Latest / Connecting the Dots / Connecting the Dots - 4-21-2025

Transcript

Connecting the Dots - 4-21-2025 · 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. Good morning, everyone. OK, so it's Monday,
• 0:18
  April 21st, 2025. Out here in Sacramento, it's
• 0:21
  looking really nice sunny, expecting a high around
• 0:24
  80 degrees. Nice day for it. Yeah, definitely.
• 0:27
  A quick market check to Dow Jones. That closed
• 0:29
  last Thursday, April 17th, at 39 ,142 .23. Okay.
• 0:35
  And the S &P 500 finished up at 5282 .7. Also,
• 0:40
  just a heads up nationally, looks like the Central
• 0:42
  U .S. is getting some pretty severe weather.
• 0:45
  Thunderstorms, lots of rain. Right. Thought something
• 0:47
  about that. Yeah. Well, welcome back, everyone,
• 0:50
  to Connecting the Dots. I'm one of your guides
• 0:51
  here from Snarfl Solutions Group. And I'm your
• 0:54
  other guide, also part of the Snarfl team. And
• 0:56
  here on Connecting the Dots, we try to take all
• 0:58
  that tech news flying around and figure out,
• 1:00
  well, how does it actually connect to your life?
• 1:02
  Exactly. What's the real impact? So today we've
• 1:04
  got three pretty interesting things on the docket.
• 1:07
  First up, some really fascinating stuff from
• 1:09
  OpenAI. Seems their AI is getting much better
• 1:12
  at understanding images, almost thinking with
• 1:14
  them. Yeah, that one's got some layers to it.
• 1:16
  Then we'll shift gears and look at India's online
• 1:19
  food delivery scene. Kind of surprising, but
• 1:22
  it seems to be facing some unexpected challenges.
• 1:26
  Bit of a slow down, maybe. And finally, something
• 1:28
  potentially very practical using AI in healthcare
• 1:32
  to, well, maybe make dealing with your doctor's
• 1:35
  office a bit less painful. Definitely something
• 1:37
  many people can relate to. Okay, so let's dive
• 1:39
  in. First topic, open AI and this new image thinking
• 1:44
  AI. What's the core idea here? Okay, so OpenAI,
• 1:47
  they've launched these two new models, 03 and
• 1:50
  05, and the big deal, the sort of aha moment,
• 1:53
  is they're saying these models can think with
• 1:55
  images. Think with images. What does that actually
• 1:58
  mean in practice, like beyond just recognizing
• 2:01
  a cat in a photo? Exactly. It's more than just
• 2:03
  object recognition. It means the AI can internally
• 2:07
  manipulate the image as part of its reasoning,
• 2:10
  cropping, zooming, maybe even rotating parts
• 2:13
  of it. Ah, OK. So it's like it can interact with
• 2:15
  the picture in its mind to figure something out.
• 2:17
  That's a good way to put it. Yeah. So this blend
• 2:19
  of visual processing and verbal reasoning lets
• 2:22
  it analyze things much more deeply. So if I upload,
• 2:25
  say, a photo of a handwritten math problem, or
• 2:29
  maybe a really blurry sign. Right, or a complicated
• 2:32
  chart, the idea is the AI can now kind of dissect
• 2:36
  it visually before giving you an answer, leading
• 2:39
  to richer, more accurate results. OK, that makes
• 2:41
  intuitive sense. Especially for things like charts,
• 2:44
  where it's not just about seeing lines, but understanding
• 2:46
  the relationships. How much better are we talking?
• 2:49
  Did they give specifics? They did. They're reporting...
• 2:52
  Pretty substantial improvements, outperforming
• 2:55
  older models in benchmarks for STEM questions,
• 2:58
  science, tech, engineering, math. Also chart
• 3:01
  reading, just basic perception, visual search.
• 3:04
  They specifically mentioned one benchmark on
• 3:07
  V, which is apparently quite challenging. V,
• 3:09
  never heard of it. Yeah, it tests complex visual
• 3:11
  reasoning. And these new models hit 95 .7 % accuracy.
• 3:16
  OpenAI is basically saying they've largely solved
• 3:18
  it. So a pretty big step. 95 .7%. That sounds
• 3:21
  impressive. But there's always a but, isn't there?
• 3:24
  What's the catch? Well, yeah, here's where it
• 3:26
  gets interesting, maybe a little worrying. OpenAI
• 3:29
  themselves reported that these new models hallucinate
• 3:31
  more often. Hallucinate? You mean they just make
• 3:34
  stuff up more frequently? Essentially, yes. Producing
• 3:36
  inaccurate or nonsensical outputs more often
• 3:39
  than the previous versions. And what's really
• 3:42
  notable is OpenAI admits they don't fully get
• 3:44
  why yet. Oh, that's less good. So it's smarter
• 3:49
  visually, but also more prone to errors, how
• 3:52
  do you balance that? That's the million dollar
• 3:54
  question, isn't it? Especially for anything critical.
• 3:56
  They also mention the model sometimes overthink
• 3:59
  thing. Overthink, like doing unnecessary image
• 4:02
  tweet. Exactly, doing manipulations that don't
• 4:04
  actually help get a better answer and potentially
• 4:06
  some reliability issues like getting different
• 4:09
  results if you ask the same visual question multiple
• 4:11
  times. Progress, but definitely still some kinks
• 4:15
  to iron out. Who can actually use these new models
• 4:18
  now? Okay, so availability. Right now it's ChatGPT
• 4:21
  Plus Pro and team users. They get both 03 and
• 4:25
  04 mini. These have replaced the older versions.
• 4:28
  Enterprise and education users should be getting
• 4:30
  access like... this week. And even free chat
• 4:33
  GPT users will get a bit of 4mini's visual smarts
• 4:37
  through a new feature they're calling Think.
• 4:39
  Think. Okay. Yeah. It really does make you start
• 4:42
  imagining, you know, how this level of visual
• 4:44
  AI could filter into everyday tools. Better image
• 4:47
  search, maybe? Or AI assistance that can actually
• 4:50
  look at a graph with you and explain it properly.
• 4:53
  Accessibility tools, too. Yeah, tons of possibilities
• 4:55
  once AI can genuinely see and interpret. visuals
• 4:58
  better. Absolutely. It really opens up a whole
• 5:00
  new landscape. Okay, let's switch gears now.
• 5:03
  Story number two takes us over to India looking
• 5:06
  at the online food delivery market. You mentioned
• 5:08
  Zomato and Swiggy dominate there. Yeah, massively.
• 5:11
  Between them, they control something like 90
• 5:14
  % of the market near Duopoly. But it seems they're
• 5:18
  hitting some turbulence. Turbulence, how? Yeah.
• 5:20
  Weren't they growing like crazy? They were. But
• 5:22
  the latest reports, looking at the last six months
• 5:25
  or so, show things are kind of stagnating. The
• 5:28
  total value of orders isn't really growing much.
• 5:31
  And people aren't ordering quite as often. Zomato
• 5:33
  apparently saw a big drop in profit, like 57
• 5:36
  % year on year. Wow. Yeah. And only tiny growth
• 5:39
  in that gross order value, even a slight dip
• 5:42
  in monthly users. Swiggy's losses have reportedly
• 5:45
  widened, too. It feels like a market that's maybe
• 5:47
  maturing faster than expected. OK, so why the
• 5:50
  slowdown? Any ideas floating around? Well, a
• 5:54
  few potential reasons. Could be a general consumption
• 5:56
  slowdown in urban India, people cutting back
• 5:59
  on non -essentials. Makes sense. Also maybe saturation.
• 6:03
  In the big metro areas, the top eight or ten
• 6:05
  cities, maybe most people who are going to use
• 6:08
  these services are already using them. Right,
• 6:11
  fewer new customers to onboard. And just getting
• 6:13
  existing customers to order more often seems
• 6:15
  to be hitting a plateau as well. And what about
• 6:17
  prices? I heard that might be a factor too. Oh,
• 6:19
  definitely. Reports suggest both Zomato and Swiggy
• 6:23
  have hiked up their fees, platform fees, commissions
• 6:26
  from restaurants. Leading to? Leading to consumers
• 6:29
  paying significantly more, like maybe 20 to 30
• 6:32
  percent costlier just between 2023 and 2024.
• 6:35
  That's a big jump. Yeah, that would definitely
• 6:37
  make you think twice, especially younger people
• 6:39
  maybe. Exactly. That kind of price hike can really
• 6:42
  dent demand. It also makes it harder for competitors.
• 6:44
  There's this government -backed thing, ONDC,
• 6:47
  Open Network for Digital Commerce. ONDC, trying
• 6:51
  to break the duopoly. Yeah, trying to offer cheaper
• 6:53
  alternatives. But even with incentives, it seems
• 6:56
  like they're really struggling to get much market
• 6:59
  share. It's tough going up against those established
• 7:01
  giants. So if regular delivery is slowing, what
• 7:05
  are Zomato and Swiggy doing? Are they just...
• 7:07
  Accepting it? Not quite. They seem to be shifting
• 7:10
  focus a bit. One big push is into quick commerce.
• 7:13
  You know, those super fast 10 -15 minute deliveries.
• 7:16
  Like groceries and snacks delivered instantly.
• 7:19
  Right. Swiggy has stuff like Bolt and snacks.
• 7:21
  Ahmad is experimenting too. The other area is
• 7:23
  their dining out business, helping people find
• 7:25
  restaurants, book tables, pay bills. That side
• 7:28
  seems to be growing quite well. Interesting.
• 7:30
  But that quick commerce model, isn't it really
• 7:33
  hard logistically and expensive? Very complex
• 7:37
  logistics, questionable profitability, at least
• 7:40
  initially. And the big question is, does it actually
• 7:43
  bring in new users or just shift orders from
• 7:46
  their main platform? Good point. And the dining
• 7:48
  out part while growing is a totally different
• 7:50
  business model, right? Completely different.
• 7:52
  So they're trying things, but it's not clear
• 7:54
  yet if these pivots will fully reignite that
• 7:57
  rapid growth. You know, even though this is specific
• 8:00
  to India, it really does feel like it highlights
• 8:02
  broader trends. Market maturity. price sensitivity.
• 8:06
  Absolutely. And the sheer difficulty for anyone
• 8:09
  new trying to break into a market dominated by
• 8:11
  a couple of huge players. These are things we
• 8:14
  could see playing out elsewhere, too. Yeah, definitely
• 8:16
  some lessons there. OK, third and final story
• 8:19
  for today. Let's move into health care. And specifically,
• 8:23
  tackling that often painful process of trying
• 8:26
  to schedule a doctor's appointment over the phone.
• 8:28
  Right. It's kind of amazing. We have all these
• 8:29
  online patient portals, but the data suggests
• 8:32
  something like 88 % of appointments are still
• 8:34
  booked via phone call. 88%, wow. And only, what,
• 8:38
  2 .4 % online? Tiny fraction online. And this
• 8:41
  reliance on phones, it's inefficient. leads to
• 8:44
  errors and contributes to this massive problem
• 8:46
  of no -shows. No -shows. How big a problem is
• 8:49
  that? Huge. Estimated cost to the U .S. healthcare
• 8:52
  system is around $150 billion a year. Billion,
• 8:56
  with a B, plus all the time wasted by staff dealing
• 8:59
  with calls. That's staggering. Okay, so how is
• 9:02
  AI potentially helping here? Well, there's this
• 9:04
  company, Assort Health. They've built an AI platform
• 9:08
  specifically designed to handle these patient
• 9:10
  phone calls. assort health. And their goal is
• 9:13
  just better scheduling. Pretty much. Improving
• 9:16
  access to care by cutting down hold times, making
• 9:19
  scheduling smoother. That's the mission. What
• 9:21
  makes their approach different? Is it just a
• 9:23
  smarter IVR system? It seems to be more than
• 9:25
  that. A really key piece is its direct integration
• 9:27
  with the big electronic health record systems,
• 9:30
  EHRs like Cerner and Epapis. Okay. Why is that
• 9:33
  integration so important? Because it means the
• 9:35
  AI isn't just taking a message. It can actually
• 9:38
  access and update patient info, check insurance,
• 9:42
  and schedule the appointment right there, in
• 9:44
  real time, in the official system. No need for
• 9:48
  a human to manually enter it later. That sounds
• 9:50
  way more efficient. Cuts down on double handling
• 9:53
  and potential errors. Exactly. And it does more
• 9:55
  than just schedule. Oh, like what? It can triage
• 9:58
  calls, figure out urgency, answer frequently
• 10:01
  asked questions tailored to the specific doctor's
• 10:04
  office and even the patient's record. It can
• 10:06
  guide patients to provide needed info up front.
• 10:09
  So collecting details before the call even needs
• 10:12
  a human. Right. And it's proactive, too. Sending
• 10:15
  out automated SMS or calls for appointment reminders,
• 10:18
  confirmations, follow ups. It even tries to recover
• 10:20
  appointments that got canceled or weren't scheduled
• 10:22
  properly in the first place. OK, this sounds
• 10:24
  genuinely useful. Do they have any data on whether
• 10:27
  it actually works? They do. Assort Health's customers
• 10:30
  are reporting some pretty impressive results.
• 10:32
  They claim a call resolution rate over 90%. 90
• 10:36
  % handled by the AI. Seems like it. And an average
• 10:40
  patient satisfaction score of 4 .6 out of 5.
• 10:43
  That's high for dealing with an automated system.
• 10:45
  It is. Plus, they're seeing about a 3 % net gain
• 10:48
  in new patients, likely because it's easier to
• 10:51
  get through. And maybe the biggest number, a
• 10:53
  reported 63 % reduction in operational costs
• 10:57
  tied to managing calls. 63 % cost reduction?
• 11:01
  Wow, that's got to get attention. Definitely
• 11:03
  seems like a strong value proposition, better
• 11:05
  for patients, much cheaper for providers. And
• 11:08
  are they expanding? What's next for them? Yeah,
• 11:10
  they just raised $22 million in Series A funding,
• 11:13
  planning to use that to grow their team, especially
• 11:15
  hiring more engineers to keep developing the
• 11:17
  platform. So thinking about this from, you know,
• 11:21
  our perspective as patients, what could this
• 11:23
  realistically mean for our experience? potentially
• 11:26
  much shorter waits on hold when you call the
• 11:28
  doctor, a less frustrating scheduling process,
• 11:30
  fewer missed appointments because of better proactive
• 11:32
  reminders. Just a generally smoother interaction,
• 11:35
  hopefully. It really shows how AI can chip away
• 11:38
  at these everyday frictions, these administrative
• 11:40
  headaches, in really important areas like healthcare.
• 11:43
  Absolutely. It's not always about, you know,
• 11:45
  moonshots. Sometimes it's about fixing the plumbing.
• 11:48
  Yeah, well put. Okay, so let's just quickly recap
• 11:50
  the ground we covered today. Sounds good. So
• 11:52
  first we looked at OpenAI's new image -savvy
• 11:55
  AI. Really impressive potential for understanding
• 11:59
  visuals, but also dealing with increased hallucinations
• 12:02
  and reliability questions right now. Right. Then
• 12:05
  we jumped to India's food delivery market, Zimato
• 12:08
  and Swiggy facing slower growth, rising prices
• 12:11
  impacting demand, and their pivot towards quick
• 12:14
  commerce and dining out. Highlighting those challenges
• 12:17
  of market maturity and competition. And finally,
• 12:21
  assort health, using AI to automate healthcare
• 12:23
  phone calls, integrating with EHRs, and showing
• 12:26
  some really significant results in efficiency
• 12:28
  and patient satisfaction. And pulling it all
• 12:31
  together, the common thread seems to be AI getting
• 12:33
  more sophisticated, doesn't it? Better at handling
• 12:35
  different kinds of information images, market
• 12:38
  data, even spoken conversations. Yeah, it's moving
• 12:41
  beyond. Just text or simple data points. It's
• 12:44
  starting to understand and interact with the
• 12:46
  world or at least representations of it in more
• 12:49
  complex ways Blurring those lines. It definitely
• 12:51
  makes you wonder as this continues how AI is
• 12:54
  ability to process You know visual information
• 12:56
  voice complex interactions how that's going to
• 13:00
  reshape other services We use every day what
• 13:03
  other parts of life are ripe for this kind of
• 13:06
  AI -driven transformation. That is a great question
• 13:09
  to ponder. Lots to think about there. And remember,
• 13:11
  if you do have more questions on any of this
• 13:13
  or anything else tech -related, we are here to
• 13:15
  help here at Snarful. So reach out on our website
• 13:17
  if you have any more questions. That's a wrap
• 13:22
  on today's top stories. Remember, tech doesn't
• 13:25
  have to be complicated and we're here to help
• 13:28
  connect the dots. If you've got questions or
• 13:30
  want to learn more, visit us at snarful .com.
• 13:34
  Thanks for listening and we'll catch you next
• 13:36
  time.