- AI Enthusiasts by Feedough.com
- Posts
- How 5-Second Calls Can Hijack Your Voice
How 5-Second Calls Can Hijack Your Voice
Your voice at risk, AI that ‘feels’ your mood, Singularity in 4 years & DNA decoded
Hey there
Ever been on a call where someone says, "Let me record 5 seconds of your voice," and then they can make you say literally ANYTHING?
Well, that nightmare scenario just became reality.
Bland AI just released voice cloning tech that can:
Record 5 seconds of someone's voice over a phone call
Clone it perfectly with emotional inflections
Make them 'say' anything you want
Deploy in AI phone calls
Yep, voice verification as a security measure? Effectively dead.
As someone building in the AI space, this both terrifies and fascinates me. The scam potential is off the charts, but so are legitimate use cases, such as accessibility tools and personalized learning systems.
And this is just ONE of the wild AI developments that dropped this week.
Here's what else caught my attention (we discuss many more in our WhatsApp group, though):
Speech-to-Speech Gets Eerily Human
Hume just launched EVI 3, an AI that can not only mimic any voice but also pick up on your mood, your tone, rhythm, and vibe.
It beat voice assistants like GPT‑4o and Gemini in blind tests. Real people said EVI 3 sounded more natural, empathetic, and expressive, and answered in under a second.
Think of it like calling a friend, not scheduling your tone like "say it sad" or "happy." It just knows and adapts.
Combine that with emotion-aware tech, and the era of "that AI voice" is ending fast.
10X AI Improvements Coming Fast
Ray Kurzweil, the singularity guy, just claimed that AI hardware and software are about to improve 10x in the next year.
Ten times better. In a year.
He's been mapping technology's exponential growth for decades. His new Singularity timeline? 2029 instead of 2045.
That's not a small shift. That's a wake-up call.
It means human-level AI might be here in just four years.
And his message is clear:
If we don't teach AI in colleges today, we're preparing people for a world that doesn't exist.
Whether you agree with his exact timeline or not, one thing's for sure—AI isn't waiting around for us to catch up.
It's happening.
And fast.
Clean 'Human' Data is Gaining Trend
Clean data is finally in the spotlight.
EleutherAI just released a huge new dataset — 8 terabytes of text that's fully licensed and written by real people.
No spammy web scraping. No copyright headaches. Just solid, authentic content.
China's Xiaohongshu is also making waves. Their new language model was trained only on real, high-quality writing. No fake or machine-made content. And they're sharing the model for everyone to use.
Compare that to other big AI players like DeepSeek and GPT.
Those models gobble up data from everywhere, including websites that aren't always reliable. They even train on content that other AI wrote. It's like copying your homework from a friend who copied from someone else.
That shortcut saves time. But it can make the models worse over time.
Clean data might take more effort to gather. But it keeps AI honest.
In a world drowning in spammy content, quality is everything.
AI Decodes the 3D Structure of DNA
AI is now piecing together the shape of life itself.
Scientists at Skoltech just used generative AI to map the 3D structure of our DNA.
Before this, microscopes could only give us fragments. Like trying to solve a puzzle with half the pieces missing.
Now? AI fills in the blanks.
This matters because how DNA folds can flip disease switches, like cancer or genetic disorders.
With this tech, we can see which genes misfire, and maybe even fix them.
Unexpected AI Use: Deciphering Dwarf Fortress
DeepSeek R1 tackled text from Dwarf Fortress, one of the densest, most chaotic simulation games out there. It didn't just read the logs. It understood them.
One Reddit user explained how they fed screenshot text into DeepSeek R1 and asked it to describe dwarf behavior.
Imagine applying this to legal docs, medical records, or financial reports. If Dwarf Fortress chaos becomes clear, so can your most tangled datasets.
Anyway, the real question is: What are YOU building with these tools?
And more importantly, is what you're building going to be made obsolete by next month's developments, or are you focusing on uniquely human needs that these technologies can enhance?
Let me know what you're working on. I'm always looking to connect with people building interesting things in this space.
Until next time,
-AP
P.S. If you found this valuable, would you mind forwarding to ONE person who might also benefit? The best ideas spread through trusted connections, not algorithms.
Your Voice AI Guidebook is Here
Thinking about Voice AI for your contact center? Discover how leading contact centers are making smarter Voice AI decisions.
This guide walks you through the key trade-offs between vendors and model types, shows how to reduce latency and cut inference costs, and outlines how to deploy a fully optimized solution in just weeks—not months.
Whether you're building or buying, get the clarity you need to move fast and scale with confidence.