⏳AI's Attention Span

Plus Voice LLMs, Huawei's New Chips, Anthropic's New Quest, and More

Welcome to another edition of the Neural Net! We’re here to help you make the most of this crazy AI journey we’re all on.

In today’s edition: AI’s attention span is improving, how to use voice LLMs, Huawei’s new chips, Anthropic’s interpretation quest, a legal face plant, Notion’s AI offer, and Duolingo joins the AI wave.

The Street

note: stock data as of market close

Measuring AI’s Attention Span

Rather than just testing whether AI can answer trivia questions or handle isolated tasks, researchers are now looking at something deeper: how long AI can stay focused on a full task, from start to finish, without losing track.

The idea is to measure AI performance based on the time it would take a human to complete the same task — using it as a powerful new proxy for real-world capability.

The takeaway from the recent findings? LLMs aren’t just getting smarter.
They’re getting better at staying on course.

📈 AI’s Progress by the Numbers

Current AI vs. Human Performance:

  • AIs complete tasks taking a human < 4 minutes almost perfectly (~100% success).

  • Current models like Claude 3.7 Sonnet can now complete 1-hour-long human tasks with around 50% reliability.

  • For tasks taking a human > 4 hours, their success rate plummets to around 10%.

Exponential Growth:

  • The maximum task length AI can complete with 50% reliability has doubled every 7 months for six years straight.

  • Continuing on that trend would mean that by 2032, AI agents could automate month-long software development projects.

Stability of the Trend:

  • Even if the absolute measurements are off by a factor of 10, the timeline to AI automation shifts by only about two years — indicating the progress we’re seeing is steady and reliable.

🛠️ Why Real-World Skills Are AI’s True Test

Real-world projects aren’t just about answering a question — they’re messy, multi-step, and time-consuming. As the researchers astutely observed in their findings:

“We think these results help resolve the apparent contradiction between superhuman performance on many benchmarks and the common empirical observations that models do not seem to be robustly helpful in automating parts of people’s day-to-day work.”

In other words, the inability to string together long-term, complex actions is why AI looks like a genius on paper — but a total intern in real life. AI currently shines in the neat confines of the data in which it was trained — but struggles to carry that success into the open-ended, multi-step chaos of everyday life.

But with rapid progress in handling long-term, complex tasks, AI is on track to shift from a easily distracted assistant to a true partner in real-world work.

AI has been sprinting forward these past few months — and if you missed it, you might be living under a very cozy rock. While expert timeline predictions may vary on when we’ll reach general artificial intelligence, one thing is clear: it’s happening — and sooner rather than later.

💡How To AI: Smarter Conversations with Voice LLMs

Voice assistants used to be pretty basic — you’d ask for the weather, and if you were lucky, you got it. Early versions worked off simple rules ("If user says X, respond with Y"). But with the rise of Voice LLMs (Large Language Models + voice), they’ve leveled up dramatically, making conversations smarter, more flexible, and way more useful.

Here’s how to leverage AI in your daily life — no typing, just talking (and maybe a little yelling at inanimate objects):

  • Real-time problem solving: Talk through complex, personalized decisions like packing, troubleshooting, or planning live.

  • Interactive learning and tutoring: Learn new skills conversationally with dynamic feedback, like practicing languages or solving math problems.

  • Creative collaboration: Brainstorm stories, songs, or projects in real time like working with a creative partner.

Voice AI Isn’t Just for Personal Use — Companies Are Leveraging It Too
More businesses are using Voice AI to handle outreach, qualify leads, and boost productivity. Here's how one company made it happen:

How Smartcat Scaled Outreach and Cut Costs

Smartcat’s sales team needed a better way to qualify leads and book demos. By partnering with Synthflow, they deployed Voice AI Agents that increased call engagement, revived cold leads, and reduced booking costs by 70%. The result? More deals closed, and reps focused on what matters most—selling.

Heard in the Server Room

Look out, Nvidia—China's Huawei is coming for a seat at the AI table. The tech giant is preparing to test its new Ascend 910D AI processor, part of its push to build homegrown alternatives to Nvidia’s prized—and now export-controlled—chips. While earlier Huawei processors couldn’t match Nvidia’s raw power one-on-one, the company is shifting its strategy. It now focuses on the ability to link hundreds of chips into large computing clusters that could scale up to rival Nvidia-powered datacenters. Ironically, the restrictions meant to slow China’s AI progress may have opened a major domestic market for Huawei’s semiconductor business, but the jury is out on if the new chips are up to the task.

Anthropic CEO Dario Amodei is sounding the alarm in his new essay on AI interpretability, highlighting how researchers still can't fully explain AI decision-making processes—a knowledge gap he aims to narrow by 2027. His ambitious goal? For Anthropic to detect and understand most AI problems before we reach more powerful milestones, something he argues requires both industry-wide research efforts and thoughtful government oversight. While Anthropic has made early progress identifying some internal AI circuits, Amodei acknowledges that fully understanding these systems remains a 5-10 year journey.

In a legal face-plant, MyPillow CEO Mike Lindell's lawyers are facing the judge's wrath after filing a document riddled with AI-generated errors in a defamation case. The judge, who identified nearly thirty incorrect citations, is now demanding Lindell's attorneys explain why they shouldn't face sanctions. Lawyer Christopher Kachouroff claims they accidentally submitted the wrong draft and argued the court should have given him a chance to review the filing before grilling him on an “unfamiliar document.” Maybe MyPillow can use this opportunity to launch a new line of scream pillows.

Notion + AI: A Smarter, More Powerful Way to Work

Even if you haven’t used Notion yet, you’ve probably heard about it.
It’s one of the most powerful (and easy-to-use) tools for organizing projects, planning goals, and now — tapping into unlimited AI. Whether you’re running a startup, managing your side projects, or just trying to keep life a little less chaotic, this is a deal worth checking out:

Want unlimited AI, ASAP?

Get up to 6 months of Plus plan + unlimited AI free!

Launch and scale your startup faster with Notion.

Visit the Notion for Startups page to get the offer.

The Owl Has Spoken: More AI, Fewer Humans

Another major CEO is pushing AI from the top: Duolingo’s Luis von Ahn just declared the company officially “AI-first,” setting major changes into motion. Our favorite green owl may still look the same, but big shifts are happening behind the scenes:

  • AI is replacing contract workers: von Ahn made it clear — "Headcount will only be given if a team cannot automate more of their work."

  • Systems are being rebuilt from scratch: “Making minor tweaks to systems designed for humans won’t get us there,” he said.

  • AI is seen as essential to scaling: Without it, von Ahn argued, it would take “decades” to manually grow Duolingo’s learning content.

  • Speed over perfection: “We can’t wait until the technology is 100% perfect. We’d rather move with urgency and take occasional small hits on quality than move slowly and miss the moment.”

Like Shopify and Amazon, Duolingo shows that CEOs aren’t just dabbling in AI — they’re mandating it as the new foundation.

That’s it for today — have a great week, and we’ll catch you Friday with more hot AI takes.

How did you like today's newsletter?

Login or Subscribe to participate in polls.

  • ❓Have a question or topic you’d like us to discuss? Submit it to our AMA!

  • ✉️ Want more Neural Net? Check out past editions here.

  • 💪 Click here to learn more about us and why we started this newsletter

  • 🔥 Like what you read? Help us grow the community by sharing the Neural Net!