Latest / Connecting the Dots / Claude Gets Faster, Honor Goes Bigger, and Sora 2 Steps Into the Spotlight

Transcript

Claude Gets Faster, Honor Goes Bigger, and Sora 2 Steps Into the Spotlight · 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. Welcome back to Connecting the Dots from
• 0:17
  Snarful Solutions Group here in Sacramento, California.
• 0:20
  It is Thursday, October 16th, 2025. Good morning.
• 0:24
  And it feels like a pivotal week, especially
• 0:26
  with AI getting, well, Faster, cheaper, and starting
• 0:29
  to feel much more physical, more integrated into
• 0:32
  the hardware we use every day. Absolutely. Lots
• 0:34
  to connect today. Before we dive into the tech
• 0:36
  news though, let's get a quick check on the conditions
• 0:38
  out there. Sounds good. Here in Sacramento, looks
• 0:41
  like some patchy fog early on, but that should
• 0:44
  burn off for a sunny day. High around 69 degrees
• 0:48
  Fahrenheit. Nice fall day here. Nationally, things
• 0:51
  are a bit more active. There are marginal risks
• 0:53
  for severe storms and maybe some flash flooding
• 0:56
  across the central and northern plains. Okay.
• 0:59
  And also seeing frost and freeze warnings popping
• 1:01
  up in parts of the west and also the northeast.
• 1:04
  So definitely jacket weather in some places.
• 1:06
  Definitely. Okay. Quick look at the markets.
• 1:08
  After Wednesday's close and heading into Thursday
• 1:10
  morning, the Dow Jones and the S &P 500 both
• 1:13
  dipped slightly down about 0 .26 % each. Pretty
• 1:17
  modest moves there. Yeah. Meanwhile, Bitcoin's
• 1:20
  having a bit of a better morning, Thursday morning
• 1:22
  that is, showing a positive daily change up almost
• 1:25
  a full percent, 0 .97 % last I checked. Interesting.
• 1:29
  Little divergence there. Well, tech keeps moving
• 1:32
  regardless. It certainly does. And speaking of
• 1:34
  moving forward, we want to give a big thank you
• 1:36
  to our sponsor, Old Glory. Ah, yes. If you're
• 1:38
  already thinking about holiday shopping, maybe
• 1:40
  for that tech savvy friend who seems impossible
• 1:43
  to buy for. Old Glory has fantastic, iconic music
• 1:47
  and sports fan merchandise. Really unique stuff.
• 1:50
  Good call. And you can get a pretty nice discount
• 1:52
  right now. Exactly. Use the promo code SNARFUL,
• 1:55
  that's S -N -A -R -F -U -L at checkout for 15
• 1:59
  % off your entire purchase. Just head over to
• 2:02
  oldglory .com. Great gear. Great deal. Okay,
• 2:06
  ready to jump into the Morningstar report. Let's
• 2:08
  do it. Story number one feels, well, pretty massive
• 2:11
  in terms of AI efficiency. Anthropic just dropped
• 2:14
  Claude Haiku 4 .5. Right, and this isn't just
• 2:16
  another model update. This feels like a step
• 2:18
  change. Haiku 4 .5 is their latest small model,
• 2:21
  but get this. It's hitting coding performance
• 2:23
  levels similar to Claude Sonnet 4. Which was
• 2:26
  the absolute state of the art, what, only five
• 2:28
  months ago? Exactly, five months. That rate of
• 2:31
  progress for a smaller model to catch up is,
• 2:34
  well, it's stunning. And it's not just performance
• 2:36
  parity, the efficiency gains are the real headline
• 2:39
  here, aren't they? Oh, absolutely. Haiku 4 .5
• 2:42
  delivers that sonnet four -level performance
• 2:45
  at, reportedly, one -third the cost. and it runs
• 2:48
  more than twice as fast. One third the cost,
• 2:50
  twice the speed. That just changes the economics
• 2:52
  completely. It feels like AI's version of Moore's
• 2:56
  Law is really hitting its stride, especially
• 2:58
  on the cost side. It really is. Think about what
• 3:00
  this means for, say, a small startup or even
• 3:02
  just one person building an app. Running large
• 3:05
  models constantly used to be, frankly, way too
• 3:08
  expensive for many. Right. The cost per token
• 3:10
  could eat you alive on high volume tasks. Precisely.
• 3:14
  But Haiku 4 .5, it's built for those real -time
• 3:17
  low latency things. Chat assistance, customer
• 3:20
  service bots, maybe even pair programming tools
• 3:22
  where you need instant responses. So it becomes
• 3:25
  the go -to workhorse for anything needing speed
• 3:27
  and affordability. Yeah, that's a good way to
• 3:29
  put it. But the really fascinating part, the
• 3:32
  sort of aha. Moment isn't just that the small
• 3:35
  model is fast and cheap. It's how it plays with
• 3:37
  the big models You're talking about the orchestration
• 3:40
  idea using different models together exactly
• 3:42
  that yeah think of sonnet 4 .5 Anthropics current
• 3:46
  big powerful frontier model. It's the brilliant
• 3:50
  strategist. It can take a really complex goal
• 3:52
  like I don't know draft a detailed marketing
• 3:55
  plan for a new gadget. Something that requires
• 3:57
  deep understanding and planning. Right. But instead
• 4:00
  of Sonnet 4 .5 doing all the grunt work itself,
• 4:03
  which is slow and expensive, it can break that
• 4:05
  huge task down into maybe hundreds of smaller
• 4:08
  subtasks. Okay. And then it hands off those smaller
• 4:11
  specific tasks to a whole team, a whole fleet
• 4:14
  of these super fast, super cheap Haiku 4 .5 agents.
• 4:18
  They execute all those pieces in parallel. So
• 4:20
  the big brain plans and the army of fast workers
• 4:23
  gets it done. That's it. It's like a project
• 4:25
  manager directing a crew of highly efficient
• 4:27
  specialists. This makes really complex AI workflows
• 4:30
  suddenly much more feasible economically speaking
• 4:33
  for almost anyone. That makes a ton of sense.
• 4:35
  And we saw some real numbers on that speed difference
• 4:36
  too, right? A developer test. Yeah, someone tested
• 4:39
  a simple command. Just appending the word test
• 4:41
  to a note, Haiku 4 .5 did it in 18 seconds. OK.
• 4:45
  And Sonnet 4 .5? 47 seconds. And the same simple
• 4:48
  task. That's a huge difference in latency for
• 4:50
  everyday operations. Wow. almost three times
• 4:53
  faster, and there was a safety aspect too. Right.
• 4:55
  Anthropic made a point of releasing this under
• 4:57
  their AI safety level 2, or ASL 2. Basically,
• 5:01
  for listeners, that means it's undergone significant
• 5:04
  testing to minimize misuse for common predictable
• 5:07
  harms, things like generating hate speech or
• 5:11
  facilitating scams. So focused on practical safety
• 5:14
  for widespread use, not necessarily the more
• 5:16
  theoretical long -term risks that higher levels
• 5:19
  address. Exactly. It's about responsible deployment
• 5:21
  for these kinds of workhorse models. Faster,
• 5:24
  cheaper, and demonstrably safer according to
• 5:26
  their metrics. It's a powerful package. It really
• 5:29
  sets a new baseline for what we should expect
• 5:31
  from efficient AI. Definitely shifts the focus
• 5:34
  to smart resource distribution. How do you use
• 5:36
  the right tool or the right combination of tools
• 5:38
  for the job? That's a perfect transition, actually,
• 5:41
  because all this amazing software needs capable
• 5:45
  hardware to run on, especially if we want AI
• 5:47
  available instantly right on the device in our
• 5:49
  pocket. Precisely. Which brings us to story number
• 5:52
  two, Honor's new Magic 8 flagship phones. They
• 5:56
  seem to be betting big on physical AI integration.
• 6:00
  Yeah, two things really jumped out. The batteries
• 6:02
  and a dedicated AI button. Let's start with the
• 6:04
  batteries because these numbers are kind of wild.
• 6:07
  They really are. The versions launching in China,
• 6:09
  the Magic 8 Pro has a 7 ,200 milliamp hour battery.
• 6:14
  The standard Magic 8 has 7 ,000. Just for context,
• 6:16
  most flagship phones you buy today are maybe
• 6:19
  around 5 ,000 mAh. So this is a lot more. A lot
• 6:22
  more. It's enabled by a new silicon carbon battery
• 6:25
  tech. Basically, it lets them pack more energy
• 6:27
  into the battery compared to traditional lithium
• 6:29
  ion. The idea is to power more intensive on -device
• 6:33
  AI processing for longer periods. Okay, but I
• 6:36
  have to ask, is this genuine need? Are on -device
• 6:39
  AI tasks really demanding that much power already?
• 6:44
  Or is it maybe covering up for software that
• 6:46
  isn't perfectly optimized yet? That's the critical
• 6:49
  question, isn't it? It's probably a bit of both
• 6:51
  right now. But think about AI assistants running
• 6:54
  constantly, analyzing sensor data, maybe processing
• 6:58
  voice commands locally. That stuff does draw
• 7:01
  continuous power. So Honor seems to be saying,
• 7:04
  we want you to use these AI features without
• 7:06
  constantly worrying about your battery. It's
• 7:09
  building the foundation for seamless, always
• 7:11
  -on AI. Right, providing the raw power needed
• 7:14
  for the kind of efficient AI software we were
• 7:16
  just talking about. and powering it all is the
• 7:18
  new Snapdragon 8 Elite Gen 5 chip. Yep, Qualcomm's
• 7:22
  latest running Honor's Magic OS X, which is based
• 7:25
  on Android 16. It's designed specifically with
• 7:28
  these next -gen AI tasks in mind. And then there's
• 7:30
  that button, an extra physical button just for
• 7:32
  AI. Yeah, on the side. We've seen dedicated assistant
• 7:35
  buttons before, like the old Bixby button, and
• 7:38
  they often felt a bit gimmicky. Right, often
• 7:40
  got disabled or remapped. Exactly. But Honor
• 7:43
  seems to be positioning this differently. It
• 7:45
  shortcuts directly to their yo -yo assistant
• 7:48
  and something called yo -yo memories, which sounds
• 7:50
  like an AI -powered way to organize photos and
• 7:53
  documents. Okay, so it's trying to be a primary
• 7:55
  interface, not just a voice trigger. And interestingly,
• 7:58
  it also doubles as a camera shutter button, which
• 8:00
  suggests they expect you to be reaching for it
• 8:03
  often enough that it becomes natural, part of
• 8:05
  your muscle memory. It's a bet on making AI interaction
• 8:07
  physical and frequent. Integrating it deeper
• 8:10
  than just software. That's interesting quick
• 8:12
  note for international listeners. Those huge
• 8:14
  battery numbers are for the China models Right
• 8:16
  due to shipping regulations likely the global
• 8:19
  pro version will apparently have seven thousand
• 8:21
  one hundred milta still massive But in EU markets,
• 8:25
  it'll be six thousand two hundred seventy milita
• 8:27
  still very large but worth knowing good clarification
• 8:29
  So the trend is clear pack the phone with power
• 8:33
  and give AI its own dedicated access point the
• 8:36
  phone as a dedicated AI machine. That seems to
• 8:38
  be the direction. Okay, so we've covered efficiency
• 8:40
  in software and dedicated power in hardware.
• 8:44
  Let's bring it back to creativity with our third
• 8:46
  story. OpenAI is leveling up Sora, their video
• 8:50
  generation model. Yeah, Sora 2 is here, and it's
• 8:52
  tackling two big limitations of earlier AI video.
• 8:56
  Length and, well, making sense. Making sense,
• 8:59
  like basic physics. Pretty much. We'll get to
• 9:01
  that. First, the length. General users on the
• 9:04
  app or web can now generate clips up to 15 seconds
• 9:07
  long. Okay, that's up from what? Typically 10
• 9:09
  seconds or less for most tools. Right. A decent
• 9:12
  bump. But for pro users on the web version, the
• 9:15
  limit is now 25 seconds. 25 seconds. How much
• 9:18
  difference does that extra 10 seconds make creatively?
• 9:20
  Is it just more video? It's actually a pretty
• 9:23
  significant difference. But think about it, 15
• 9:25
  seconds is enough for a quick shot, a visual
• 9:27
  idea, maybe a social media snippet, but 25 seconds,
• 9:31
  that's enough time to actually craft a scene.
• 9:33
  You can establish a mood, have action unfold,
• 9:36
  maybe even include a line of dialogue with proper
• 9:38
  pacing, and then resolve the moment. It moves
• 9:40
  from cool effect territory towards actual storytelling.
• 9:43
  Okay, I see. It allows for a narrative beat,
• 9:46
  not just a visual. Exactly. And to support that,
• 9:49
  pro users also get a new storyboard feature in
• 9:51
  the web tool. So you can plan out your sequence,
• 9:54
  figure out the pacing before you hit generate,
• 9:56
  and spend your credits. That's about making the
• 9:59
  creative process more efficient too. Smart. Reducing
• 10:01
  trial and error. But you mentioned making sense
• 10:04
  earlier. What's changed there? This is maybe
• 10:07
  the biggest news. Sora 2 now generates synchronized
• 10:10
  audio dialogue that matches lip movements, background
• 10:13
  soundscapes, specific sound effects. It's catching
• 10:16
  up to competitors like Google's VO3 in adding
• 10:19
  that crucial audio layer. Finally, video without
• 10:22
  sound always felt incomplete. Definitely. But
• 10:25
  beyond audio, it's the visual consistency and
• 10:27
  logic. OpenAI is calling this their GPT 3 .5
• 10:31
  moment for video. Okay, comparing it to when
• 10:33
  GPT -3 .5 made large language models genuinely
• 10:36
  useful for complex tasks, that's a bold claim
• 10:39
  for video. What does it mean in practice? It
• 10:41
  means the AI seems to have a much better grasp
• 10:43
  of basic physics and cause and effect within
• 10:46
  the generated clip. Their example was a basketball
• 10:48
  shot. In older models, if you ask for a missed
• 10:51
  shot, the ball might hit the rim and then just
• 10:54
  sort of reappear in the player's hands, or bounce
• 10:56
  weirdly. The AI knew missed shot conceptually,
• 11:00
  but not how gravity and momentum work. Yeah,
• 11:02
  those little glitches totally break the illusion.
• 11:05
  Exactly. But with Sora 2, they claim much better
• 11:08
  physical accuracy. If the shot misses, the ball
• 11:11
  rebounds realistically off the rim or backboard.
• 11:14
  It understands inertia, gravity, object permanence
• 11:17
  within that 25 -second scene. That's huge for
• 11:20
  making the video feel believable. Wow, okay.
• 11:22
  That kind of internal logic is crucial. And wasn't
• 11:24
  there something about putting yourself in the
• 11:26
  video, too? Yeah, a quick mention of a new iOS
• 11:28
  social app feature, letting users insert themselves
• 11:31
  using cameos. So personalized AI video creation
• 11:34
  is becoming more accessible too. Wild. So longer,
• 11:38
  smarter, sounds better, and more physically consistent.
• 11:41
  That really does sound like a major step forward
• 11:43
  for AI video. It feels like the moment AI video
• 11:45
  starts becoming a truly viable creative tool,
• 11:47
  not just a novelty generator, the efficiency
• 11:49
  meets the realism. What a fascinating set of
• 11:52
  updates today. Okay, that wraps up our Morningstar
• 11:54
  report. Before we do our final recap, just want
• 11:57
  to remind you about our other sponsor. Yes, got
• 11:59
  to mention SCONI. If you're a golfer or no one,
• 12:03
  SCONI makes premium golf shoes known for comfort
• 12:05
  and performance. And just like with Old Glory,
• 12:08
  you can get 15 % off your entire purchase at
• 12:10
  SCONI using our promo code SNARFUL. That's S
• 12:13
  -N -A -R -F -U -L. Check them out at SCONI .com.
• 12:17
  Great shoes. Treat your feet on the course. Okay,
• 12:19
  so let's quickly tie today's threads together.
• 12:22
  The big theme that emerged was this push towards,
• 12:24
  I guess, smarter efficiency through specialization
• 12:27
  and dedication. Yeah, absolutely. First we saw
• 12:30
  an AI software with Antropic. The most powerful
• 12:33
  AI, Sonnet 4 .5, isn't doing everything itself.
• 12:37
  It's becoming the director, orchestrating fleets
• 12:39
  of cheaper, faster Haiku agents to do the specialized
• 12:42
  work. Decentralized intelligence almost. Right.
• 12:44
  Then second, we saw hardware stepping up with
• 12:47
  the Anamagic 8, dedicating physical components,
• 12:49
  that massive battery, that specific AI button
• 12:52
  to make sure AI runs efficiently and instantly
• 12:54
  right there on your device. Giving AI a physical
• 12:57
  home and the power it needs. And finally, with
• 13:00
  OpenAI Sora 2, we saw creative tools becoming
• 13:03
  more efficient and capable. Not just longer videos,
• 13:06
  but videos with internal logic, synchronized
• 13:09
  sound, better physics, moving towards being a
• 13:12
  real production tool by handling those complex
• 13:15
  details more reliably. It's like every layer
• 13:18
  of the stack is getting optimized for speed,
• 13:21
  cost, and importantly, useful results. The efficiency
• 13:25
  gains are enabling new capabilities across the
• 13:27
  board. So maybe the closing thought for you to
• 13:29
  ponder is this. If the cheapest, fastest AI models
• 13:32
  are already catching up to what was state -of
• 13:34
  -the -art just months ago, like Haiku 4 .5 did,
• 13:38
  how much further and faster will the next generation
• 13:40
  of those top -tier models push the boundaries?
• 13:43
  Are we getting to a point where the main bottleneck
• 13:45
  isn't the AI's capability but really our own
• 13:48
  imagination in figuring out how to best apply
• 13:50
  these incredibly efficient new tools? Yeah, it's
• 13:53
  a great question to chew on. The tools are evolving
• 13:55
  so fast, keeping up with the possibilities is
• 13:57
  the real challenge now. We're here to help at
• 13:59
  Snarful, so reach out on our website if you have
• 14:01
  any more questions. That's a wrap on today's
• 14:07
  top stories. Remember, tech doesn't have to be
• 14:09
  complicated and we're here to help connect the
• 14:11
  dots. If you've got questions or want to learn
• 14:14
  more, visit us at snarful .com. Thanks for listening
• 14:18
  and we'll catch you next time.