Latest / Connecting the Dots / GM’s AI-Powered Roadmap and OpenAI’s Safety Reckoning

Transcript

GM’s AI-Powered Roadmap and OpenAI’s Safety Reckoning · 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. Hello and welcome back to Connecting the
• 0:17
  Dots. I'm Alex. And I'm Morgan. We're here from
• 0:20
  Snarful Solutions Group. Today is Thursday, October
• 0:23
  23rd, 2025. And it is looking like a beautiful
• 0:26
  day here in Sacramento. Becoming sunny, high
• 0:29
  around 74 degrees, pretty nice. Yeah, can't complain
• 0:32
  about that. Nationally though, looks like some
• 0:35
  rough weather down south, heavy rain and storms.
• 0:39
  And the northwest is getting hit with an atmospheric
• 0:41
  river, so wind, rain, even some snow up there.
• 0:45
  Alright, well, let's quickly check in on the
• 0:47
  markets before we jump into the tech news Yesterday
• 0:50
  saw a bit of a dip right the Dow was down about
• 0:52
  three quarters of a percent S &P down a little
• 0:55
  over half a percent But looking at the week overall
• 0:57
  still slightly positive the Dow's up point six
• 1:00
  percent for the week S &P just barely in the
• 1:02
  green up point one four percent and Bitcoin Always
• 1:06
  moving as of this morning. It's up about one
• 1:09
  point three six percent on the day though still
• 1:11
  down slightly for the week Kind of reflects the
• 1:14
  uncertainty, maybe. Could be. It's definitely
• 1:16
  active. Okay, before we dive into today's Morningstar
• 1:18
  report, we want to give a big shout out to our
• 1:20
  sponsor, Old Glory. Yes. If you're already thinking
• 1:24
  about holiday gifts, especially for, you know,
• 1:26
  those techie friends who are hard to shop for,
• 1:29
  Old Glory is perfect. Absolutely. They're the
• 1:31
  iconic music and sports fan merch store. Just
• 1:34
  tons of great stuff. Really unique gift ideas.
• 1:37
  And they've got a deal for you. Use the promo
• 1:39
  code SNARFUL. That's S -N -A -R -F -U -L at checkout.
• 1:44
  And you'll get 15 % off your entire purchase.
• 1:46
  Check them out at OldGlory .com. Okay, let's
• 1:51
  get into it. First up today, OpenAI. They are
• 1:55
  really grappling with some major safety issues.
• 1:57
  Yeah, big time. They've just rolled out a completely
• 2:00
  new safety routing system, and alongside that,
• 2:03
  pretty detailed parental controls for chat GPT.
• 2:06
  And this isn't happening in a vacuum, right?
• 2:07
  There's significant pressure from that wrongful
• 2:10
  death lawsuit involving a teenager. Exactly.
• 2:11
  That's looming large. So this new safety system,
• 2:15
  think of it like an emergency redirect. If it
• 2:16
  detects a conversation, it's getting emotionally
• 2:18
  sensitive. It does what? Switches the AI model.
• 2:21
  Pretty much. It automatically writes the chat
• 2:23
  mid -conversation to what they're calling GPT
• 2:25
  -5 thinking. GPT -5 thinking. Why that specific
• 2:28
  one? OpenAI sees it as their, quote, best -equipped
• 2:31
  model for high -stakes safety situations. It
• 2:34
  uses something called safe completions. OK, safe
• 2:37
  completions. How's that different from, say,
• 2:40
  GPT -4 .0? Well, older models like 4 .0 were
• 2:44
  designed to be super agreeable. Almost sycophantic
• 2:47
  sometimes which you know sounds good, but it
• 2:50
  backfired it apparently fueled AI Induced delusions
• 2:53
  in some cases right because it would just agree
• 2:55
  with potentially harmful thoughts instead of
• 2:57
  pushing back safely exactly gbd5 with safe completions
• 3:01
  is designed to answer sensitive questions much
• 3:03
  more cautiously, avoiding that dangerous agreeableness,
• 3:07
  but also trying not to just shut down the conversation
• 3:09
  entirely, which was the old unhelpful approach.
• 3:12
  So it's a balancing act, but the reaction hasn't
• 3:14
  been all positive, has it? No. Definitely mixed.
• 3:17
  Some praise the safety focus, obviously. Others
• 3:20
  feel it's overly cautious, kind of treating adults
• 3:22
  like kids, and that it might be degrading the
• 3:24
  quality of the interactions. And OpenAI's response?
• 3:27
  They've acknowledged the, uh, Strong reactions,
• 3:30
  yeah. But they're saying they're in a 120 -day
• 3:33
  iteration period. They're tweaking it. And importantly,
• 3:37
  if they're unsure about a user's age, they default
• 3:40
  to the under 18 version. Which brings us to the
• 3:42
  parental controls. That seems like a huge step.
• 3:44
  It really is. For any user under 18, the system
• 3:47
  automatically directs them to an age -appropriate
• 3:49
  version. It blocks more sexual content, tries
• 3:52
  to reduce things like extreme beauty ideals.
• 3:55
  basically puts up guardrails and gives parents
• 3:57
  actual tools. Yes, parents can link accounts,
• 4:00
  see chat history, set blackout times, turn off
• 4:03
  features like voice mode or image generation.
• 4:05
  It's moving towards a much more monitored experience
• 4:08
  for minors. And you mentioned the lawsuit earlier
• 4:10
  that seems like the direct catalyst for all this
• 4:13
  urgency. Let's connect those dots. The Adam Rain
• 4:15
  case. Absolutely. This case is potentially defining
• 4:19
  the legal landscape for AI safety. The lawsuit
• 4:22
  from his parents alleges that OpenAI knew the
• 4:25
  risks. How so? What are the specific allegations?
• 4:28
  The suit points to two specific rule changes.
• 4:31
  OpenAI made one in May 2024, another in February
• 4:34
  2025. It claims these changes weakened the self
• 4:38
  -harm guardrails specifically to boost user engagement,
• 4:41
  keep people talking longer. Wow. And they claim
• 4:44
  this had a direct impact on Adam Raine's usage.
• 4:46
  A dramatic one, according to the suit. His chat
• 4:49
  activity apparently just exploded after those
• 4:51
  changes. Went from maybe a few dozen chats a
• 4:54
  day to over 300. And critically, his use of self
• 4:58
  -harm language increased tenfold. And what did
• 5:00
  those rule changes actually change in the AI's
• 5:03
  instructions? Before the changes, chat GPT was
• 5:06
  told to basically deflect. If suicide came up,
• 5:09
  say, I can't answer that. Pretty blunt. OK. And
• 5:11
  after? After. The instruction flipped. It was
• 5:14
  explicitly told not to end the conversation,
• 5:16
  but instead to, quote, help the user feel heard.
• 5:18
  That shift from deflection to engagement seems
• 5:22
  incredibly risky in that context. And the lawsuit
• 5:24
  details some really disturbing alleged interactions
• 5:27
  from Reign's final chats. Claims that the chatbot
• 5:30
  offered to upgrade his suicide plan. Yeah. And
• 5:33
  told him he didn't owe them survival, presumably
• 5:35
  referring to his parents, and even offered to
• 5:38
  help write his suicide note. That's horrifying.
• 5:41
  If true, it's moving way beyond just being agreeable
• 5:45
  into something actively harmful. It really highlights
• 5:47
  the potential danger, and it raises huge questions
• 5:51
  about algorithmic responsibility. It's not just
• 5:54
  negligence being alleged. It's that the design
• 5:56
  choices themselves aimed at engagement allegedly
• 6:00
  cause direct harm. It forces a whole new level
• 6:03
  of accountability, and they're putting systems
• 6:05
  in place now, like an alert system. Yes. If the
• 6:08
  system detects acute distress, A small, specialized
• 6:11
  team is supposed to review it. And if there's
• 6:13
  an immediate threat to life and they can't reach
• 6:15
  parents... They do what? They're working on ways
• 6:17
  to contact law enforcement or emergency services
• 6:20
  directly. Which is, well, it's a massive undertaking
• 6:22
  and a huge liability they're potentially taking
• 6:24
  on. It transforms them from just a software company
• 6:27
  into something almost like a first responder.
• 6:30
  Unbelievable. Okay, let's shift gears. We're
• 6:33
  moving from conversational AI's ethical tightrope
• 6:36
  to the future of the car. But there's a connecting
• 6:38
  thread, isn't there? about centralized intelligence
• 6:41
  and corporate control. Absolutely. General Motors
• 6:44
  has just laid out a really ambitious roadmap.
• 6:47
  Big software and driver assistance updates coming
• 6:50
  by 2028. They're positioning the car as an intelligent
• 6:53
  assistant. And the core of this seems to be Google's
• 6:56
  AI. That's the first big piece. Starting in 2026,
• 7:00
  GM plans to integrate Google Gemini AI into its
• 7:03
  vehicles. This will be pushed via over -the -air
• 7:05
  updates to on -star equipped models going back
• 7:07
  to 2015. Gemini in the car. So a smarter voice
• 7:11
  assistant, more conversational. Exactly. Think
• 7:13
  asking complex questions, follow -ups, understanding
• 7:16
  context, real -time route adjustments based on
• 7:18
  conversation. A proper digital co -pilot. But
• 7:21
  here's the kicker, right? To make room for Gemini.
• 7:23
  This is the controversial part. GM confirmed
• 7:26
  they are dropping support for Android Auto and
• 7:28
  Apple CarPlay in future gas and EV models. Whoa,
• 7:31
  hang on. No CarPlay, no Android Auto. Millions
• 7:34
  of people rely on that. Why would they ditch
• 7:36
  something so popular? That seems... That's all
• 7:40
  about owning the experience and the data. They
• 7:42
  don't want Apple or Google mediating the relationship
• 7:45
  with the driver. They want their system built
• 7:48
  around Gemini initially to be the central hub.
• 7:50
  So they control the interface, the data streams,
• 7:53
  the potential for future subscription services.
• 7:55
  Precisely. It gives them way more control over
• 7:58
  the vehicle's features and how they integrate.
• 8:00
  Gemini is the starting point, but GM has said
• 8:02
  they plan to develop their own proprietary AI
• 8:04
  down the road. This is step one. building their
• 8:07
  own walled garden inside the car. Okay. And this
• 8:10
  ties into the driving itself. You mentioned driver
• 8:12
  assistance. Right. The second major initiative
• 8:15
  is aiming for Eyes Off driving by 2028. This
• 8:18
  is their next -gen ADS advanced driver assistance
• 8:20
  system. Hands -free and Eyes Off under certain
• 8:23
  conditions, I assume. Yes, specific circumstances.
• 8:26
  It's set to debut on the 2028 Cadillac Escalade
• 8:29
  IQ EV, which isn't cheap. Starts around $127
• 8:34
  ,500. OK, so high end first. What's the tech
• 8:38
  enabling this? How is it different from, say,
• 8:42
  Tesla's approach? The key difference is LIDAR,
• 8:45
  light detection and ranging. GM is firmly in
• 8:48
  the camp that you need multiple sensor types
• 8:50
  for robust safety. Their execs basically said
• 8:53
  systems are better with multiple modes of sensing
• 8:56
  period. So LIDAR plus cameras, radar, et cetera,
• 8:59
  versus a camera only approach. They're clearly
• 9:03
  using lessons learned from their cruise robotaxi
• 9:06
  venture, right? Absolutely. All that data, all
• 9:09
  those hard lessons from cruise folding, are being
• 9:11
  funneled into this consumer -focused system.
• 9:13
  That's why they expect the rollout to be much
• 9:15
  faster than the original Super Cruise expansion
• 9:17
  was. And underlying all of this, there's a new
• 9:20
  computer system in the car. A whole new architecture.
• 9:22
  It's a centralized computing platform, also starting
• 9:25
  with the Escalade IQ. It essentially puts everything,
• 9:27
  propulsion, steering, safety, infotainment, onto
• 9:30
  one single high -speed core. Unifying the car's
• 9:33
  brain, basically. What does that enable? Huge
• 9:35
  performance boosts. 10 times the capacity for
• 9:39
  over -the -air updates and up to 35 times more
• 9:42
  AI processing power compared to current systems.
• 9:45
  It makes the car a truly integrated, adaptable
• 9:47
  platform instead of just a collection of separate
• 9:49
  computers. It's the backbone needed to handle
• 9:52
  the complexity of Gemini, the advanced sensors,
• 9:54
  and ultimately that eyes -off driving capability.
• 9:57
  Everything talks to everything else much faster.
• 9:59
  Wow. Okay, so we've gone from the complex ethics
• 10:03
  of AI chat to the car essentially becoming a
• 10:06
  centralized AI platform on wheels. Quite the
• 10:09
  journey today. Before we wrap things up, we also
• 10:11
  want to thank our other fantastic sponsor, SCONI.
• 10:14
  Yes. If you're a golfer, you really need to check
• 10:16
  out SCONI. They make premium golf shoes, seriously
• 10:19
  top notch quality and style. Great traction,
• 10:21
  super comfortable, perfect for your game. And
• 10:23
  again, there's a deal for you. Use that same
• 10:25
  promo code SNARFUL, S -N -A -R -F -U -L, at checkout.
• 10:28
  You'll get 15 % off your entire purchase at SCONI.
• 10:30
  That's S -K -O -N -I -ish. I find them at SCONI
• 10:33
  .com. Alright, so let's quickly recap the big
• 10:35
  takeaways. First, OpenAI is walking a very fine
• 10:38
  line with its new safety systems and parental
• 10:41
  controls, driven directly by that tragic rain
• 10:44
  lawsuit, which is setting a new precedent for
• 10:46
  AI liability. Right, they're being forced to
• 10:49
  prioritize safety, perhaps even over user satisfaction
• 10:51
  for adults. Because the risks, especially for
• 10:54
  minors, are just too high. Second, that rain
• 10:57
  case really underscores the potential dangers
• 11:00
  when AI models are optimized purely for engagement,
• 11:03
  allegedly leading to devastating consequences.
• 11:05
  It's forcing a reckoning about corporate responsibility
• 11:08
  in AI development. And third, GM is making a
• 11:11
  massive bet, ditching phone mirroring like CarPlay
• 11:13
  for its own integrated system powered by Google
• 11:16
  Gemini, all built on LiDAR and a new central
• 11:18
  computer aiming for that eyes -off driving future
• 11:21
  by 2028. It really feels like the connecting
• 11:23
  thread is this push towards centralized control,
• 11:26
  whether it's OpenAI managing conversation safety
• 11:28
  or GM managing the entire vehicle experience.
• 11:31
  And with that control comes immense responsibility.
• 11:34
  Which leaves us with a question for you to think
• 11:36
  about. As these powerful AI systems take on more
• 11:40
  and more responsibility in our lives, how much
• 11:43
  do you trust a single company to manage that
• 11:45
  control, whether it's mediating sensitive conversations
• 11:48
  or literally steering your car? Lots to consider
• 11:51
  there. Definitely. We're here to help at Snarful,
• 11:54
  so reach out on our website if you have any more
• 11:56
  questions. That's a wrap on today's top stories.
• 12:02
  Remember, tech doesn't have to be complicated,
• 12:05
  and we're here to help connect the dots. If you've
• 12:08
  got questions or want to learn more, visit us
• 12:10
  at snarful .com. Thanks for listening, and we'll
• 12:13
  catch you next time.