[darkwater]

> Couldn’t agree more. Crypto was a monetizable solution in search of a problem.

True

> AI solves real problems

AI research and real products, most probably. All those new companies/startups that are wrappers around ChatGPT? Not that much. Maybe someone will do something that really brings value but most of them sound and act like the Blockchain startups we had til 12 months ago.

[doctor_eval]

GPT wrapper products are the equivalent of static corporate HTML pages from the 90s. We build them because we are the early adopters of a brand new technology and we’re working out how it works.

I’m working with such a company and yes, the first product is a (very nuanced and very domain specific) wrapper around GPT. But as we gain experience, we expect to develop far more complex and interesting products. But we gotta start somewhere.

2 years ago, nobody even knew about ChatGPT. We are at the very start of this wave and we’re just still plucking the low hanging fruit. The fact that GPT wrappers are useful speaks directly to the potential of the technology.

Contrast this with crypto, which wasn’t even capable of solving the problems it was supposed to solve, and where losses incurred by things like smart contracts were almost inevitable. The products being built with AI are real products that real people are paying for. The difference couldn’t be more stark.

[vineyardmike]

> 2 years ago, nobody even knew about ChatGPT

So chatGPT isn’t 2 years old, but i recently discovered that if you search HN for GPT-2, you’ll discover years of people discussing it, below hype levels. You’ll find “did GPT2 write this” type comments that we see today with 3,4. You’ll see people train it against their texts to talk to themselves. A lot of the ChatGPT induced hype hypotheticals were already done.

[doctor_eval]

> A lot of the ChatGPT induced hype hypotheticals were already done

I really disagree with this.

The more broadly useful and available this technology becomes, the more interesting and unexpected uses will be discovered for it. The people I'm working with now are not remotely software people but they are generating real value from GPT in ways I would never have imagined. I look at what they're doing and I think to myself - a seasoned software engineer - OMG that's incredible, I would never have thought to use it that way.

That's why I say that GPT4 is the 56K modem of AI. Until now it's just been largely software geeks. But now that it's getting out into the broader world, people are coming up with novel ways to use it that are unexpected and occasionally ingenious. Ways to help people that most software engineers wouldn't have the domain experience to think of.

And they're just getting started.

[apexalpha]

Have you seen the new AI photoshop where you can just AI fill anything? And it just works?

The amount of people who can edit photo's is now basically unlimited. That's thanks to generative AI.

Same like how the computers enabled billions to do the most complex calculations within a second, while before you needed to learn that skill over years or practise.

[ben_w]

> Have you seen the new AI photoshop where you can just AI fill anything? And it just works?

> The amount of people who can edit photo's is now basically unlimited. That's thanks to generative AI.

It's also available at zero cost, thanks to the ecosystem around Stable Diffusion.

You probably know the meme that gets parroted randomly about OpenAI needing a moat; well, what's Adobe's moat, now that this exists?

[fxtentacle]

Adobe's moat is that GIMP still cannot properly edit 4-channel CMYK images. And none of the competitors can print with ICC profile correction. I kid you not, there is no way around Adobe for any kind of serious image editing.

Adobe's second moat is PDF. And it builds upon their imaging moat because companies want their product boxes to be sharp and with accurate colors, so there is an entire industry built around PDF-based print and cut and glue manufacturing.

So whether or not they throw an AI bone to the kids is probably irrelevant to their bottom line. My guess would be that Adobe added those AI features mainly to make their company look more sexy to potential future hires.

[zirgs]

CYMK and ICC profiles are irrelevant in gamedev and vfx industries though.

[fxtentacle]

You need ICC if you want your digital to film printer to generate accurate results.

But yeah, for gamedev Adobe Photoshop doesn't have that much benefit over specialized tools like Substance Painter .. oh wait, they recently bought that company ;)

[rickdeckard]

Adobe's business model is to provide you the benefits of wild-west AI with the legal stability of an Adobe service.

With Firefly they offer a tool an art agency can use daily without having to worry about hidden legal implications of using AI...

[dbmikus]

The design world runs on Creative Suite, and the Figma acquisition improved that position. There's a network effect to the software, the file format, and creative team collaboration tools.

Given this, competitors need to be meaningfully better, not just equivalent or somewhat better. Adobe is still in the position to be the go to product suite for creative professionals.

[stale2002]

> well, what's Adobe's moat, now that this exists?

Although this is debatable, lets accept it for the sake of argument that adobe has no moat.

If this is the case, this may be bad for adobe, but it is amazing for the design world. Producing more art for cheaper is real value.

And we may not know who specifically will capture that value, but it is undeniably value none the less that will go somewhere. That still benefits the economy as a whole, even if we can't guess which AI startup will win the race.

[ben_w]

> That still benefits the economy as a whole, even if we can't guess which AI startup will win the race.

One possible outcome is that the entire economy benefits without any AI startup winning — even if Stability AI goes under for whatever reason, the various Stable Diffusion models are still out there getting used to make pictures.

[trashtester]

> Producing more art for cheaper is real value.

Except that if art production becomes too easy, the prices go down, and the only party benefiting is the consumer/customer.

[stale2002]

Fortunately, the only party that matters is the consumer.

Businesses may exist and be necessary to serve the consumer. But they are not valuable in and of themselves.

[zirgs]

As if that's a bad thing.

[trashtester]

Generally, it's not.

Though it tends to disappear into effects that are taken for granted.

Like the improvement in mobile phone quality since 1990.

Just like how easy access to music through Spotify may make us appreciate music less than if we have to actually put a vinyl record on, easily accessed art may also reduce our perceived appreciation.

[vineyardmike]

Well stable diffusion’s parent company is nearly dead due to funding issues. So competing isn’t something they’re succeeding at.

The moat memo misses the truth that google knows: you need existing customers. Ideally paying customers.

Firefly has doesn’t have a moat, it has a well. It has a way to sustain and fund development and growth and servers. That’s why google doesn’t need a moat. They have the funds to take on anyone, and probably win.

[martin_a]

I'm regularly using the Photoshop function to test its limits and applications to our printing business. It's kind of impressive at first but it lacks quality where it matters, especially with the more complicated jobs.

Also "editing photos" is more than what the AI does. We have solutions for that, on another level though, for the last 15-20 years already.

[Kaibeezy]

> It's kind of impressive at first but it lacks quality where it matters, especially with the more complicated jobs.

Same with the higher-complexity written tasks I’ve been chipping away at. The quality dropoff is a cliff. Crossing that chasm feels like it will take more than incremental improvement of current tools; it feels like an entirely new technology is needed.

[bombolo]

I'd expect the real implementation to not be nearly as good as the demo shows.