GPT-4o launches, Glue demo, Ohalo breakthrough, Druck's Argentina bet, did Google kill Perplexity?

Within roughly two years of May 2024, the basic building blocks of AI models and capabilities will have standardized across the industry, and only after that point (i.e., starting around 2026) will the dominant, large-scale AI application companies (“the Facebook of AI”) be created; as of May 2024, that dominant consumer AI application company does not yet exist.

“So I think where we are is probably within two years of where the basic building blocks are standardized. And then I think the real businesses get built. So I will maintain my perspective here, which is the quote unquote Facebook of AI has yet to be created.” View on YouTube

AI foundation models will continue to become roughly 90% cheaper to run each year, implying that over any two-year period they will be approximately 99% cheaper and better than models from two years prior.

“these new models are so much more efficient that you actually can throw the old model in the garbage, because it's so inefficient, and these are now becoming about 90% cheaper every year, which means every two years these things are gonna be 99% cheaper and better.” View on YouTube

Over time, OpenAI will shift its primary business model away from direct B2C subscriptions toward B2B monetization, mainly by charging for and monetizing applications and services built on top of its models.

“Well, I think you're assuming there that the long term business model of OpenAI is in B2C subscriptions. And I think that's probably the least attractive business model they have available to them... So I suspect they're going to move in more of a B2B direction over time, because that's where the real money is. And probably the way they do that is by monetizing all the apps that are built on top of it.” View on YouTube

OpenAI’s long‑term strategy for ChatGPT/GPT access will be to make core individual usage effectively free or near‑free, and to monetize primarily by charging for multi‑user or collaborative (“multiplayer”) versions or features.

“OpenAI ... on a strategic level, is going to make all this free or close to free, and maybe just charge for a multiplayer version. That seems to be where it's heading.” View on YouTube

Because of GPT‑4o and similar advances in foundation models, most of the R&D invested over the last 2–5 years into building proprietary conversational/virtual customer support agents (especially work focused on core conversational ability and speed) will be rendered obsolete, and new customer support agent startups built on the latest models from 2024 onward will have a structural advantage over those earlier efforts.

“there are a whole bunch of startups out there that were creating virtual customer support agents... I think their product roadmaps just became obsolete... my guess is that hundreds of millions of dollars of R&D just went out the window. And probably this is the best time to be creating a customer support agent company if you're doing it two years ago, five years ago. Your work is just like been obsolete.” View on YouTube

For application developers who build features that overlap too much with what frontier AI models are learning to do natively, those overlapping features will be rendered obsolete within a few months of major model upgrades.

“If you're an app developer, the key thing to understand is where does model innovation end and your innovation begin? Because if you get that wrong, you'll end up doing a bunch of stuff that the model will just obsolete in a few months.” View on YouTube

In the AI era, companies that build internally useful but non‑core AI systems (such as customer support agents like Klarna’s) will increasingly choose to open-source those systems rather than maintain them privately, leading to a stronger push toward open-source infrastructure in AI than in prior software markets.

“That's why I think the incentive for these folks is going to be to push this stuff into the open source... Release it in the open source, guys. Let the rest of the community take it over so that it's available to everybody else. Otherwise you're going to be stuck supporting it... So I also think the incentive to just push towards open source in this market, if you will, is so much more meaningful than any other market.” View on YouTube

Future AI models used within Glue (and similar enterprise chat tools) will become capable of automatically inferring which internal data source or repository to query, without users having to explicitly specify it, making AI assistance more seamless than the May 2024 state.

“in the future it's going to figure it out on its own. So it's going to become more and more seamless, but.” View on YouTube

By the time a GPT‑5‑class model is available, enterprise chat tools like Glue will be able to support effectively “promptless” AI behavior, where the AI autonomously interjects into conversations when it detects it has relevant, helpful information, without being explicitly summoned by a user.

“That's where I want to go with it is I call that prompt list, which is I want the AI just to chime in when it determines that it has relevant information and can help the team, even if it hasn't been summoned yet. But we need some model improvement for that, frankly. I mean, we'll be able to get there by GPT five, but that's totally where this is headed.” View on YouTube

Ohalo’s “boosted breeding” technology will be applied across nearly every major crop worldwide, leading to higher yields and cheaper seed production, which in turn will lower food prices while improving sustainability metrics (less land, water, and energy per unit of food) compared to current agriculture.

“So we're going to be applying this boosted technology that we've discovered across nearly every major crop worldwide. It'll both increase yield, but it will also have a massive impact on the ability to actually deliver seed and help farmers and make food prices lower. Is it improve sustainability? No, it's actually cheaper. So higher yield, lower cost.” View on YouTube

Compared with conventional crop varieties, boosted‑breeding crops from Ohalo will require less water, less land, and less energy per unit of output to achieve their higher yields.

“Do you need more water. … Less water, less land, less energy?” View on YouTube

Once Ohalo’s boosted‑breeding crops reach commercial trials and deployment, fertilizer required per pound of harvested output will decrease significantly compared with current crop systems, due to improved efficiency despite higher biomass.

“Fertilizer usually scales with biomass. But these sorts of systems should be more efficient. So fertilizer use per pound produced should go down significantly. As we get to commercial trials with all this stuff.” View on YouTube

Within approximately 5–10 years from May 2024, typical company organizational structures and workflows will be transformed by AI and related tools to such an extent that they will be barely recognizable compared to how companies operate today (much smaller headcount relative to output, far lower opex, heavy automation/delegation).

“The way that companies will work in five and ten years, I don’t think guys, any of us are going to recognize what it’s going to look like.” View on YouTube

Over the coming years of the AI-driven productivity boom (implicitly the next several years after 2024), overall unemployment in the U.S. will remain low rather than rising significantly, with the primary adjustment being the creation of many more (often smaller) companies rather than widespread joblessness.

“The unemployment rate is going to stay very low. We’re just going to have more companies.” View on YouTube

A new type of company will emerge (and become relevant at scale) that specializes in clearing/licensing publisher content rights for AI-answer products like Google’s AI Overviews, acting as an intermediary between content creators and AI platforms.

“There’s going to need to be a new company that clears this content so that Google can do answers like this.” View on YouTube

Isomorphic Labs, Google’s drug-discovery/biotech AI unit, will ultimately be worth on the order of multiple hundreds of billions of dollars in enterprise value (i.e., comparable to a standalone multi-hundred-billion-dollar company) once its business is mature.

“Between last week’s announcement of Isomorphic Labs, which let’s be honest, that’s a that’s just a multi hundred billion dollar company.” View on YouTube

Gemini-powered AI Overviews/one-box style results will become the dominant interface for Google Search, effectively displacing the traditional ‘10 blue links’ paradigm as the primary way users consume search results.

“It’s very clear now that Gemini powered one box is the future of Google search. People just want the answer. I think that this feature is going to eat the rest of Google’s search.” View on YouTube

Perplexity.ai, as an independent search/answer engine, will struggle severely or fail competitively now that Google has copied its core product approach; its most viable positive outcome is to be acquired rather than to succeed at large scale as a standalone competitor.

“They got caught completely flat footed. And they’ve now, I guess, caught up by copying perplexity and sucks for perplexity. I think they’re kind of screwed now unless they get over an acquisition deal.” View on YouTube

Alphabet/Google’s market capitalization will reach approximately $5 trillion at some point in the future, despite earlier AI missteps, driven in part by successfully copying and integrating AI features into search and other products.

“Same thing here with Google. They completely screwed up AI. They invented the transformer. Completely missed LLMs…Doesn’t matter…they’re probably going to become a $5 trillion company now.” View on YouTube