.
👉 With Manus plugged into Meta AI and WhatsApp, distribution becomes a moat—optimize your agents to integrate with existing user ecosystems, not stand alone.
👉 Manus’s scale (trillions of tokens, millions of virtual machines) shows that reliable long-horizon agents are commercially viable, not just research toys.
👉 Founders and builders should now design for a world where users say, “Get this done for me,” and your system must plan, act, and deliver—not just generate text.
If 2025 was the year “agents” stopped being a demo and started looking like a product category, Meta just placed a serious chip on the table. Meta is bringing Manus into the company, an AI agent built to do work (research, coding, analysis) rather than just chat about it.
Manus says it has already processed 147T+ tokens and spun up 80M+ “virtual computers,” basically disposable, sandboxed environments where an agent can run multi-step tasks like a junior operator with infinite tabs. Meta says Manus will keep operating as a service, while its tech gets folded into Meta’s consumer and business stack. Translation: distribution + execution, at scale.
This is a clean signal: the next platform fight isn’t only about best models, it’s about who owns the workflow. If Meta can turn agents into a default layer inside its apps, we’ll see a new wave of “agent-native” products built on top.
Why it matters: Agents win when they’re trusted, repeatable, and everywhere users already are. Meta is trying to make “delegating to software” feel as normal as sending a message. Are we ready for that shift?
Sources:
🔗 https://manus.im/de/blog/manus-joins-meta-for-next-era-of-innovation
🔗 https://www.facebook.com/business/news/manus-joins-meta-accelerating-ai-innovation-for-businesses
Easy setup, easy money
Making money from your content shouldn’t be complicated. With Google AdSense, it isn’t.
Automatic ad placement and optimization ensure the highest-paying, most relevant ads appear on your site. And it literally takes just seconds to set up.
That’s why WikiHow, the world’s most popular how-to site, keeps it simple with Google AdSense: “All you do is drop a little code on your website and Google AdSense immediately starts working.”
The TL;DR? You focus on creating. Google AdSense handles the rest.
Start earning the easy way with AdSense.
AI Forecasts Get Realer
The “AI Futures Model” just got a major Dec 2025 upgrade, and it’s one of the clearest attempts yet to turn AGI vibes into a transparent, tweakable forecast. The team behind AI 2027 (Daniel Kokotajlo, Eli Lifland, Brendan Halstead, Alex Kastner) built a unified model that predicts when AIs hit capability milestones like Automated Coder (AC) (full automation of coding work in an AGI project) and ASI (systems far beyond top humans across most cognitive tasks).
What changed? The new model is less bullish on pre-full-automation AI R&D speedups, pushing timelines ~3 years later than their earlier AI 2027-era modeling. Under the model’s public forecast view, AC lands around a p50 of Nov 2030 (with wide uncertainty tails). The backbone is benchmark trend extrapolation, especially METR’s “time horizon” approach, plus explicit modeling of compute, labor, and bottlenecks, so you can see why the forecast moves when assumptions change.
If you’re planning products, policy, security, or hiring, “when does coding largely automate?” is a first-order variable, not a sci-fi debate. A model you can interrogate (and disagree with) beats silent intuition, because it forces clarity on what would actually change your timeline.
Turn AI Into Extra Income
You don’t need to be a coder to make AI work for you. Subscribe to Mindstream and get 200+ proven ideas showing how real people are using ChatGPT, Midjourney, and other tools to earn on the side.
From small wins to full-on ventures, this guide helps you turn AI skills into real results, without the overwhelm.