AI Voices Are Getting Too Real—Should We Be Worried?

📌TL;DR

• Sesame’s AI voice sounds freakishly human—but can it scale?

• AI still changes its answers unpredictably—so much for perfect prompting.

• We want AI that sounds real and thinks reliably—but we’re not there yet.

• More AI news…

Estimated reading time: 5 - 6 minutes.

TOGETHER WITH HUBSPOT

Use AI as Your Personal Assistant

Ready to save precious time and let AI do the heavy lifting?

Save time and simplify your unique workflow with HubSpot’s highly anticipated AI Playbook—your guide to smarter processes and effortless productivity.

Download the free guide today.

PURR-FECT CATCH

Sesame’s AI Voice Sounds Uncannily Human

AI voices have never sounded this real.

San Francisco-based startup Sesame just dropped a voice model so natural and expressive that some users say it’s unsettling.

PCWorld’s Mark Hachman literally had to step away from his computer because the AI reminded him of an old friend.

The Verge’s Sean Hollister called it the first AI assistant he actually enjoyed talking to.

This isn’t your usual robotic text-to-speech.

Sesame’s Conversational Speech Model (CSM) is built to sound human… REALLY human.

Why Is Everyone Freaking Out?

🔊 It’s Expressive
The AI pauses, changes tone, and reacts emotionally based on what’s being said.

💬 It Remembers
Unlike most AI that forgets the conversation mid-way, this one keeps track so it sounds natural.

⏳ It’s Fast
The model responds in under 500ms (average 380ms), meaning it feels real-time.

🧠 It’s Smart
No need to over-explain things. The AI understands context and adjusts mid-conversation like a human.

👓 AI Glasses Incoming?
Sesame is developing lightweight AI glasses that could pair with this voice model, essentially creating a real-time AI companion that sees what you see. (Let’s just say privacy concerns are loading…)

Right now, CSM only supports English, but Sesame says 20+ languages are coming in the next few months.

My Take:

I checked out the voice samples on Sesame’s X (Twitter), and yeah… it’s impressive.

The way it expresses emotion, emphasizes words, and sounds completely natural is probably the best I’ve heard so far.

It’s just as good—if not better—than Google’s NotebookLM voice.

But here’s the big question: Can this work at scale?

• Latency isn’t an issue (380ms is fast), so that’s a win.

• It’s still English-only—multi-language support is planned, but how well it adapts is TBD.

• Cost matters. High-quality AI voice generation isn’t cheap. Will this be affordable for creators and businesses?

Right now, Sesame’s AI voice is insanely good.

And you can try the demo out here.

But rolling it out in real-world apps, across different languages, and without breaking the bank?

That’s the real test.

A MESSAGE FROM THE RUNDOWN AI

Learn how to make AI work for you

AI won’t take your job, but a person using AI might. That’s why 1,000,000+ professionals read The Rundown AI – the free newsletter that keeps you updated on the latest AI news and teaches you how to use it in just 5 minutes a day.

Sign up to start learning.

EXTRA TREAT

AI Prompting Myths Debunked

If you’ve spent any time in AI circles, you’ve probably seen countless "expert" takes on prompt engineering.

Use the right phrasing, be polite, and AI will reward you with better answers.

Except… turns out that’s not entirely true.

I chanced upon an article that Neuron wrote on this topic (shout out to Neuron, an AI newsletter that’s also obsessed with cats!), and they covered some surprising findings from a recent AI study.

After running 19,800 tests on GPT-4o and GPT-4o mini, researchers found that a lot of what we assume about prompting AI is wrong.

What the Study Actually Found

🔹AI is inconsistent
Ask it the same question 100 times, and don’t be surprised if it changes its mind.

🔹 Politeness is unpredictable
Saying “please” sometimes improves responses… and sometimes makes them worse.

🔹 Formatting matters more than we think
AI performed noticeably better when given structured, well-formatted prompts.

🔹 AI benchmarking is flawed
The way we measure AI’s accuracy dramatically changes its perceived performance.

At first glance, this might seem like a minor technicality.

But for anyone who relies on AI—whether for writing, research, or content creation—it raises an important question:

If AI’s performance changes based on how we prompt and test it, how do we actually trust the answers it gives us?

The Bigger Problem

One of the biggest misconceptions about AI is that it’s like a calculator.

You input something, and it gives you the right answer.

In reality, AI is more like a very smart intern who sometimes gets things completely wrong but says them with absolute confidence.

This study highlights something that AI users have probably experienced firsthand:

• You ask AI a question. It gives an answer.

• You rephrase it slightly. It gives a different answer.

• You ask it a third time, and suddenly it’s convinced you were wrong all along.

That’s not exactly the reliability we’d hope for from AI that’s increasingly being used for everything from research to decision-making.

What Does This Mean for AI Users?

✅ Test your prompts, don’t just assume
If an answer seems off, rephrase your question or ask multiple times.

✅ Politeness isn’t the secret hack
Clarity and structure matter way more than saying “please.”

✅ Format your prompts properly
AI performs best when given clear instructions with structured formatting.

AI is evolving fast, but studies like this are a good reminder that we’re still figuring out how to work with it effectively.

BYTE-SIZED BUZZ

Here’s a quick roundup of what’s making waves in the AI world this week.

🎬 Sora Video AI Is Coming to ChatGPT
OpenAI is bringing Sora, its AI video-generation tool, directly into ChatGPT. That means AI-powered video clips without switching apps.

But there’s a catch. This ChatGPT version won’t have Sora’s full editing suite. It’s designed for quick, simple video generation rather than high-level production.

The Big Deal: AI video inside ChatGPT could speed up content creation, but will it be a true creative tool or just another “cool but kinda useless” feature?

🛒 Opera’s AI Browser Operator: The Future of Browsing?
Imagine a browser that does the clicking for you; booking flights, finding deals, or handling research while you sip your coffee.

That’s the promise of Opera’s Browser Operator, an AI agent that interacts with websites on your behalf.

Unlike cloud-based assistants, this one runs directly on your device, making it faster and (supposedly) more private.

The Big Deal: AI-powered browsing sounds amazing—but will it work as advertised, or are we about to see accidental bulk purchases of socks?

📱 Deutsche Telekom’s AI Phone with Magenta AI
Deutsche Telekom (T-Mobile’s parent company) is teaming up with Perplexity, Google Cloud AI, ElevenLabs, and Picsart to create an AI-powered phone that ditches traditional apps.

Instead of hopping between Uber, Gmail, and Amazon, this phone’s Magenta AI will book your flights, send emails, and shop for you… just by voice command.

The Big Deal: If done right, this could be a huge leap in AI-driven interfaces. But after the Rabbit R1 disaster, let’s just say… skepticism is warranted.

🔊 ElevenLabs’ Scribe: Speech-to-Text That (Finally) Works
Transcriptions are about to make sense for once!

Thanks to Scribe, ElevenLabs’ new AI model that supports 99+ languages, detects who’s speaking, and even tags background sounds like laughter or applause.

It’s also beating OpenAI’s Whisper and Google Gemini 2.0 Flash in accuracy tests.

The Big Deal: If AI captions and transcriptions finally stop butchering sentences, this could be a game-changer for podcasters, content creators, and accessibility advocates.

📸 Google Gemini Live: Real-Time AI Assistance
Forget typing. Google’s Gemini Live now lets you point your camera at anything for real-time AI help.

Need fashion advice? Show Gemini your outfit. Fixing something? Aim your camera at the mess and pray for a good suggestion.

It’s rolling out this month for Gemini Advanced subscribers, but naturally, privacy concerns are in the air.

The Big Deal: This could make AI incredibly useful—or it could be the moment we all realize we’ve given Google a literal window into our lives.

SUGGESTION BOX