Latest / Connecting the Dots / Connecting the Dots - 6-27-2025

Transcript

Connecting the Dots - 6-27-2025 · 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, welcome, welcome to Connecting
• 0:17
  the Dots. I'm Alex, thrilled to be kicking things
• 0:19
  off today from our studio right here in sunny
• 0:22
  Sacramento, California. And I'm Morgan. Great
• 0:25
  to be here with you all. It is Friday, June 27th,
• 0:28
  2025. And yeah, it's a hot one here in Saxeale,
• 0:31
  97 degrees Fahrenheit, predicted today. Stay
• 0:34
  cool out there, folks. Elsewhere, looks like
• 0:37
  that heat wave's continuing over in the Ohio
• 0:39
  Valley and mid -Atlantic. Yeah, and some severe
• 0:41
  thunderstorm chances up in the northern plains
• 0:43
  and upper Midwest through the weekend. So keep
• 0:45
  an eye on that. Okay, let's take a quick peek
• 0:47
  at the markets closing out the week. The Dow
• 0:50
  Jones finished Friday at 43 ,596 .56. That's
• 0:55
  up about 91 points for the day. And pretty solid
• 0:58
  for the week, too. Up what? 1418 points. It's
• 1:00
  over 3 .3%. Yeah, good week. S &P 500 closed
• 1:03
  at 6 ,159 .32. Also up slightly today. And up
• 1:07
  about 190 points for the week. Just over 3 %
• 1:10
  gain there, too. And Bitcoin, let's see. Closed
• 1:13
  at 106 ,886 .20. down just a tiny bit today,
• 1:17
  like 95 points, but still up to the week, over
• 1:19
  1 .25%. So mostly positive signs closing out
• 1:22
  June. Mostly positive, yeah, unless you're a
• 1:24
  San Francisco Giants fan like me. Tough loss
• 1:28
  yesterday against the Marlins, five to 12. Ouch,
• 1:30
  yeah, I saw that score. Yeah, Rafael Devers just
• 1:33
  lit us up, three for four with a homer. Oh well,
• 1:37
  sometimes you just gotta shake it off, look ahead.
• 1:39
  Exactly, and in the tech world, there's always
• 1:41
  something new ahead. We've got our Morningstar
• 1:43
  report lined up, packed with today's biggest
• 1:45
  stories. All right, let's dive in. First up,
• 1:49
  Metta. They are making some serious power plays
• 1:51
  in AI. We've heard rumblings, right? Even OpenAI
• 1:54
  CEO Sam Altman mentioned their poaching of talent.
• 1:59
  Yeah, and it seems that's not just talk. Their
• 2:01
  latest hire is really significant. Trebit Bansal.
• 2:04
  He was a key researcher at OpenAI, helped develop
• 2:06
  their O1 reasoning model. O1? Okay, remind us
• 2:09
  quickly what that is. So O1 is all about getting
• 2:11
  AI to think and plan more like humans, using
• 2:14
  logic. It's a big step. And Bansal worked closely
• 2:17
  with Ilya Sitskever, OpenAI's co -founder, especially
• 2:19
  on reinforcement learning. His LinkedIn now says
• 2:22
  he joined Meta this month. Wow. Okay, so he's
• 2:24
  not just any researcher. And where's he landing
• 2:27
  within Meta? He's joining their new AI super
• 2:29
  intelligence team. And get this, it's being led
• 2:32
  by Alexander Wang, the CEO of Scale AI. Scale
• 2:36
  AI, didn't Meta just invest heavily in them?
• 2:39
  They did. A massive $15 billion investment, giving
• 2:42
  Meta a 49 % stake. So you can see the strategy,
• 2:46
  right? Securing talent, securing capabilities
• 2:49
  through investment. It's a major push. It's definitely
• 2:52
  connecting the dots there. What else is part
• 2:54
  of Meta's grand AI plan? Well, beyond talent,
• 2:58
  they're reportedly working on their own AI search
• 3:00
  engine, indexing the web for over a year, apparently.
• 3:03
  An AI search engine? Interesting. And they're
• 3:06
  developing their own AI chips, too. They're in
• 3:08
  the testing phase, aiming for self -reliance,
• 3:11
  cutting costs, and maybe getting off the Nvidia
• 3:12
  train eventually. That's a huge undertaking.
• 3:15
  Building your own silicon. It is. And they've
• 3:17
  even mentioned potentially sharing some tech
• 3:19
  with the US government. So the ambitions are
• 3:21
  broad. OK. But the really juicy part might be
• 3:24
  this new Lama 3 .1 model they just announced,
• 3:27
  the 405B version. Zuckerberg is calling it the
• 3:29
  path forward for open source AI. Right. He's
• 3:32
  really pushing this idea that open models are
• 3:34
  the way to go, powering their meta AI assistant,
• 3:37
  benefiting everyone. But is it really open source?
• 3:40
  I feel like there's always a catch with that
• 3:42
  term these days. That's exactly the debate. The
• 3:44
  open source initiative, the OSI, and analysts
• 3:47
  like Stephen O 'Grady are saying, well, hold
• 3:49
  on a minute. It's not truly open source in the
• 3:52
  traditional sense. Why not? What's the issue?
• 3:54
  The licensing. It specifically restricts big
• 3:57
  competitors, think Amazon, Google, Microsoft,
• 4:00
  from using it commercially. The OSI director
• 4:03
  even called it a liability for developers because
• 4:05
  it's opaque and meta ultimately controls it.
• 4:08
  So it's open for some, but not for their main
• 4:11
  rivals. Kind of an open -ish strategy, then.
• 4:14
  Exactly. It's strategically open. It benefits
• 4:16
  Meta because lots of smaller players can use
• 4:18
  it, build on it, innovate with it. But it keeps
• 4:21
  the biggest threats at bay. That's clever, if
• 4:24
  a bit misleading on the open source label. But
• 4:26
  despite that debate, these models are having
• 4:28
  real -world effects, right? Oh, absolutely. Take
• 4:31
  Dr. Cal Clark at Zaron Labs. He's using Meta's
• 4:34
  law mark to build something called Guardian AI.
• 4:36
  He describes it as a kind of spell -checker for
• 4:39
  radiologists. Wow, okay. How does that work?
• 4:43
  Well, you know, radiologists look at thousands
• 4:45
  of scans a year, maybe 10 ,000, 15 ,000, and
• 4:50
  there's an estimated error rate, maybe 4%, which
• 4:53
  means millions of patients might be affected
• 4:55
  by false negatives. That's a lot. It is. So Guardian
• 4:59
  AI uses Llama to help spot those potential misses,
• 5:02
  those false negatives, and flags them for a second
• 5:04
  look by a human radiologist. That's incredible.
• 5:07
  And that's enabled by having access to a powerful
• 5:09
  model like Llama. Precisely. It shows the power
• 5:12
  of making these models available, even if it's
• 5:14
  open -ish. It helps level the playing field for
• 5:17
  innovators like Dr. Clark. Plus, there's the
• 5:19
  benefit of processing data locally. Ah, right.
• 5:22
  Crucial for healthcare with privacy rules like
• 5:24
  IPI. You can't just send patient scans to the
• 5:26
  cloud necessarily. Exactly. Jan LeCun, Metta's
• 5:29
  chief AI scientist, really emphasizes this collaborative
• 5:32
  aspect. He talks about building on each other's
• 5:34
  ideas, getting more people involved to make progress
• 5:37
  faster. But it also helps Metta, doesn't it?
• 5:39
  When others build on Llama, it makes Llama better,
• 5:42
  strengthens their ecosystem. Totally. It's a
• 5:44
  feedback loop. Developers create tools, find
• 5:47
  efficiencies, integrate it. It helps Llama catch
• 5:49
  up to or even surpass other models. It's mutually
• 5:53
  beneficial, but, you know, Meta's still holding
• 5:55
  the reins to that license. It really makes you
• 5:57
  think about what open even means anymore in tech.
• 6:01
  Meta's playing the long game here. Talent, chips,
• 6:04
  strategic openness. It's fascinating. Okay, let's
• 6:07
  shift gears from AI models to how we actually
• 6:11
  use social media day -to -day. Our next story
• 6:13
  is about platforms expanding their reach even
• 6:15
  onto your TV screen, potentially. And it starts
• 6:17
  with something I gotta say, it was a long time
• 6:19
  coming, Instagram finally, an iPad app. Took
• 6:23
  them long enough, right? After years of users
• 6:25
  asking and Adam Masseri, Instagram's head, kind
• 6:28
  of dismissing it, saying there wasn't enough
• 6:29
  demand or resources. Yeah, yeah. But is this
• 6:32
  just about pleasing iPad users or is there more
• 6:35
  to it? Oh, there's definitely more. This seems
• 6:37
  to be part of a bigger strategy, especially with
• 6:40
  the potential TikTok ban looming in the US. Instagram
• 6:43
  wants to be ready to scoop up those users and
• 6:46
  creators. Ah, makes sense. What else are they
• 6:48
  doing on that front? They're launching a new
• 6:49
  video editing app called Edits. Sounds like a
• 6:53
  direct competitor to ByteDance's CapCut, which
• 6:56
  is huge with TikTok creators. Okay, so... providing
• 6:59
  the tools creators need. Right. And they've increased
• 7:02
  the length limit for reels to three minutes matching
• 7:05
  TikTok. Plus some interface tweaks to make it
• 7:08
  feel, well, more like TikTok. Classic meta playbook
• 7:11
  maybe. Emulate the successful competitor. You
• 7:13
  could say that. Yeah. But it's also about how
• 7:15
  social media is challenging the old guard, like
• 7:17
  Google. Oh, so. Well, a Google executive, Prabhakar
• 7:21
  Raghavan, admitted something pretty startling
• 7:23
  recently. He said younger users, like 18 to 24
• 7:26
  -year -olds, are ditching Google search and maps
• 7:29
  for discovery. Really? Where are they going instead?
• 7:32
  TikTok and Instagram. He said almost 40 % of
• 7:34
  young people, when they're looking for, say,
• 7:36
  a lunch spot, they go to TikTok or Instagram
• 7:39
  first, not Google. 40%. Wow. That has to be worrying
• 7:44
  for Google. That's their core business, search
• 7:47
  and discovery. and the ads that go with it. Exactly.
• 7:51
  It's a potential long -term threat. Younger generations
• 7:54
  prefer, as he put it, visually rich forms of
• 7:57
  information. He made the point that many younger
• 8:00
  users have never even seen a paper map. Yeah,
• 8:03
  true. Yet a lot of mapping apps are still designed
• 8:06
  like digitized paper maps. Maybe that visual
• 8:08
  language just doesn't click the same way. That's
• 8:11
  a really interesting point about generational
• 8:13
  shifts and information seeking. So what's Google
• 8:17
  doing to fight back? They can't just let that
• 8:19
  happen. No, they're definitely responding. They're
• 8:21
  adding more augmented reality features to maps,
• 8:24
  making it more visual, more immersive with 3D
• 8:26
  views. Trying to make maps more engaging than
• 8:28
  just a flat map. Right. And significantly, they're
• 8:31
  now indexing Instagram and TikTok videos directly
• 8:34
  in Google search results. Oh, OK. So if I search
• 8:37
  for something, I might get a TikTok video as
• 8:38
  a result. Yeah, and they're using AI to analyze
• 8:41
  the content of those videos. So if you search
• 8:44
  how to change a tire, Google might show you a
• 8:47
  video result and let you jump straight to the
• 8:49
  part where they, I don't know, loosen the lug
• 8:51
  nuts. That's actually really useful instead of
• 8:53
  scrubbing through a whole video. Exactly. It's
• 8:55
  trying to make video content as searchable and
• 8:58
  useful as text. So maybe this is the next phase
• 9:01
  of search, right? More visual, more AI -driven
• 9:04
  analysis of different content types. It certainly
• 9:06
  feels like search is evolving beyond just typing
• 9:09
  words into a box. It's adapting to how people,
• 9:12
  especially younger people, discover and consume
• 9:15
  information. Fascinating stuff. Definitely. Okay,
• 9:18
  for our last story today, let's talk about the
• 9:20
  hardware that powers so much of this AI boom.
• 9:24
  NVIDIA. They've been flying high, but it looks
• 9:26
  like they're hitting some turbulence with their
• 9:28
  next big thing. Ah, the Blackwell chips. I heard
• 9:31
  those were supposed to be incredible, like an
• 9:32
  iPhone moment for NVIDIA. Twice as fast as the
• 9:35
  current Hopper chips. What's the issue? Well,
• 9:37
  there seem to be a couple of significant problems
• 9:39
  causing delays. First, there were reports about
• 9:42
  design flaws in the... processor die that connects
• 9:46
  the two GPUs together on the main GB200 chip.
• 9:49
  Their manufacturer, TSMC, apparently flagged
• 9:52
  it. Okay, a design flaw. That sounds serious.
• 9:54
  Yeah. And more recently, another issue popped
• 9:56
  up. Overheating. Not necessarily the chips themselves,
• 10:01
  but the server racks designed to hold them. They're
• 10:03
  apparently getting too hot. requiring some late
• 10:06
  stage design changes. Overheating racks. So it's
• 10:09
  a system level problem. These things generate
• 10:11
  immense heat, I suppose. They do. And these aren't
• 10:13
  small issues. They're causing real delays. So
• 10:15
  what's the intact? I know all the big tech companies,
• 10:17
  Microsoft, Google, Metta, they've ordered billions
• 10:20
  of dollars worth of these Blackwell chips. Huge
• 10:23
  impact. It's creating a domino effect. Microsoft,
• 10:26
  for example, was planning to get like 55 ,000
• 10:28
  to 65 ,000 of these GV 200 GPUs for OpenAI by
• 10:32
  early 2025. That timeline looks shaky now. Wow,
• 10:36
  that directly impacts OpenAI's plans. Right.
• 10:39
  And Google apparently ordered over 400 ,000 of
• 10:41
  them. That's like a $10 billion order. Meta also
• 10:44
  put in a $10 billion order. Significant shipments
• 10:47
  for these big orders aren't expected now until
• 10:49
  the first quarter of 2025, maybe later. So everyone's
• 10:53
  AI timelines are getting pushed back because
• 10:56
  of these chip delays. It seems likely, yeah.
• 10:58
  It affects how quickly they can train their next
• 11:00
  -gen models, deploy new AI services. It's a major
• 11:02
  bottleneck. How's NVIDIA handling this publicly?
• 11:05
  Are they admitting to the problems? Their official
• 11:08
  line is more measured. They're saying production
• 11:09
  is on track to ramp and that these kinds of engineering
• 11:13
  challenges are normal and expected for cutting
• 11:16
  -edge tech. They did highlight that SoftBank
• 11:18
  is set to be the first customer using Blackwell
• 11:21
  to build the supercomputer in Japan early next
• 11:23
  year. Trying to put a positive spin on it maybe,
• 11:26
  but these delays must be creating an opening
• 11:28
  for Nvidia's competitors, right? They have like
• 11:31
  80 % of the market. Absolutely. This could be
• 11:33
  a real opportunity for companies like Intel with
• 11:35
  their gaudy AI chips, which are reportedly much
• 11:38
  cheaper. Cheaper is always attractive. And AMD,
• 11:42
  too. They're pushing their open source ROCM software
• 11:45
  platform as an alternative to NVIDIA's dominant
• 11:48
  proprietary CUDA platform. If customers can't
• 11:52
  get Blackwall when they need it, they might start
• 11:54
  looking seriously at Intel or AMD. So NVIDIA's
• 11:57
  dominance might actually be tested here because
• 11:59
  of these production hiccups. It's possible. The
• 12:02
  demand is astronomical. But if you can't deliver,
• 12:05
  customers will explore alternatives. It just
• 12:08
  shows how intense this whole AI hardware race
• 12:10
  is. The stakes are incredibly high. Yeah, one
• 12:13
  slip up in manufacturing these complex systems
• 12:15
  can send ripples across the entire industry.
• 12:17
  It means the AI future everyone's racing towards
• 12:19
  might arrive on a slightly different schedule
• 12:21
  and maybe with a more diverse set of players
• 12:23
  involved. Exactly. It's never a smooth path.
• 12:26
  when you're dealing with this level of complexity
• 12:28
  and demand. OK, quite a journey through tech
• 12:30
  today. We looked at Meta's big AI push -poaching
• 12:33
  talent, investing heavily, and that whole open
• 12:36
  -ish source strategy with Llama 3 .1 and what
• 12:39
  that really means. Then we shifted to social
• 12:42
  media's evolution, with Instagram finally getting
• 12:44
  an iPad app. But more importantly, how platforms
• 12:48
  like Instagram and TikTok are challenging Google's
• 12:50
  dominance in search and discovery, especially
• 12:53
  with younger users. And finally, the speed bumps
• 12:55
  for NVIDIA. Those delays with the super -hyped
• 12:58
  Blackwell AI chips due to design and overheating
• 13:01
  issues, and how that impacts the timelines for
• 13:04
  giants like Microsoft, Google, and Meta, potentially
• 13:07
  opening the door for competitors. These stories
• 13:09
  really show how interconnected everything is.
• 13:12
  Software ambitions depend on hardware realities,
• 13:15
  user behavior shapes, platform strategies, and
• 13:17
  the definition of open can be a competitive tool.
• 13:20
  It all connects back to how we work, how we find
• 13:23
  information, how businesses operate and innovate.
• 13:25
  These aren't just abstract tech trends, they
• 13:27
  shape the tools and services you use every single
• 13:30
  day. We're here to help at Snarful, so reach
• 13:32
  out on our website if you have any more questions.
• 13:50
  Thanks for listening, and we'll catch you next
• 13:52
  time.