each of these are known >$100m opportunities:

Generative Text for writing - Jasper AI going 0 to $75m ARR in 2 years

Generative Art for non-artists - Midjourney/Stable Diffusion

Copilot for knowledge workers - both GitHub’s Copilot X and “Copilot for X”

Conversational AI UX - ChatGPT / Bing Chat, with a long tail of Doc QA startups

(shameless plug from https://www.latent.space/p/agents)

> Generative Text for writing - Jasper AI going 0 to $75m ARR in 2 years

And laying off a ton of staff in July..... the latest ARR numbers are not nearly as good.

character.ai

I think there are some killer products, but the surface area here is not that big. I can definitely see how we can improve on all of the above, but at the rate things are developing right now, you're looking at smaller and specialized models as the underlying infrastructure. Keeping the conversation focused on Anthropic, unless there is promise of some specific functions outside this range that require big powerful models, they're in trouble.

There is a world in which the dust settles and the current "era" of AI doesn't actually result in a significant amount of productivity/value creation and capture thereof. Everything rests on the current assumption that emergent behaviors and some semblance of consciousness can extrapolate infinitely. You have to believe that in order to justify the investments we're currently seeing in some of the big players.

Do they have a moat though? When MS Office integrates a more powerful model than is even available to Jasper right into the word processor, what happens?

If you can train a machine using video, and get it to repeat what you have taught it...and then manufacture millions of them - you have an (almost)-zero-marginal-cost revolution just like you had for software development.

Plus, it's replacing $$$ humans. 4X the productivity while paying for itself in a few months.

>If you can train a machine using video, and get it to repeat what you have taught it

I might be wrong, but from what I know about Transformer models and neural nets in general, what you are describing sounds a lot closer to something like an AGI.

When you say “train a machine” here it sounds a lot closer to “teach a machine” rather than “finding weights and baises for an existing function”.

I think we are still a ways off from AGI.

I dont think AGI is required. But multimodality for sure and some piece to plan/execute actions. Prompting techniques now, would land you clear tasks, validations and such, given some objective "replace table leg". A unit to process this would be required.