Block Says Dessa Learned From Speech Alone | AI News Briefing

Welcome to your Superpower Daily AI Briefing We are unpacking four major tech developments you absolutely need to know for today October 26th 2026 Let's get right into it Block's new speech only voice AI an MIT swimming robot powered by actual living muscle GrokImagine's brand new Odyssey episode and crucial expert advice on avoiding a terrifying AI voice kidnapping scam Block's speech only voice AI Dessa learns from audio alone Block just introduced Dessa and it's a huge deal

because it was built from scratch using only human conversation audio No text data no transcripts zero pre trained models That means Dessa is what we call full duplex Instead of acting like a walkie talkie where you speak wait and then it responds Dessa can listen and speak simultaneously It processes your pauses handles interruptions and throws in little acknowledgements like yeah or got it right in the middle of your sentence without waiting for a rigid handoff It's

engineered to feel exactly like a real human conversation Block argues that when we convert speech to text we throw away a massive amount of context A text transcript completely loses sarcasm a frustrated sigh or a cracking voice To prove their point they released some wild company produced demos In these recordings you hear Dessa laughing mid sentence instantly raising its volume when asked to speak up and even reacting with a totally convincing emotional gasp And get this

Block says this is all running faster than real time completely locally on a high end MacBook Pro M4 Max But here's the crucial catch Block is keeping early access strictly limited to a wait list and a handful of safety researchers Why Because while Dessa sounds unbelievably expressive we simply don't know how efficiently a speech only model actually learns hard facts or acquires new languages compared to traditional text trained models The researchers are digging into potential risks

right now like testing if the AI can reliably spot genuine human distress instead of just coming back with a totally inappropriate response Muscle powered swimming robot MIT engineers use living cells and light Moving over to the physical engineering world MIT researchers just dropped a fascinating study in advanced functional materials They've built a tiny flexible robot that is actually propelled through water by a thin layer of living engineered muscle To pull this off the team had to

build a very specific foundation First they created a firm half millimeter thick gelatin base called Gelma It had to be rigid enough to support the tissue but light enough not to sink Next they used microscopic square bottomed grooves to perfectly align the living muscle cells so they'd fuse and contract in unison And finally they basically sent this tissue to the gym strengthening the muscle with repeated light pulses before putting the whole thing together And the results

are genuinely incredible They genetically modified these muscle cells to react to light meaning researchers could just shine a beam on the robot's fins to make them flap At its absolute maximum speed this tiny biohybrid prototype swam approximately four times its own body length in a single minute But we have to be clear about an important limitation here We are looking at externally guided swimming versus independent route finding The robot did successfully navigate a simple maze sure

but it only did that because a human researcher was manually moving a light source by hand above the water to steer it Right now this prototype possesses zero autonomous navigation capabilities So while it's an absolute breakthrough for thin tissue propulsion it is not an independent underwater vehicle just yet Grok Imagine's Odyssey episode a seven minute AI adaptation debuts Looking at the wild intersection of AI and Hollywood Grok Imagine just dropped a seven minute AI generated adaptation

of Homer's ancient Greek epic The Odyssey straight onto X Let's look at the timeline to contextualize this Back in July of 2026 Elon Musk pledged that Grok Imagine would deliver a full length historically accurate movie of The Odyssey This new October 23rd release is explicitly labeled as Episode 1 giving us our first real look But it is just a short slice The concrete deadline for the promised full feature is the end of 2026 and that is

fast approaching What's interesting is this release actually credits human creative partners specifically Wonder Studios and director Dev Walker But early critical response is proving that AI filmmaking is a lot harder than just generating pretty pictures TechCrunch's Anthony Ha reviewed it and pointed out that while the ocean shots looked incredibly real the film ultimately suffered from a quote long voiceover introduction stilted dialogue and lack of narrative momentum It really highlights that while Grok Imagine can nail the

convincing visuals the actual core storytelling human interest and emotional weight are still falling flat Finally story four AI voice kidnapping scam A mother loses thousands in a terrifying ordeal This is a deeply disturbing trend we all need to be aware of A mother named Deborah Del Mastro was recently targeted by a fake virtual kidnapping scam Using a voice she swore was an exact clone of her daughter Sarah Because Sarah had just gone through a bad breakup

the crying on the other end of the line was totally convincing to her mom The scammer started by demanding $20,000 They kept Del Mastro trapped on an open phone line for five and a half hours They threatened her they monitored her every move and they even dictated a cover story for her to tell bank tellers In the end they extorted $5,400 across four separate transactions using MoneyGram and Western Union And the truly devastating part her daughter

was perfectly safe at work the entire time Del Mastro only found out when the scammers finally let her off the hook and she called her daughter directly So how do we safeguard against this The Federal Trade Commission warns that scammers literally only need a few short seconds of audio from a public social media video to perfectly clone a voice Because of this the FBI and researchers like Kimberly Hyun at the University of Cincinnati are telling families

to take action today First establish a private family code word immediately Second if you get a call like this always try to reach your loved one on a known number or call someone who you know is with them And most importantly you absolutely have to break the caller's high pressure tactics long enough to independently verify the emergency Just recognizing a familiar voice is simply no longer a reliable identity check Let's synthesize today's explainer to lock in

our key takeaways One Block's DESA shows massive promise for emotional full duplex voice AI but we still need to see if it can reliably learn hard facts Two MIT's light powered muscle robot is an engineering marvel for thin tissue propulsion but it still lacks autonomous navigation Three Grok's Odyssey project has great visuals but is struggling with basic storytelling as its feature film deadline looms And four the terrifying rise of AI voice scams means you absolutely must

rely on independent verification and family code words It leaves us with a critical thought to end on As artificial intelligence continues to perfectly mimic the voices of the people we love and generate entire visual worlds from scratch how will we adapt our fundamental baseline for trust in our daily lives Read more at superpowerdaily com Stay curious