Build by AI

Chat is Dead: OpenAI's Radical ChatGPT Overhaul I 8th June

Share episode

Use Left/Right to seek, Home/End to jump to start or end. Hold shift to jump forward or backward.

0:00 | 32:36
OpenAI is reportedly planning to kill the chat interface that made ChatGPT famous, marking the biggest overhaul since launch. Meanwhile, SpaceX is quietly becoming an AI infrastructure powerhouse with massive deals from Google and Anthropic ahead of its IPO. We dive into what these seismic shifts mean for the future of AI interaction and who's really winning the infrastructure wars. Plus: why South Korea just became the hottest battleground for AI sovereignty.
SPEAKER_00

The company that made chatbots mainstream is reportedly about to kill the chat interface entirely. Like I'm staring at headlines that literally say chat is dead from OpenAI's biggest chat GPT overhaul since launch.

SPEAKER_02

Dude, the irony is incredible. They spent two years training the entire world to talk to AI through chat, got everyone comfortable with that back and forth conversation model, and now they're basically saying, actually, never mind, we're doing something completely different.

SPEAKER_00

It's like if Facebook suddenly announced they were getting rid of the newsfeed. You don't just casually overhaul the core interaction model that defines your entire product.

SPEAKER_02

Right, and this isn't some minor UI update we're talking about. If they're really moving away from chat, that suggests they've figured out something fundamentally better about how humans should interact with AI. The question is, what the hell is it?

SPEAKER_00

And here's what's really wild. This is all happening while SpaceX is quietly becoming an AI infrastructure giant. We've got reports of them securing deals with Google and Anthropic worth billions. Like since when is the rocket company in the AI business?

SPEAKER_02

That's what I'm saying. The whole AI landscape is getting turned upside down. We're not just talking about better chatbots anymore. We're talking about satellites processing AI in space, countries building their own AI infrastructure to avoid being dependent on US or Chinese systems.

SPEAKER_00

If even half of what we're seeing today is accurate, we might be looking at the biggest shift in how AI works since ChatGPT launched. And frankly, I'm not sure anyone is prepared for how fast this is moving. You're listening to Build by AI. I'm Alex Shannon, and we're diving into what might be the biggest AI interface revolution since ChatGPT launched.

SPEAKER_02

And I'm Sam Hiton. Today we're not just talking about OpenAI's radical ChatGPT overhaul, but also how SpaceX is quietly becoming an AI infrastructure giant, and why every major tech company is suddenly obsessed with South Korea.

SPEAKER_00

It's June 8th, 2026, and honestly, the pace of change in our infrastructure is getting wild.

SPEAKER_02

Buckle up, because if these stories are accurate, we might be looking at the end of the AI interface era as we know it.

SPEAKER_00

to describe what's happening.

SPEAKER_02

Yeah, and that's not clickbait hyperbole. The reporting suggests they're fundamentally redesigning the chat interface itself. Think about how radical that is. The entire chat GPT experience is built around that conversational back and forth.

SPEAKER_00

But Sam, help me understand this. Why would you kill the thing that made you successful? Chat interfaces work. People get them. They feel natural.

SPEAKER_02

Um, okay, but but here's what people are missing. Chat interfaces are actually pretty limiting when you think about it. You know, you're forcing complex AI capabilities into this linear, um, turn-taking model that's designed for human conversation. What if the AI could show you information, manipulate data in real time, or work alongside you in a completely different way?

SPEAKER_00

So you're thinking more like a collaborative workspace than a chatbot. Like instead of asking Chat GPT to write code and then copying it out, maybe it's working directly in your development environment.

SPEAKER_02

Exactly. Or imagine instead of describing an image you want, the AI is manipulating visual elements in real time as you guide it. The chat model forces everything through this narrow text pipeline when AI capabilities are way broader than that.

SPEAKER_00

Okay, that makes sense from a capability standpoint, but I'm worried about the user experience angle. Chat interfaces have this huge advantage. They're immediately familiar. Everyone knows how to have a conversation.

SPEAKER_02

So that's a fair point. But remember, before the iPhone, everyone knew how to use physical keyboards too. Sometimes you have to break familiarity to unlock the next level of capability. If OpenAI has cracked a more intuitive way to interact with AI, the learning curve might be worth it.

SPEAKER_00

The other thing I'm thinking about is competitive pressure. Google, Anthropic, everyone else is still doing chat interfaces. If OpenAI moves away from chat and it doesn't work, they're handing their competitors a massive advantage.

SPEAKER_02

But if it does work, they're not just ahead. They're playing a completely different game. This could be an iPhone moment where suddenly everyone else's chat interfaces look ancient. It's a huge risk, but the potential payoff is enormous.

SPEAKER_00

What should people who are building on chat GPT APIs be thinking about right now? Like if you've built your whole product around chat interactions with GPT, are you potentially screwed?

SPEAKER_02

I think the APIs will probably stay stable for a while. OpenAI can't just break everyone's integrations overnight. But yeah, if you're building new products, you might want to design them to be flexible about interaction models. Don't assume chat is forever.

SPEAKER_00

Let me play devil's advocate for a minute, though. What if this is just open AI getting caught up in their own hype? Like chat interfaces aren't perfect, but they're good enough for most use cases. Are we solving a problem that doesn't really exist?

SPEAKER_02

That's definitely possible. And honestly, it wouldn't be the first time a tech company threw away something that worked in pursuit of something revolutionary that nobody actually wanted. Remember when everyone tried to kill the desktop metaphor for tablets?

SPEAKER_00

Right. And users can be surprisingly resistant to change, even when the new thing is objectively better. I mean, how long did it take people to stop using Internet Explorer?

SPEAKER_02

But you know, here's the counter-argument. OpenAI has access to usage data from millions of Chat GPT users. If they're making this bet, they've probably identified real pain points in how people interact with the current system. They're not just guessing.

SPEAKER_00

That's a good point. They can see where conversations break down, where users get frustrated, where the chat model stops being helpful. If the data shows chat is fundamentally limiting, then this makes total sense.

SPEAKER_02

Plus, think about power users, the people who are pushing ChatGPT to its limits. I bet those users are constantly running up against the constraints of the chat interface. OpenAI probably wants to unlock those advanced use cases.

SPEAKER_00

And from a business perspective, if you can create a fundamentally better way to interact with AI, that's incredibly valuable intellectual property. Even if it's risky in the short term, it could define the industry for the next decade.

SPEAKER_02

Absolutely. And timing-wise, this might be their best window. They're still the market leader, they have the resources to experiment, and they haven't locked themselves into legacy systems yet. Better to cannibalize your own product than let someone else do it.

SPEAKER_00

For our listeners who are building AI-powered products right now, what's the takeaway? Should you wait to see what OpenAI does, or just keep building with current chat interfaces?

SPEAKER_02

I'd say keep building. But stay flexible. Design your products so they're not completely dependent on one interaction model. And pay attention to where your users hit friction. Those might be the same problems OpenAI is trying to solve.

SPEAKER_00

Now here's another piece of the ChatGPT puzzle that I think ties into this overhaul story. There are reports that ChatGPT is implementing continuous memory functionality that persists across all chats. So instead of every conversation starting fresh, the AI remembers context and information from all your previous sessions.

SPEAKER_02

Dude, this is huge. And it completely changes the value proposition. Right now, ChatGPT is like having a brilliant conversation partner who has amnesia. Every chat starts from zero, you have to re-establish context, explain your projects again. Persistent memory turns it into more of a long-term collaborator.

SPEAKER_00

But Sam, doesn't this also tie into the interface overhaul we just talked about? Like if ChatGPT remembers everything about you and your work, maybe the chat model becomes even more limiting.

SPEAKER_02

Oh, absolutely. That that's a brilliant connection. If the AI knows all your ongoing projects, your preferences, your work style, it doesn't make sense to interact with it through these discrete chat sessions. You'd want something more like a persistent workspace where it can proactively surface relevant information.

SPEAKER_00

I'm starting to see how these pieces fit together, but I have to ask about privacy. Continuous memory means open AI storing and connecting data across all your interactions. That's a pretty significant privacy shift.

SPEAKER_02

Yeah, that's the trade-off. You get this incredibly personalized contextual I experience, but you're essentially giving OpenAI a complete record of your digital thought process. Some people will love that, others will find it creepy as hell.

SPEAKER_00

And what about in enterprise settings? If ChatGPT remembers everything from your work conversations, how do you handle confidential information, client data, stuff that legally can't be stored by third parties?

SPEAKER_02

That's going to require some serious infrastructure changes. The memory feature is game-changing, but the data governance complexity is going to be massive.

SPEAKER_00

From a user perspective, though, assuming they solve the privacy concerns, this could be incredible for productivity. Imagine an AI that knows all your projects, remembers your coding style, understands your business context.

SPEAKER_02

Exactly, and it gets smarter about you over time. Instead of being a generic AI assistant, it becomes your personalized AI colleague. That's the kind of stickiness that makes people never want to switch to a competitor.

SPEAKER_00

But here's what I'm wondering. How do you handle the evolution of preferences? Like if I was working on web development six months ago and now I'm doing data science, I don't want the AI to be stuck on my old preferences.

SPEAKER_02

The memory system needs to be smart about what to remember, what to forget, and what to wait based on recency. Oh, if it just accumulates everything forever, you'll end up with an AI that's confused about what you actually want.

SPEAKER_00

And what about shared accounts? If you're working on a team project and multiple people are using the same Chat GPT account, whose preferences and memory does it prioritize?

SPEAKER_02

Oh man, that's a nightmare scenario. You'd probably need user-specific memory profiles within accounts or completely separate instances. The complexity just keeps growing when you think about real-world usage patterns.

SPEAKER_00

But let's say they figure all that out. The competitive implications are massive. Once you've built up months or years of memory with ChatGPT, switching to Claude or Gemini means starting over from scratch.

SPEAKER_02

That's the vendor lock-in aspect, and it's incredibly powerful. It's like how hard it is to switch from one email provider to another, except this is your entire AI relationship. OpenAI would essentially be building a moat around each user.

SPEAKER_00

Which makes me wonder if there will be regulatory pressure around data portability. Like, should users have the right to export their AI memory and take it to a different provider?

SPEAKER_02

That's a fascinating question. In the EU, you probably would have that right under GDPR. But but good luck getting that memory to work properly in a different AI system. It's not just data, it's how that data is structured and used by the AI.

SPEAKER_00

For developers building on top of ChatGPT, this memory feature could be either amazing or terrifying. Amazing because your app gets smarter over time. Terrifying because your user's data is even more locked into OpenI's ecosystem.

SPEAKER_02

Yeah, and and you you lose some control over the user experience. If ChatGPT is remembering things from other apps and conversations, your carefully designed onboarding flow might be irrelevant for returning users.

SPEAKER_00

Keep an eye on this because if OpenAI nails both the interface overhaul and the persistent memory, they're not just improving ChatGPT. They're potentially defining what the next generation of AI interaction looks like. Alright, let's shift gears to something that caught me completely off guard. SpaceX has secured a Google AI compute deal, and this is coming after they already made a deal with Anthropic. Both of these are happening ahead of SpaceX's anticipated IPO, and I'm honestly confused about why SpaceX is suddenly in the AI infrastructure business.

SPEAKER_02

Okay. So this is actually brilliant when you think about it. SpaceX has Starlink, which is this massive global network of satellites. What if they're not just providing internet connectivity, but actually offering edge computing through their satellite constellation? That would be insane.

SPEAKER_00

Wait, are you saying they could run AI models in space? Like instead of sending data down to Earth-based data centers, you're processing it on the satellites themselves.

SPEAKER_02

Exactly. Think about the latency advantages, especially for global applications. If you're in rural Africa or the middle of the ocean, instead of your data traveling thousands of miles to a data center and back, it's processed right there in orbit. For real-time AI applications, that could be game-changing.

SPEAKER_00

But hold on, that seems technically impossible. Satellites have power constraints, thermal management issues, limited processing capabilities. Can you really run sophisticated AI models in that environment?

SPEAKER_02

That's the engineering challenge, but remember, AI inference is getting incredibly efficient. With model optimization and specialized chips, you might not need massive data center-level hardware. Plus, SpaceX is already planning bigger satellites for Starlink V2.

SPEAKER_00

The timing with the IPO is interesting too. If SpaceX can position itself as not just a space company, but as a unique AI infrastructure provider, that completely changes their valuation story.

SPEAKER_02

Right, because suddenly they're not competing with other rocket companies. They're competing with AWS and Google Cloud. But with capabilities, those companies literally cannot replicate. You can't launch AWS into orbit.

SPEAKER_00

And the customers make sense too. Google needs global AI compute capacity. Anthropic is scaling up their model deployment. If SpaceX can offer global coverage with lower latency than terrestrial networks, that's incredibly valuable.

SPEAKER_02

What's wild is this could be the first time space infrastructure becomes directly relevant to everyday AI applications. Like your ChatGPT query might literally be processed in space.

SPEAKER_00

But let's think about the practical challenges. How do you update AI models on satellites? How do you handle hardware failures in orbit? What happens when you need more compute capacity? Do you just launch more satellites?

SPEAKER_02

Those are all massive engineering problems. But SpaceX has gotten really good at launching things cheaply and frequently. If you can treat satellites like disposable compute nodes that you refresh every few years, maybe it works economically.

SPEAKER_00

And there's a geopolitical angle too. If SpaceX provides AI compute through satellites, that's harder for foreign governments to regulate or shut down compared to terrestrial data centers.

SPEAKER_02

Oh, that's a really interesting point. For companies operating in countries with strict AI regulations, space-based computing could provide a way around local restrictions. That adds a whole new dimension to the value proposition.

SPEAKER_00

But what about bandwidth? Even if you're processing AI models in space, you still need to get the input data up to the satellites and the results back down. Isn't that still a bottleneck?

SPEAKER_02

For some applications, sure. But think about use cases where you're processing local data, like satellite imagery analysis, autonomous vehicle coordination, IoT device management. If the data is already nearby or generated locally, space-based processing makes more sense.

SPEAKER_00

For investors watching the SpaceX IPO, this adds a completely new dimension to consider. They're not just betting on space transportation and satellite internet. They're potentially betting on the future of distributed AI computing.

SPEAKER_02

And if this works, every other satellite company is going to be scrambling to add compute capabilities to their constellations. SpaceX might be creating an entirely new market category here.

SPEAKER_00

The scale could be incredible too. Starlink is planning tens of thousands of satellites. If even a fraction of those have AI compute capabilities, you're talking about a distributed supercomputer that spans the entire planet.

SPEAKER_02

And unlike terrestrial data centers, satellites are constantly moving, so you could optimize compute placement based on demand patterns. Need more AI processing power over Asia during business hours? Route compute tasks to satellites that are currently over that region.

SPEAKER_00

The more I think about this, the more it makes sense why Google and Anthropic would be interested. This isn't just about buying compute capacity. It's about accessing a completely new type of infrastructure that could enable AI applications we haven't even thought of yet. Now let's talk about something that's flying under the radar but could be really significant. Early reports suggest that NAVA, which is basically the Google of South Korea, is partnering with Nvidia to form what they're calling an AI factory alliance. This is part of a broader sovereign AI push. And there are also reports of NVIDIA striking multiple AI infrastructure deals with other South Korean tech firms.

SPEAKER_02

Yeah. And this sovereign AI angle is fascinating because it's not just about technology. South Korea is basically saying we're not going to be dependent on American or Chinese AI infrastructure. We're building our own.

SPEAKER_00

Help me understand the sovereign AI concept, though. What does it actually mean in practical terms? Like are they building their own chips, their own models, their own data centers?

SPEAKER_02

It's all of the above. Sovereign AI means controlling your entire AI stack from chips to models to applications, usually within your national borders. Think about it like energy independence, but for artificial intelligence. You don't want to be cut off from critical AI capabilities because of geopolitical tensions.

SPEAKER_00

That makes sense given what we've seen with chip export restrictions and trade wars. But why is Nvidia involved if this is supposed to be about independence? Isn't NVIDIA still an American company?

SPEAKER_02

That's the pragmatic reality. South Korea can't build cutting-edge AI chips overnight, so they're partnering with NVIDIA for the hardware while building everything else domestically. They get the chips, but they control the data, models, and applications.

SPEAKER_00

And NAVAR is interesting because they're not just a search company. They've got cloud services, e-commerce, messaging, even autonomous vehicles. So this AI factory could potentially power a huge chunk of South Korea's digital economy.

SPEAKER_02

Exactly. And that's probably why other countries are watching this closely. If South Korea can successfully build sovereign AI capabilities that actually compete with American and Chinese systems, it becomes a model for other mid-sized tech powers.

SPEAKER_00

I'm curious about the economics though. Building sovereign AI infrastructure has to be incredibly expensive. Is South Korea big enough as a market to justify that investment?

SPEAKER_02

That's where the export angle comes in. If they build world-class Korean AI models and infrastructure, they can export that technology to other countries that also want alternatives to US and Chinese AI. Think of it as creating a third pole in the global AI ecosystem.

SPEAKER_00

For businesses operating in Asia, this could actually be really important. Having high-quality AI infrastructure that's not subject to US-China tensions might be worth paying for, even if it's more expensive initially.

SPEAKER_02

And for developers and AI companies, this creates new opportunities. If South Korea builds genuinely competitive AI infrastructure, suddenly you have more options for where to deploy your models and serve your customers.

SPEAKER_00

But let's talk about the technical challenges. Nvidia provides the chips, but what about the software stack? Training large language models requires incredibly sophisticated infrastructure and expertise.

SPEAKER_02

That's probably where Neighbor's existing tech talent comes in. They already have teams working on search algorithms, recommendation systems, natural language processing for Korean. Building on that foundation with NVIDIA hardware could actually work.

SPEAKER_00

And there's a cultural advantage too. AI models trained specifically for Korean language and culture might perform better for Korean users than general. Purpose models from OpenAI or Google.

SPEAKER_02

Absolutely. Think about how much context is lost when you're using an AI trained primarily on English text to understand Korean business practices, cultural references, local regulations. A truly Korean AI could be significantly better for local applications.

SPEAKER_00

This also makes me wonder about data sovereignty. If all your AI processing happens on Korean infrastructure with Korean data, that's a lot more control over sensitive information.

SPEAKER_02

Right, and that's particularly important for government and enterprise applications. You don't want your military communications or corporate strategy documents being processed by foreign AI systems, even if they're technically better.

SPEAKER_00

For other countries watching this, what's the lesson? Should every nation be building sovereign AI capabilities, or is this only viable for countries with South Korea's level of technical sophistication?

SPEAKER_02

I think it depends on your strategic priorities and technical capabilities. Smaller countries might be better off forming regional AI AI alliances or partnering with trusted providers. But for mid-sized tech powers like South Korea, Japan, or European countries, sovereign AI starts to make sense.

SPEAKER_00

And the timing is interesting. If you're going to build sovereign AI capabilities, now is probably the window to do it. In five years, the technology gap might be too large to catch up.

SPEAKER_02

Exactly. Now the AI industry is still young enough that a focused national effort can potentially compete, but that window is closing as the leaders pull further ahead in terms of compute resources and data.

SPEAKER_00

Sam, this feels like a big scaling moment for Anthropic.

SPEAKER_02

Absolutely, and the timing is perfect given all this SpaceX infrastructure news. Anthropic is clearly preparing for massive scale, and having 150 enterprise customers gives them real-world feedback to improve the model. Plus, enterprise customers have different requirements than consumers, better security, compliance, customization options. Getting that right with 150 organizations is incredibly valuable data for Anthropics product development.

SPEAKER_00

And 150 is a meaningful number. That's not just a limited beta, that's real commercial deployment. These organizations are probably paying significant amounts for access to Claude Mythos.

SPEAKER_02

Right, and each of those 150 organizations probably represents hundreds or thousands of end users. So Anthropic is getting usage data from potentially tens of thousands of professional AI users. That's gold for model improvement.

SPEAKER_00

Next story. This seems to be referencing some major investment deal between Anthropic and Google.

SPEAKER_02

Wait, 2.2 billion? That's not just an investment, that's a statement. If Google is putting that kind of money into Anthropic, they're basically betting that Claude becomes a serious competitor to GPT.

SPEAKER_00

And the SpaceX IPO connection suggests that investors are viewing AI infrastructure as the key value driver. Like the rocket business is proven, but the AI infrastructure potential is what could make this IPO massive.

SPEAKER_02

Right, because 2.2 billion in I deals shows there's real demand for the compute capacity that SpaceX could provide through their satellite network. It's validation of their AI infrastructure strategy. Sure.

SPEAKER_00

But here's what's interesting. If Google is investing $2.2 billion in Anthropic and also making deals with SpaceX for AI compute, Google is essentially hedging their bets across the entire AI value chain.

SPEAKER_02

They're also investing in potential competitors and alternative infrastructure. If any part of the AI ecosystem takes off, Google has a position.

SPEAKER_00

For SpaceX IPO investors, this validates that there's serious enterprise demand for alternative AI infrastructure. These aren't speculative bets. These are multi-billion dollar commitments from established tech giants.

SPEAKER_02

And it shows that the AI infrastructure wars are really heating up. Companies are willing to pay premium prices for compute capacity that gives them strategic advantages, whether that's lower latency, better security, or just alternative supply chains.

SPEAKER_00

If you zoom out and look at everything we covered today, there's this fascinating theme about the next phase of AI competition. It's not just about who has the best models anymore. It's about who controls the infrastructure and who defines how people interact with AI.

SPEAKER_02

Exactly. And what's wild is how different the strategies are. Open AI is betting on revolutionary interfaces. SpaceX is betting on space-based computing. South Korea is betting on sovereign infrastructure. And Google is just throwing billions at everyone to make sure they have a seat at every table.

SPEAKER_00

The infrastructure wars are really heating up. Five years ago, AI was about algorithms and data. Now it's becoming about satellites, chip factories, and geopolitical alliances.

SPEAKER_02

And for regular people and businesses, this fragmentation could actually be good news. Instead of being locked into one or two AI providers, we're heading toward a world with multiple competing infrastructures and interaction models.

SPEAKER_00

The question is whether all this innovation leads to better AI experiences or just more complexity. Like, do people really want AI in space if their current AI works fine?

SPEAKER_02

I think it depends on what new capabilities emerge. If space-based AI enables real-time translation for global calls, or if new interfaces make AI collaboration seamless, people will adopt it quickly. The key is whether the technology improvements are actually noticeable to end users.

SPEAKER_00

What's really striking to me is how quickly the landscape is changing. Two years ago, ChatGPT was this revolutionary chat interface. Now OpenAI is reportedly killing chat, and we're talking about AI processing in orbit.

SPEAKER_02

And that pace of change is only accelerating. The companies that are making big infrastructure bets now, like SpaceX with satellites, South Korea with sovereign AI, they're betting on where AI will be in 2030, not where it is today.

SPEAKER_00

The geopolitical dimension is fascinating too. AI is becoming like energy or telecommunications. Critical infrastructure that countries need to control for national security reasons.

SPEAKER_02

Right, and that's driving a lot of the sovereign eye initiatives we're seeing. It's not just about having the best AI models. It's about ensuring you're not dependent on potentially hostile nations for critical AI capabilities.

SPEAKER_00

For businesses trying to navigate all this, the key seems to be avoiding lock-in while still taking advantage of new capabilities. You want to benefit from innovations like persistent memory or space-based compute without betting your entire company on any single provider.

SPEAKER_02

Absolutely. And I think we're going to see more demand for interoperability and standards. Businesses want to be able to move their AI workloads between providers, export their data, integrate multiple AI systems.

SPEAKER_00

The other thing that strikes me is how capital intensive this all is. We're talking about billions of dollars for satellite constellations, sovereign AI infrastructure, major model training runs. This is not a garage startup kind of game anymore.

SPEAKER_02

Which might actually create opportunities for smaller players to focus on specialized applications or niche markets. While the big players fight over general-purpose AI infrastructure, there's probably room for focused solutions in specific industries or use cases.

SPEAKER_00

Looking ahead, I think we're entering a phase where AI infrastructure becomes as important as the AI models themselves. The companies that control the pipes might be more valuable than the companies that create the content.

SPEAKER_02

Oh, and that's why these infrastructure plays, you know, whether it's SpaceX's satellites, South Korea's sovereign AI, or your open AI's interface innovations are so strategically important. They're positioning themselves to capture value from the entire AI ecosystem, not just model performance.

SPEAKER_00

That's a wrap for today's build by AI. Honestly, if even half of these infrastructure changes happen, 2026 is going to be remembered as the year AI went from chatbots to something completely different.

SPEAKER_02

Yeah, and we're just getting started. Make sure to subscribe wherever you listen to podcasts because tomorrow we'll probably be talking about AI robots in space or something equally wild.

SPEAKER_00

Thanks for listening, and we'll see you tomorrow.