WTF is happening w AI, pace of progress is wild
Literally pushed out my post last night talking about how it wouldn't be long before voice was integrated, now I see this on Twitter:

Audio being integrated into LLMs TODAY.

Folks, I'm serious. 2023 is the year the Turing Test will be passed, human equivalent digital personal assistants may end up being even closer than I thought.

We are at the knee of the exponential curve of progress.

AverageInvestor's avatar
I wonder how profitable creating AI like this will be. Do we start to see niche AI for specific industries, paywalled advanced AI or will this follow a data-selling model (targets ads and whatnot). It’s absolutely revolutionary technology that essentially
gives everyone a “coach” to work with. How are OpenAI and competitors going to turn a long term profit and continue to push the innovation envelope?
Josh Rozin's avatar
@averageinvestor There's an app I've been following since inception, . They've just hit $1m MRR and have considerably consistent growth previous to this recent AI wave.
Benjamin Buchanan's avatar
@averageinvestor This is a reallllllly good question and one I don't have a high confidence answer for.

My guess is two fold: 1) tons of apps will be built on top of LLMs with specific use cases, much like mobile apps today. I walked through one potential use case in my last post - companionship, which I believe will turn into a $10B+ industry within 10 years. Education will likely be larger (via tutoring apps built on LLMs) sooner

2) there will be a few general purpose LLMs that become digital personal assistants that will have a subscription/ad supported model. My guess is the companies with operating systems end up dominating this space, so that would be Android, iOS, and Windows. I'm conflicted about whether Apple ends up using its own LLM (seems likely if it can pull it off). Google obviously won't and Microsoft already is using OpenAI and seems to be taking that path (regularly hearing about deeper integration and they are the largest investor in OpenAI and have an exclusive license to the source code).

I will also say that until recently I would never have believed that a start-up like OpenAI could build something so vastly superior to what the Tech-Giants can build. But Microsoft going deep with Open AI and Meta's HUGELY embarrassing launch of their own LLM last November both leave me with a degree of uncertainty.

As to how OpenAI would monetize though, given how variable the costs are of running GPUs (HUGE variability of 7X between a query that takes ChatGPT 27 seconds to respond to vs 5 seconds or so) - seems that the business model will almost have to be tied to use - similar to AWS and snowflake

Continuing with the napkin math from my post, if ChatGPT has a cost per user per day (for someone using as a full blown digital personal assistant) of $1.5 today, that might drop by half by say two years from now. If their cost is $.75 they could charge a DPA app $1.5 and have a 50% gross margin (which would continue increasing as they scale and improve performance efficiency). That would mean the DPA itself could have a 50% gross margin if it charged $3.0 per day, or $90/month. That's not a ridiculous price for what it would be capable of.

Something more limited (for example only allows you to do X queries or generate X tokens per day) could cost 1/3 of that (less than a cell phone bill) and be ad supported?

Second-to last thought, I think now that ChatGPT has shown what's possible - other companies might catch up quickly, especially the tech giants themselves (Google might be ahead as I explained in a previous post, just hasn't released the product yet). Big-Tech has effectively unlimited money and I have to imagine they'll be able to catch up even if it requires paying exorbitant sums to steal engineers.

My last thought is - as I explained in my post yesterday (have you read it? If not it's linked on my profile) - having an LLM come between big-tech and their users is something they will prevent at all costs - this might end up resulting in the price and margins both being lower than you would otherwise expect, because the LLM itself would end up becoming required to protect their moat. If one company offers one for free then everyone might have to do the same. Similar to how Amazon uses zero margin retail to defend their 3PL and ad business, they may end up using LLMs to defend their other sources of revenue.
Benjamin Buchanan's avatar
@josh Do you know what LLM it's built off of?
Benjamin Buchanan's avatar
@averageinvestor I actually have one more thought I'd like to share. Big Tech is TERRIFIED of being seen as creepy and invasive. They don't want to be associated with products that feel sentient (though ChatGPT may change this, easier to do it if someone else has done it first).

The absolute KILLER app that I believe will eventually turn into a $1 trillion market are customized digital personal assistants that not only do everything an assistant would, but also have personalities that are customized to ours, voices that are specific to our DPA, and Avatars that are specific to our DPA (life-like). Imagine how powerful it would be to essentially have an AI best friend to not only make you insanely productive, but to encourage you when you are sad, tell you jokes it knows you will think are funny at random times, etc. People WILL love this and befriend these things.

Because this will seem so creepy (at first, until people adapt) - Big Tech might be unwilling to go there - this might open the door to a start-up to take a significant share of the market - especially the younger people who wouldn't think twice about using something like this. I know I would prefer to use something like this as opposed to a DPA that looked and appeared like everyone else's - sort of like how everyone knows the Siri voice.

Just spit ballin...
AverageInvestor's avatar
@01core_ben some of your points lead me to a search which lead me to this twitter thread that breaks down the cost. Very interesting if true.

AverageInvestor's avatar
@01core_ben this is next level stuff! I’m over here trying to figure out how to use OpenAI to increase my productivity and you’re already envisioning a realistic growth trajectory. Great stuff!



Already have an account?