> The amount of money it's burned on this is giant
It's big, but it's honestly not that big. Most importantly, costs will quickly come down as we realize the limits of the models, the algorithms are optimized and even more-dedicated hardware is built. There's no reason to think it isn't sustainable, it will add up just fine.
But yes, it will attract a ton of advertising, the same curve every service goes through, like Google Search, YouTube, Amazon, etc. Still, just like Google and Amazon (subtly) label sponsored results, I expect LLM's to do the same. I don't think ads will be built into the main replies, because people will quickly lose trust in the results. Rather they'll be fed into a separate prompt that runs alongside the main text, or interrupts it, the way ads currently do, and with little labels indicating paid content. But the ads will likely be LLM-generated.
This is honestly why I struggle to get excited for anything in our industry anymore. Whatever it is it just becomes yet another fucking vector for ad people to shove yet more disposable shit in front of me and jingle it like car keys to see if I'll pull out a credit card.
The exception being the Steam Deck, though one could argue it's just a massive loss-leader for Steam itself and thus game sales (though I don't think that would hold up to scrutiny, it's pretty costly and it's not like Valve was hurting for business but anyway) but yeah. LLMs will absolutely do the exact same, and Google's now fully given up on making search even decent, replacing it with shit AI nobody asked for that will do product placements any day now, I would bet a LOT of money on it.
> I don't think ads will be built into the main replies,
> because people will quickly lose trust in the results.
The 'best' ads will be those the public doesn't recognize. Surf the internet without an ad blocker and you will die from a heart attack. This is a matter of conditioning users. It will take some time. Case in point: people already give up on privacy because "Google knows about everything already", which reflects a normalization of abuse, as we started from trust and norms ("don't be evil").
Actually, I feel like most of the money will come from enterprises. Every company will need an LLM subscription to stay competitive. I think it's possible that consumers will get a free ride with a small amount of quota without ads.
Product placement in movies and such has been a thing for a long time now. At best you can hope that your prompts will be classified as factual-vs-entertainment, and the product placement will only happen in the entertainment ones.
It's big, but it's honestly not that big. Most importantly, costs will quickly come down as we realize the limits of the models, the algorithms are optimized and even more-dedicated hardware is built. There's no reason to think it isn't sustainable, it will add up just fine.
But yes, it will attract a ton of advertising, the same curve every service goes through, like Google Search, YouTube, Amazon, etc. Still, just like Google and Amazon (subtly) label sponsored results, I expect LLM's to do the same. I don't think ads will be built into the main replies, because people will quickly lose trust in the results. Rather they'll be fed into a separate prompt that runs alongside the main text, or interrupts it, the way ads currently do, and with little labels indicating paid content. But the ads will likely be LLM-generated.