Latest / Connecting the Dots / AI Makes Music and Moves Faster Than Ever

Transcript

AI Makes Music and Moves Faster Than Ever · Connecting the Dots

• 0:00
  Today's tech news moves fast, but understanding
• 0:03
  it shouldn't be hard. That's why Snarful Solutions
• 0:06
  Group brings you Connecting the Dots, where we
• 0:09
  break down what's happening in tech and why it
• 0:11
  matters to you. Now, let's dive in with your
• 0:14
  hosts. Good morning, everybody. Welcome back
• 0:17
  to another edition of Connecting the Dots. I'm
• 0:19
  Alex. And I'm Morgan. And we are coming to you
• 0:21
  live from the very rainy headquarters of Snarful
• 0:25
  Solutions Group right here in Sacramento. Honestly
• 0:30
  putting it lightly it is pouring out there Yeah,
• 0:33
  it really is if you're just tuning in today is
• 0:36
  Tuesday February 24th 2026 and you are definitely
• 0:41
  gonna want an umbrella today. Absolutely. We're
• 0:43
  looking at a high of just 61 degrees here in
• 0:46
  Sacramento with rain likely all day. Which is
• 0:49
  all thanks to this massive atmospheric river
• 0:52
  that's basically just parked itself over northern
• 0:54
  California right now. Right, and they're talking
• 0:56
  about flash flood threats too, so please be careful
• 0:59
  on your morning commute. But you know, it's not
• 1:01
  just us dealing with the extremes today. If you're
• 1:03
  listening from the Cascades or up in the Northern
• 1:06
  Rockies, you're getting hit with heavy mountain
• 1:08
  snow. And then over in the Great Lakes they've
• 1:10
  got a clipper system just dumping snow on them
• 1:13
  right now. Yeah, but then you look at the south
• 1:15
  central U .S. The complete opposite. Exactly.
• 1:18
  Down in southeast Colorado, the southern plains,
• 1:21
  the Ozarks. It's windy. It's dry. High fire weather
• 1:24
  concerns down there. Yeah. So just a real mix
• 1:27
  of extremes across the national map today. For
• 1:29
  sure. Well, let's pivot to something a little
• 1:31
  more stable, hopefully. It's time for the Money
• 1:34
  Minute. What are the markets doing this morning?
• 1:36
  Well, looking at yesterday's close, the Dow Jones
• 1:38
  had a decent day. It closed at 49 ,132 .76. Nice!
• 1:44
  Yeah, that's up 0 .62%. The S &P 500 also ticked
• 1:49
  up just a bit, about 0 .14%, to close at 6 ,847
• 1:54
  .12. Okay, so traditional markets are holding
• 1:56
  steady. What about crypto? Because my feed was
• 2:00
  buzzing a bit this morning. Yeah, a little more
• 2:02
  turbulence there. Bitcoin is taking a bit of
• 2:04
  a hit. Unsurprising. Right. It's down 2 .15%,
• 2:08
  hovering roughly around $63 ,240. Okay, so if
• 2:12
  you're holding, you know, you might be feeling
• 2:13
  a little pinched today. Just a little one. But
• 2:15
  usually when we see a dip like that, without
• 2:17
  some major regulatory news, it's just standard
• 2:20
  volatility. Still something to keep an eye on
• 2:22
  though. Definitely. And hey, if looking at that
• 2:24
  crypto chart has you feeling a little sluggish
• 2:26
  or stressed this morning, we actually have the
• 2:28
  perfect fix to get you back on track. We absolutely
• 2:31
  do. This show is brought to you by strong coffee
• 2:35
  listen We all need energy to get through the
• 2:37
  morning routine, but I personally cannot stand
• 2:40
  the jitters you get from regular coffee. Nobody
• 2:42
  likes the jitters. No, that is why I love strong
• 2:46
  coffee. It is premium instant coffee. And I know
• 2:49
  instant usually sounds like a compromise, but
• 2:51
  this genuinely tastes like a traditional brew.
• 2:53
  It really does. And it's packed with 15 grams
• 2:55
  of protein, MCTs and adaptogens like ashwagandha
• 2:59
  to give you clean, sustained energy without the
• 3:02
  jitters. Promo code SNARFL gets you 20 % off
• 3:05
  your entire purchase at strongcoffeecompany .com.
• 3:08
  Alright, the coffee is brewed, the mics are hot.
• 3:10
  Let's connect the dots on today's top three stories.
• 3:13
  Let's do it. We are opening up The Morning Star,
• 3:16
  which is our internal SnarfleTech article report,
• 3:18
  and the first story today is huge for anyone
• 3:21
  who has ever hummed a melody and thought, you
• 3:23
  know, I could be a pop star. Or, honestly, anyone
• 3:26
  who's tried to start a band and realized that
• 3:28
  getting four adults in the same room is physically
• 3:31
  impossible. Right. Scheduling is the real enemy
• 3:34
  of art. Truly. So the headline here is that Google
• 3:37
  has officially acquired producer AI and they
• 3:40
  are folding it directly into Google Labs. Which
• 3:43
  is a big move. It is. But we've seen a million
• 3:48
  of these AI music generators over the last year.
• 3:51
  What makes this specific one worth buying? So
• 3:54
  that's the crucial question. According to the
• 3:56
  Google Labs team, the big differentiator is that
• 3:59
  producer AI isn't just a slot machine. A slot
• 4:02
  machine? Yeah, like most generative AI up to
• 4:04
  this point has been what they call prompt and
• 4:06
  pray. Oh, right. You type in sad acoustic song
• 4:09
  about a lost dog, you pull the lever, and whatever
• 4:12
  it spits out is what you get. Exactly. And if
• 4:14
  the guitar sounds great, but the lyrics are weird,
• 4:16
  you're out of luck. You just have to roll the
• 4:18
  dice again. Right. But producer AI is built entirely
• 4:21
  around collaboration. You are actually chatting
• 4:24
  with an AI agent that's powered by Gemini. So
• 4:26
  it's more like having a co -producer sitting
• 4:28
  in the studio with you. Yeah. You start with
• 4:31
  an idea, and then you just talk to it. You can
• 4:33
  workshop the lyrics together. You can say, hey,
• 4:35
  take this track, but isolate the drums and make
• 4:37
  them punchier. Oh, wow. So it has stems. Exactly.
• 4:40
  You can swap out a single instrument or rewrite
• 4:42
  just the bridge, and the AI iterates with you
• 4:45
  back and forth. That is massive. Because the
• 4:48
  granular control has always been the missing
• 4:50
  piece for professionals. Totally. And the tech
• 4:53
  stack behind this is wild. It's powered by a
• 4:56
  preview of Google's Liria 3 model, but they've
• 4:59
  also integrated NanoBanana. NanoBanana, I love
• 5:02
  your startup name. I know, it's so good. But
• 5:04
  NanoBanana is for generating the album art, and
• 5:07
  then they've also tied in Vio for generating
• 5:09
  music videos. So wait, you can write the song,
• 5:12
  produce the beat. make the album cover, and shoot
• 5:15
  a music video all in one interface. All on your
• 5:17
  phone. It's basically an entire record label
• 5:19
  in your pocket. That's insane. And they've got
• 5:21
  some serious industry backing too, right? Because
• 5:23
  I saw a quote from Alex Paul from the Chainsmokers.
• 5:26
  Yeah, he's totally on board. He said the tool
• 5:28
  is crafted around the musician's experience,
• 5:31
  which is high praise from someone who actually
• 5:33
  writes massive hits. Right, and it's rolling
• 5:35
  out fast. It's already live at over 250 countries.
• 5:38
  They're doing free credits to start, and then
• 5:40
  subscriptions range from $8 for a starter tier
• 5:43
  up to $64 a month for the heavy users. So I guess
• 5:46
  this means we can finally launch that band we've
• 5:49
  been joking about? The snarfle sounds. The snarfle
• 5:53
  sounds. I mean, our biggest issue has always
• 5:55
  been that neither of us can play an instrument
• 5:57
  to save our lives. Speak for yourself. I was
• 6:00
  elite at the recorder in third grade. Oh, my
• 6:02
  apologies. I forgot I was sitting next to a prodigy.
• 6:05
  Thank you. But seriously, having an AI producer
• 6:09
  that never gets annoyed when you ask it to change
• 6:12
  the bass line for the 30th time. That lowers
• 6:16
  the barrier to entry so much. It completely democratizes
• 6:19
  creativity. You don't need to rent a studio for
• 6:22
  a thousand dollars a day. You just need a prompt
• 6:24
  and some good taste. Which is amazing, but of
• 6:27
  course this brings us to our second story, because
• 6:29
  there's always a catch in tech. We just talked
• 6:31
  about this incredible producer AI interface,
• 6:34
  but the actual engine running underneath it,
• 6:36
  that Lyria 3 model you mentioned, it has some
• 6:39
  really strict guardrails. Yeah, this is the reality
• 6:42
  check portion of the morning. Because we've seen
• 6:44
  the shiny interface, but Google is playing it
• 6:47
  incredibly safe with the engine itself. Safe
• 6:50
  how? Because competitors like Suno and Udio are
• 6:53
  out there generating full three -minute radio
• 6:56
  -ready songs. Right. But Lyria 3 has a hard 30
• 6:59
  -second ceiling. 30 seconds? Yep. That's not
• 7:02
  a song. That's a jingle. Or like a ringtone?
• 7:05
  It's very short. And Google's official reasoning
• 7:07
  is that they are focusing on shareable moments.
• 7:10
  Think TikTok sounds, YouTube shorts, or their
• 7:13
  new DreamTrack feature. Okay, sure. They're explicitly
• 7:16
  not trying to help you write the next Billboard
• 7:18
  Hot 100 hit right now. They want to help you
• 7:20
  make a catchy hook for your vacation vlog. Right,
• 7:22
  but we both know that's only half the story.
• 7:25
  The real reason for that 30 -second limit is
• 7:27
  legal fear. Oh, 100%. The copyright lawsuits
• 7:30
  are breathing down their necks. Yeah. For anyone
• 7:33
  who hasn't been following the legal drama, the
• 7:36
  RIAA, the Recording Industry Association of America,
• 7:39
  is currently suing Suno and Udio into the ground.
• 7:42
  They really are. And the whole argument centers
• 7:44
  on training data. The major labels are saying,
• 7:47
  look, you scraped decades of copyrighted music.
• 7:50
  everything from Michael Jackson to Taylor Swift
• 7:53
  to teach your AI what music sounds like. And
• 7:56
  they view that as massive scale copyright infringement.
• 7:59
  Exactly. The RIAA stance is essentially, you
• 8:02
  stole our artists' work. to build a machine that's
• 8:05
  going to replace our artists. And the AI companies
• 8:06
  are fighting back using the fair use defense.
• 8:09
  They're arguing that the AI isn't directly copying
• 8:12
  and pasting the songs. It's just listening to
• 8:14
  them and learning the mathematical patterns of
• 8:17
  genres, which they argue is the exact same thing
• 8:20
  a human musician does when they listen to the
• 8:22
  radio and get inspired. It's the Napster debate
• 8:24
  all over again, but on steroids. It really is.
• 8:27
  And the stakes are astronomical. The statutory
• 8:30
  damages could be up to $150 ,000 per infringed
• 8:34
  work. Which would bankrupt almost any startup.
• 8:36
  In fact, didn't IDEO just pivot? Yeah, IDEO just
• 8:40
  announced they're pivoting to a fully licensed
• 8:41
  model, presumably to settle with the labels and
• 8:44
  stay alive. Wow. So with all that happening,
• 8:47
  Google is looking at the landscape and saying,
• 8:49
  we are Way too big of a target to mess around
• 8:52
  with this exactly Google has infinitely deeper
• 8:55
  pockets than these startups a Multi -billion
• 8:58
  dollar lawsuit is not worth the risk for them
• 9:00
  right now So leery 3 is the corporate safe legally
• 9:04
  bulletproof version of AI music which explains
• 9:07
  the other limitations, too because they explicitly
• 9:10
  prevent the model from mimicking specific artists.
• 9:13
  Right. You can use an artist's name to describe
• 9:16
  a mood, but you can't say, generate a new Drake
• 9:18
  song. And I also read that it heavily favors
• 9:20
  mainstream genres. Like it's great for pop, R
• 9:23
  &B, Afrobeat, but if you try to get weird and
• 9:26
  prompt it for like underground Mongolian throat
• 9:28
  singing mixed with dubstep. It kind of just steers
• 9:30
  you back to a standard pop beat. Like a nervous
• 9:33
  radio DJ. Exactly. And the final piece of the
• 9:35
  safety puzzle is Synth -A -D. Every single piece
• 9:38
  of audio generated by Lyria 3 is watermarked
• 9:41
  at the audio wave level. Which is imperceptible
• 9:43
  to the human ear, right? Right. You can't hear
• 9:46
  it, but detection tools can scan it and instantly
• 9:48
  tell you, yes, this was generated by Google AI.
• 9:51
  I honestly think that's the most important part
• 9:53
  of this whole rollout with the amount of deep
• 9:56
  fakes and AI voice cloning out there, having
• 9:59
  that digital fingerprint is crucial for trust,
• 10:03
  especially for parents who are worried about
• 10:05
  what media their kids are consuming. It establishes
• 10:08
  provenance. It's just necessary at this point.
• 10:10
  So to sum it up, Google is giving us an incredible
• 10:13
  high quality studio in our pocket, but they are
• 10:16
  absolutely keeping the training wheels on. You're
• 10:18
  getting an appetizer, not the full meal. An appetizer
• 10:21
  that you can't get sued over. Fair enough. Okay,
• 10:24
  let's connect the dots to our final story, which
• 10:26
  moves us away from music and into the fundamental
• 10:29
  way artificial intelligence actually processes
• 10:32
  information. This is the nerdy stuff and I love
• 10:34
  it. It is. We are looking at a startup called
• 10:37
  Inception, which just raised $50 million to basically
• 10:40
  change the speed of thought for AI. And just
• 10:43
  to clarify, no relation to the Christopher Nolan
• 10:45
  movie with the spinning top, though the concept
• 10:47
  is almost as mind -bending. Right, so this company
• 10:50
  is led by Stefano Ehrman, a professor out of
• 10:53
  Stanford, and they're introducing a new model
• 10:55
  called Mercury 2. Which is specifically designed
• 10:58
  for coding, but the underlying tech is what's
• 11:01
  revolutionary here. Because they're moving from
• 11:03
  autoregressive models to diffusion models. Yes.
• 11:06
  Okay, we need to unpack that for the folks listening
• 11:07
  who aren't machine learning engineers. What does
• 11:10
  autoregressive mean in the context of, say, chat
• 11:13
  GPT or Gemini? Okay, so... Think about how a
• 11:16
  typewriter works, or even just how you and I
• 11:18
  are speaking right now. Sequentially. Right.
• 11:20
  One word at a time. The current AI models predict
• 11:24
  the next token or word based on the words that
• 11:27
  came before it. It's a linear process. It physically
• 11:30
  cannot generate the end of a sentence until it
• 11:32
  has generated the beginning. Like laying bricks
• 11:35
  one by one to build a wall. Exactly. But Inception
• 11:38
  is throwing that method out and using diffusion
• 11:40
  for text. Now, I know diffusion from image generators,
• 11:44
  like mid -journey or delay, where it starts a
• 11:47
  static noise and then slowly sharpens into an
• 11:50
  image of cat riding a skateboard. But how do
• 11:52
  you apply that to words? Because pixels can blend
• 11:55
  together, words can't really be blurry. And that
• 11:58
  has been the holy grail of engineering for the
• 12:00
  last two years. Pictures are continuous data,
• 12:04
  text is discrete data. But Inception finally
• 12:07
  cracked the math on this. So what does it look
• 12:10
  like? Instead of typing out an answer word by
• 12:13
  word, the model generates the entire response
• 12:16
  simultaneously. The whole thing at once. The
• 12:18
  whole block of text. It drafts a blurry structural
• 12:21
  version of the whole answer and then refines
• 12:24
  it in parallel until every word is sharp and
• 12:27
  correct. So if the old way was laying bricks,
• 12:29
  This is like 3D printing the entire wall in one
• 12:32
  shot. Or developing a Polaroid picture, the entire
• 12:34
  scene appears at the same time and just gains
• 12:37
  resolution. OK, that is super cool, conceptually.
• 12:40
  But what is the practical benefit? Why did Menlo
• 12:43
  Ventures, Microsoft's M12, and Nvidia just throw
• 12:47
  $50 million at this? Two words, speed and cost.
• 12:50
  Because Mercury 2 isn't constrained by linear
• 12:53
  generation, it is benchmarked at over 1 ,000
• 12:56
  tokens per second. A thousand words a second?
• 12:58
  Roughly, yeah. That is blisteringly fast. It's
• 13:00
  about 10 times faster than current state -of
• 13:03
  -the -art models. And because it computes in
• 13:05
  parallel, it requires drastically less latency
• 13:08
  and compute power, which makes it incredibly
• 13:10
  cheap to run. Okay, I want to push back on this
• 13:13
  a little bit. Go for it. Speed is great, but
• 13:15
  from a consumer perspective, does it really matter
• 13:18
  if my AI writes a work email in 0 .1 seconds
• 13:22
  versus 1 second? It feels like we're just chasing
• 13:24
  benchmark scores at this point. For an email?
• 13:27
  No, you're right, it doesn't matter. But think
• 13:28
  about voice interactions. Oh. Think about talking
• 13:31
  to Siri or a customer service voice agent. Right
• 13:34
  now, there is always that awkward beat of silence.
• 13:38
  You ask a question, you wait one or two seconds
• 13:40
  while it processes, and then it replies. It's
• 13:42
  that uncanny valley of conversation. It constantly
• 13:45
  reminds you that you're talking to a machine.
• 13:47
  Exactly. But with diffusion text, at a thousand
• 13:50
  tokens a second, that latency drops to zero.
• 13:53
  So it's actual real -time conversation. Yes.
• 13:56
  The AI can interrupt you. It can laugh at exactly
• 13:58
  the right millisecond. It can react. Instantly,
• 14:02
  it makes the digital agent feel entirely human.
• 14:05
  And from a business side, if you drop the cost
• 14:08
  of running these models by 90%, you can suddenly
• 14:11
  afford to put a genius level AI in everything.
• 14:15
  Every toy, every appliance, every drive -through
• 14:17
  menu. The economics totally change. And of course,
• 14:20
  NVIDIA is backing this. They never miss an opportunity
• 14:22
  to fund the next leap in compute efficiency.
• 14:24
  They really don't. Moving from the typewriter
• 14:27
  era to the Polaroid era of AI is a massive paradigm
• 14:31
  shift. It really is. And it's wild how these
• 14:34
  stories all connect. You've got AI helping us
• 14:37
  collaborate on music, the legal realities forcing
• 14:39
  that music into 30 second bytes, and now a brand
• 14:42
  new engine that's going to make all of this processing
• 14:44
  happen 10 times faster. The tech landscape is
• 14:46
  just moving at warp speed right now. It definitely
• 14:49
  is. And we'll keep tracking all of it for you.
• 14:51
  But before we wrap up and give you the final
• 14:53
  recap, We want to mention one more sponsor that
• 14:55
  can help you upgrade your real -world style while
• 14:58
  we wait for the digital future. Yes, we do. Big
• 15:00
  shout out to Old Glory. If you have an iconic
• 15:03
  music or sports fan in your life, or let's be
• 15:05
  real, if you are that fan, Old Glory is your
• 15:08
  go -to spot. It's the best. And since we're just
• 15:11
  talking about AI pop stars and these digital
• 15:14
  30 -second songs, Old Glory is actually great
• 15:17
  stuff for your techie friend who still appreciates
• 15:19
  physical media and vintage band tees. Oh, absolutely.
• 15:22
  There is something about the cotton and the print
• 15:24
  of a classic band t -shirt that digital goods
• 15:27
  just can't replace. You can't wear a digital
• 15:29
  watermark. Exactly. And they have a massive selection,
• 15:33
  everything from classic rock to hip hop to sports
• 15:36
  fan merch. So treat yourself or your techie friend.
• 15:40
  If listeners use promo code SNARFL at checkout,
• 15:42
  you can get 15 % off your entire purchase. The
• 15:45
  website is OldGlory .com. Grab a hoodie, stay
• 15:48
  dry in this rain. Seriously. All right, let's
• 15:50
  quickly recap what we covered today for everyone
• 15:52
  finishing up their drive to work. Sounds good.
• 15:54
  So first, we broke down Google's acquisition
• 15:56
  of producer AI. They're bringing an AI studio
• 15:58
  buddy straight to your phone, complete with GeminiChat,
• 16:01
  Nano Banana for art, and VO for video. The chain
• 16:04
  smokers love it and it's shifting music generation
• 16:06
  from a slot machine to a true co -pilot experience.
• 16:10
  Yep. Second, we looked under the hood at Lyria
• 16:13
  3. which is the engine powering that new studio.
• 16:16
  It's high quality, but it's capped at 30 seconds
• 16:19
  to focus on shareable moments and, more importantly,
• 16:22
  to avoid the massive copyright lawsuits hitting
• 16:25
  competitors like Suno and Udio. Safety first
• 16:27
  with synthi watermarks on every track. Exactly.
• 16:31
  And finally, we talked about Inception's new
• 16:34
  Mercury 2 model. They are fundamentally changing
• 16:37
  how AI reads and writes by moving from sequential
• 16:40
  autoregressive models to parallel diffusion models.
• 16:45
  The typewriter to the Polaroid. Right. It makes
• 16:47
  the AI 10 times faster and cheaper, which is
• 16:50
  going to make real time, zero latency voice agents
• 16:53
  a reality very soon. Incredible stuff. You know,
• 16:55
  thinking about all this, one thing really sticks
• 16:57
  with me from that music segment, the idea of
• 16:59
  the 30 second song. Yeah. We're seeing technology
• 17:02
  physically shape the format of art. I mean, historically,
• 17:04
  the three minute pop song only became the standard
• 17:07
  because of the physical limitations of a 10 inch
• 17:09
  vinyl record. That's all the space they had.
• 17:11
  Oh, that's true. So it makes you wonder, are
• 17:13
  we about to see an entire generation of music
• 17:15
  that is designed to be exactly 30 seconds long?
• 17:20
  Simply because that's the safe limit the AI models
• 17:24
  will generate. Wow. The technology dictating
• 17:27
  the culture again. Exactly. The medium is the
• 17:29
  message. It's definitely something to chew on
• 17:31
  today. For sure. Well, that is all the time we
• 17:33
  have for this morning's Exploration of the News.
• 17:35
  Thank you so much for tuning in and letting us
• 17:37
  be a part of your day. Stay dry out there, Sacramento.
• 17:40
  We're here to help at Snarful, so reach out on
• 17:42
  our website if you have any more questions. That's
• 17:48
  a wrap on today's top stories. Remember, tech
• 17:51
  doesn't have to be complicated and we're here
• 17:53
  to help connect the dots. If you've got questions
• 17:56
  or want to learn more, visit us at snarful .com.
• 17:59
  Thanks for listening and we'll catch you next
• 18:02
  time.