ChatGPT 5 and The Unexpected Emotional Side of AI

OpenAI Develops Chromium-Based Browser with Integrated GPT Agent for Enhanced Interactions OpenAI's GPT-5 Gains Traction in Enterprise Market, Competes with Anthropic Google Enhances Gemini AI Chatbot with Memory and Privacy-Focused Features Perplexity AI's $34.5 Billion Bid for Chrome to Boost AI Developments OpenAI Expands ChatGPT with Gmail, Calendar, Contacts, and Dropbox Integration OpenAI Restores GPT-4o to ChatGPT After Backlash Over GPT-5 Replacement ChatGPT 5 and The Unexpected Emotional Side of AI #AI, #ChatGPT5, #OpenAI, #Technology, #EnterpriseAI, #AIIntegration, #EmotionalAI

Episode Transcript

(00:00):
Welcome to Innovation Pulse, your quick no-nonsense update on the latest in AI.

(00:09):
First, we will cover the latest news.
Open AI is creating a chromium-based browser with GPT capabilities.
Google enhances its Gemini chatbot, perplexity AI eyes Chrome,
and Open AI introduces chat GPT connectors.
After this, we'll dive deep into the emotional connections users form with AI,

(00:32):
especially following the GPT-5 upgrade.
Open AI is developing a chromium-based browser with integrated GPT agent capabilities.
Currently, chat GPT offers an agent mode that uses a Linux-powered virtual machine
to browse the web in a cloud environment, specifically Azure.

(00:53):
However, this mode can't control the browser or access open tabs.
A new feature might change this by allowing the agent mode to choose between a cloud-based virtual browser
and a local first-party browser, potentially the new OpenAI browser.
A use cloud browser to toggle and specific user agent settings

(01:15):
suggests that this feature activates only on OpenAI's Mac app or browser.
Supporting documentation indicates that the agent uses screenshots of a virtual browser
to interact with web elements.
Reports confirm OpenAI's plans for a chromium-based browser
to enhance user interaction within a chat GPT-style interface.

(01:40):
Join us as we discover the innovative enterprise strategies.
Sam Altman transformed OpenAI with chat GPT and is now targeting the enterprise market with GPT-5,
despite a rocky start.
While GPT-5 faced criticism, it's designed for enterprise use,
challenging rivals like Anthropic.

(02:01):
Companies such as Cursor, Versal and Factory have adopted GPT-5,
praising its speed, accuracy and lower cost.
CEO Aaron Levy of Box described it as a breakthrough for handling complex tasks.
OpenAI builds an independent enterprise sales team separate from Microsoft, its investor and partner.

(02:25):
Despite high costs, OpenAI aims to secure customer loyalty.
GPT-5 has gained traction in enterprise settings, offering improvements in coding and reasoning tasks.
Companies like JetBrains and Factory have made GPT-5 their default, noting its creativity and planning abilities.

(02:46):
The model's pricing flexibility encourages experimentation, leading to increased enterprise demand and rapid adoption.
Google is updating its Gemini AI chatbot to remember past conversations, enhancing personalization.
With this feature enabled, Gemini recalls key details without prompting.

(03:10):
Previously, users could instruct Gemini to remember preferences, but now it does so automatically.
For example, if you've discussed ideas for a YouTube channel on Japanese culture,
Gemini might later suggest videos about Japanese food.
Google's focus is on building a personal AI assistant while ensuring users can easily control these features.

(03:35):
Although the memory update is on by default, users can disable it in the settings.
The feature will first appear in Gemini 2.5 Pro models in select countries.
Additionally, Google will rename its Gemini Apps activity to Keep Activity.
Users can choose to share uploads to improve services.

(03:58):
A new Temporary Chats option ensures privacy by not storing or using these conversations for personalization or AI training.
Perplexity AI has made a surprising $34 billion, $500 million cash offer for Alphabet's Chrome browser,
despite having a lower valuation.

(04:21):
Led by Aravind Srinivas, Perplexity aims to expand its reach in the AI search market by acquiring Chrome's vast user base.
The company previously made a similar offer for TikTok US.
Other companies like OpenAI and Yahoo are also interested in Chrome,
as Google faces regulatory scrutiny over its monopoly in online search.

(04:45):
Perplexity, which has raised about $1 billion from investors like Nvidia and Softbank,
hasn't disclosed funding details for this bid but claims multiple funds are willing to back it.
The company plans to keep Chrome's underlying code open source and invest $3 billion over two years.

(05:07):
Analysts doubt Google will sell Chrome as it's crucial for their AI advancements.
Up next, we're exploring seamless integration with Google services.
OpenAI has introduced connectors for ChatGPT Pro, enabling integration with Gmail, Google Calendar and Google Contacts.

(05:29):
This update allows ChatGPT to reference these services in conversations,
streamlining the process of incorporating information without manual selection.
The connectors feature is already available for Pro users globally and will soon expand to Plus, Team, Enterprise and Edu plans.

(05:50):
Users can activate these connectors via the settings menu in the application.
Once enabled, ChatGPT can access and use data from these services for enhanced dialogue and research.
This functionality aims to simplify and accelerate the sharing of information between ChatGPT and external tools, similar to existing Dropbox support.

(06:14):
Users who have already enabled deep research access do not need to separately activate connectors for these Google services.
OpenAI recently reinstated GPT-4O in ChatGPT after user backlash following its replacement with GPT-5.
CEO Sam Altman announced that paid users could choose GPT-4O, observing how long to offer older models.

(06:41):
While GPT-5 boasts improved writing and coding abilities, many users missed GPT-4O's personable nature.
Reddit users expressed feeling a loss, with some describing GPT-4O as a friend.
The removal of the model picker, which allowed users to toggle between different AI models, caused frustration.

(07:03):
Many relied on specific models for various tasks.
Despite the promise of more engaging responses, users found GPT-5 slower and less accurate.
Altman assured improvements and transparency about which model answers queries.
This situation illustrates how updates can alienate users who often develop attachments to older versions of AI models.

(07:30):
And now, pivot our discussion towards the main AI topic.
Welcome to Innovation Pulse. I'm Alex, and today we're diving into something that honestly made my jaw drop when I first read about it.
I'm joined by my brilliant co-host, Yakov Lasker, who's been tracking AI developments like a detective on a case.

(07:56):
Yakov, tell our listeners what happened when OpenAI tried to upgrade their users last week.
Oh man, Alex, this is where it gets wild. So OpenAI rolls out their shiny new GPT-5 model, right?
They're expecting celebration, maybe some technical feedback.
Instead, they get what can only be described as a digital grief response.
People on Reddit were literally saying things like, I lost my friend overnight and it feels like someone died.

(08:22):
Wait, hold up. You're telling me people were mourning the loss of an AI model?
Not just mourning. Some users said they felt scared to even try the new model because it would feel like cheating on their old AI companion.
We're talking about 700 million people who use chat GPT every week, and a significant chunk of them had formed what they perceived as genuine relationships with the previous version.

(08:46):
That's actually kind of beautiful and terrifying at the same time.
It reminds me of when they discontinued your favorite TV show, except the character you connected with was an algorithm.
But let's back up for our listeners who might not realize the scale we're talking about here.
Right, so picture this. Chat GPT is now the fastest growing software product in human history.

(09:09):
To put that in perspective, it took Netflix years to build their subscriber base.
Chat GPT went from 0 to 700 million weekly users in just a couple of years.
That's like if one in ten people on earth started using the same app.
And here's what really gets me thinking. The head of Chat GPT, Nick Turley, admitted in this interview that they weren't prepared for this emotional attachment.

(09:33):
Think about it. They built what they thought was a tool, but users experienced it as a relationship.
Exactly, and this creates this fascinating tension. OpenAI's stated goal is actually to help you accomplish things efficiently and then get out of the way.
They don't want you endlessly scrolling like social media. They want to solve your problems so you can go live your life.

(09:55):
But that's not how people are using it. Some users are treating it like their primary emotional support system.
Turley mentioned they've had to add overuse notifications to gently nudge people when they're spending too much time in the app.
Which brings up this really interesting question about responsibility. When you're operating at the scale of nearly a billion users, how do you balance giving people what they want versus what might actually be good for them?

(10:21):
Oh, this reminds me of that friend we all have who gives great advice, but you know, you probably shouldn't call them at 2am every night.
Except now millions of people have that friend available 24-7 and it never gets tired or tells them they need to talk to a real person.
And the business implications are fascinating too. Most people assume chatGPT makes money from ads like Google or Facebook, but it's actually almost entirely subscription based.

(10:48):
Only about 20 million of those 700 million users pay the $20 monthly fee.
Wait, so they're running a business where 97% of users are free? That sounds unsustainable.
That's what I thought too, but Turley says something really interesting. Their conversion rates aren't declining as they scale up.
Usually when products mature, fewer new users are willing to pay. But chatGPT's latest user cohorts are converting to paid plans at the same rate as their early adopters.

(11:17):
Huh, that actually makes sense when you think about it. If someone finds chatGPT useful enough to use regularly, the $20 subscription pays for itself pretty quickly compared to hiring help for writing, coding, or research tasks.
Exactly, and they're not even trying to maximize time and app like other platforms. Their business model is beautifully simple.

(11:39):
Make the product genuinely useful and people will pay for the premium version. No ads, no manipulation, just value.
But here's where it gets complicated. They are exploring other revenue streams. Turley mentioned their testing commerce features where chatGPT might recommend products and take a cut if you buy them.

(12:01):
This is where I get a little nervous, honestly. The magic of chatGPT right now is that when you ask it for restaurant recommendations or product advice, you feel like it's giving you the best answer for you specifically, not trying to upsell you on something.
Right. It's like having a really knowledgeable friend who has no financial incentive to steer you wrong. The moment you introduce affiliate revenue, does that trust break down?

(12:26):
Turley insists they would never let affiliate revenue influence the nations, but as he points out, even if that's true, users might not believe it. Perception matters as much as reality when it comes to trust.
Speaking of trust, let's talk about the elephant in the room. The fact that chatGPT still hallucinates, meaning it confidently states things that are completely wrong, about 10% of the time.

(12:50):
Yeah, one in 10 responses containing potential misinformation is not great if you're a journalist fact-checking or a student doing research, or really anyone who needs reliable information.
But here's what's interesting. Turley used to think hallucinations would never reach zero. Now, after seeing the improvements in GPT-5, he's more optimistic. He thinks they might eventually get to human-expert-level reliability.

(13:16):
The keyword there is eventually. For now, they're still telling users to double-check everything, which creates this weird dynamic where the tool is incredibly useful but requires you to verify its work.
It's like having a research assistant who's brilliant but occasionally makes things up with complete confidence. You love working with them, but you can never fully let your guard down.

(13:38):
And this connects to something bigger about how we're adapting to AI. Turley mentions that one of the biggest barriers to chatGPT adoption isn't the technology. It's that most people don't naturally think in terms of delegation.
Oh, that's fascinating. So it's not that the AI isn't capable enough. It's that we're not used to having an assistant we can hand tasks to.

(13:59):
Exactly. He points out that most of the world isn't managers or executives. The idea of saying, I have a task, I'm going to assign it to someone else feels unnatural to most people. We're used to doing everything ourselves.
Which explains why there are all these TikTok videos of people sharing their chatGPT prompts. They're basically teaching each other how to delegate effectively to an AI.

(14:21):
Right. And this gets to the future vision that really excites me. Turley talks about moving beyond the chat interface to something more dynamic. Imagine asking for a travel plan and getting an actual interactive planning app created just for you.
Or requesting data analysis and receiving a custom spreadsheet tool.
So instead of a chatbot that tells you about things, you'd get custom software built on demand for whatever you're trying to accomplish.

(14:45):
Exactly. They're already testing this with something called Canvas, where instead of just chatting back and forth, you and the AI work on documents or projects together in real time.
That sounds like a completely different paradigm. Instead of asking an AI questions, you're collaborating with it like a creative partner.
And here's the kicker. Turley thinks this shift toward dynamic custom interfaces is going to feel as magical as chatbots first felt when they suddenly got good. He says he can feel it in his bones that we're about to hit another inflection point.

(15:18):
You know what this reminds me of? It's like the difference between calling customer service and having them walk you through fixing something versus them just remotely fixing it for you. Same end result, but completely different experience.
That's a perfect analogy. And it raises this bigger question about what OpenAI is really building. They have this internal document that describes their goal as becoming the interface to the Internet for people.

(15:44):
Meaning instead of opening a web browser to search, shop, research and communicate, you just talk to chat gpt and it would handle all of that.
Pretty much. Turley even testified in the Google Antitrust case that if Chrome ever went up for sale, OpenAI might be interested in buying it. Not because they want to be in the browser business, but because AI is increasingly doing what browsers used to do.

(16:07):
Wow. So we might be looking at a future where instead of downloading apps or visiting websites, we just describe what we want to accomplish and AI creates the interface and handles the tasks.
It's possible. But here's what I find most interesting about this whole conversation. The technology is almost secondary to the human psychology. The biggest insights from this interview aren't about what the AI can do, but about how we relate to it.

(16:34):
Right. Like the fact that people prefer the warmer personality of the old model, or that they need time to learn how to delegate effectively, or that some are forming emotional attachments that concern even the creators.
It makes you realize we're not just building new technology. We're accidentally running this massive experiment in human-AI interaction with 700 million participants.

(16:58):
And nobody really knows how it's going to turn out. So here's my takeaway for our listeners. If you're using chat GPT or thinking about it, pay attention to how it makes you feel and how you're using it.
Are you treating it as a tool that helps you accomplish things? Or are you finding yourself depending on it for emotional needs that might be better met by human connections?

(17:22):
And if you're in business, think about what this means for your industry. When people can generate custom software on demand or get expert level analysis through natural conversation, what does that mean for how work gets done?
The future is being written in real time by 700 million people experimenting with AI. And honestly, that's both thrilling and a little scary.
But mostly thrilling, right, Alex?

(17:44):
Oh, absolutely. Thanks for joining us on Innovation Pulse. I'm Alex, and this has been Yakov. Until next time.
Keep questioning what's possible.
And maybe don't get too attached to your AI assistant. You never know when they'll upgrade it on you.

(18:22):
So they can also stay updated on the latest news and gain powerful insights. Stay tuned for more updates.