zachthewf

what do you think people should do when their countries experience hyperinflation?

What company are you at? I've been working on a product to solve this problem.

If you're up for a chat my email is on my page here https://www.zachocean.com/contact/

Were you able to find good candidates from your post eventually?

I mean, inference costs have decreased like 1000x in a few years. OpenAI is the fastest growing startup by revenue, ever.

How foolish do you have be to be worrying about ROI right now? The companies that are building out the datacenters produce billions per year in free cash flow. Maybe OP would prefer a dividend?

Exactly right. Traditional VCs come up with a thesis and then buy 20% of the company they think will be the winner who fits that thesis when they're at like $1M to 10M revenue (series A)

YC can instead get ~10% of every plausible winner they come across when they're at $0 revenue

What's your use case?

Can't wait for them to move up to Seattle. We're basically San Francisco with slightly easier hills and more rain/occasional ice, so seems like a good candidate.

I love Tesla, I own the stock, but I don't think they can be considered comparable until they have a live autonomous taxi service that regular people can call from their phones and regularly rave about.

Ginormous market, 10x better product, no competition is currently close.

Plus although they have the capital cost of owning the cars they don't have the COGS of paying the drivers so presumably much higher margin per ride.

Cool, looks like this is trained on 16 million hours of audio (500B tokens at ~.11 seconds per token).

Even the large open source TTS models (see F5 TTS, Mask GCT) are mostly trained on very small audio datasets (say 100k hours) relative to the amount of audio available on the internet, so it's cool to see an open source effort to scale up training significantly.

Pedantically, literally everyone can be a founder.

At this moment, you can decide that you are starting a company. You are now a founder.

(This is why Founder is historically an unprestigious job that smells a lot like Unemployed, though the startup hype cycle of the last 15 years changed that.)

I don't remember exactly, but somewhere in the neighborhood of $1,000.

We had a great experience doing this in a much less hardcore way through this company: https://withtheseringshandmade.com/

I'd recommend to anyone who is interested but does not have the skills to make a ring on your own. Great weekend trip from Seattle too.

I think the only way to avoid race to the bottom is to offer relatively high end services, competing on price is a recipe for disaster.

What do you mean by selling Education?

(I'm the author)

The "why now" is any significant new efficiency gains that are available without a ton of cash.

> "Is the kernel of this advice to start with a known problem which is already solved with a paid service, and then just sell that service more efficiently?"

Yep, this captures what I'm saying.

> "why is this a better strategy for a startup"

If you can build a differentiated product it's great to do that, but it's a lot more difficult/expensive than it used to be.

Incumbent product companies will pursue full automation, but probably won't compete directly with service businesses because it's not their wheelhouse. E.g. GitHub built Copilot and maybe they'll try to ship a "fully automated" AI software engineer. But until their AI software engineer is actually fully automated (which I would guess is 10 years away at least), they won't be competing with software development agencies.

Even if AI doesn't help with any aspect of the service itself there are a ton of efficiency gains to be had from "the rest" of the work, which will make service margins less ugly. Every business still neeeds some form of sales and/or marketing, support, and back office administration.

And for a lot of services there'll be creative ways to squeeze efficiencies from tech/AI. For example roofing businesses used to send a guy to your house to give you an evaluation + quote; increasingly, they now send drones (https://www.roofer.com/).

Nice, what do you guys do?

I wonder what (if anything) is the impact of the leading spaces on each line of the multiline string, which are an artifact of wanting to keep the prompt pretty within code.

Hopefully not much, but I've heard horror stories about trailing spaces...

Don't get it twisted—the MIT Startup is still based in SFBA

The publicity might be a short term win but there is a dangerous narrative for Apple that it feeds: that they are no longer a design-obsessed company that prizes art and creativity and channels that obsession to build the best products.

How cool! It's like Foldit (https://en.wikipedia.org/wiki/Foldit) for surgery videos.

GPT3.5-level models are now insanely cheap: $1 buys you something like a Moby Dick's worth of text. STT and TTS models are getting cheaper too. I'm guessing that they are betting usage over the lifetime of the device (maybe 2 years tops?) amortizes out to a decent margin without a subscription.

It's also a lot easier to stomach a beautiful shiny teenage engineering toy that doesn't do much if you don't have to pay every month.

what do you think should happen when you start a company that grows to $30b market cap like cloudflare?

Expressivity and naturalness are still not good enough. It's getting there, but today you'd never have a voice conversation with an AI and come away thinking "boy, what an engaging, charismatic personality that AI had!" This is partly due to latency, partly due to imperfect speech-to-text which is too slow and doesn't understand conversational dynamics or nonverbal signals, but yes partly due to text-to-speech.

When it comes to naturalness for dialog, OpenAI actually has the best voices I've heard—significantly better than Azure or 11 Labs which are second and third best. But at least the voices they expose through ChatGPT are fairly muted and inexpressive.

Hard for me to imagine that this could be solved in text space. I think the prediction task needs to be done on the audio.

Good way to start is to force yourself to post (as yourself, ideally with your work) on the internet every day

You are right, but using the 409a pricing would not really be honest either. (In the sense that the 409a is basically "as low as we could reasonably value this stock" in order to keep the strike prices low.)

I didn't know that. Maybe it's not good enough so they figured they had to train on customer data to ship something high quality, hence this change.