Hacker Newsnew | past | comments | ask | show | jobs | submitlogin
YC offers early interviews for AI companies (ycombinator.com)
49 points by craftsquick on July 15, 2023 | hide | past | favorite | 81 comments


The AI feeding frenzy in VC land is reminiscent of the early days of the public access internet aka the .com boom. That corrected itself sharply but eventually most of those .com boom dreams came out, just a bit later and lots of VCs lost lots of money because they got the timing wrong.

Every time there is a major shift in tech you see this pattern repeat but I don't recall seeing it on that scale again since ~2000, even though lots of stuff has happened since then (for instance: the mobile revolution).

Investing in AI for the sake of companies using AI because it increases their chances of funding is the hallmark of a bubble, and that will probably end the same way it did before. But if you manage to pick the winners early it will pay off in spades. The big question to me is whether or not those winners will be part of the first wave or the second, usually it takes a while before a new technology really comes into its strength.

Until then there is a pretty big risk that what looks to be an early winner will be displaced by a party from the second wave which is better positioned to take advantage of later, more mature versions of that technology, which by then may well be able to displace those 'early winners' by a couple of guys with an open source package.

Not an easy time to be an investor.


Total addressable market and value proposition are key.

A specific technology doesn't really change either of these.

"I am going to sell pet supplies to people who need pet supplies"

Okay, how many people need pet supplies and why would they use you instead of the store down the street (in the 90s)?

"I am going to sell books to people who need books"

Is obviously a better proposition, even before we get into the fact that it's "online" or uses any specific technology!


Why is the second one a better proposition?


Because the TAM of people who want books is a lot larger than people who want pet supplies, and there are compelling reasons why people would rather buy books from centralized warehouses than local inventories (namely, that books have a long tail and pet supplies do not).


This makes 0 sense. What makes you think the market wasn’t already adequately serving people who wanted books? New technology is the only reason this sales pitch worked. It’s not enough to provide books. You need to provide more books at cheaper prices than existing incumbents. Without knowing the competitive landscape, you can’t say whether or not selling books is better than selling pet supplies.


Types of dog and cat foods are more substitutable than book titles, which decreases the value proposition of warehousing a larger variety.

And on the profit side, books don't spoil and are small and light.

Bezos is on record in an interview somewhere as saying that he didn't care what they sold on the internet -- they picked books (and later music and movies) because they were good fits to store and ship.


This was your original statement:

> Total addressable market and value proposition are key.

> A specific technology doesn't really change either of these.

Bezos chose books because they ship well. This is because he was going to use a new technology (the web) to sell goods and needed items that would work with this model well. So the decision to sell books had little to do with TAM and everything to do with technology. Basically the opposite of what you originally asserted.


Technology can't create a value proposition where none possibly exists.

There was no Pets.com business model that would have worked, with any technology shy of a free energy transporter.

In contrast, Amazon proceeded thusly.

1. There are a lot of people who buy books.

2. People who buy books would value a wider selection. (than current local options make available to them)

... therefore, what combination of technologies exist that allow Amazon to deliver on the above? (web storefront + centralized warehousing + efficient shipping)

The point of bubble businesses is that the first two components, neither of which have much to do with technology, are often absolutely missing.

Fundamentally sound business plan + new technology >> ?? + new technology

Too many bubble businesses are the latter, and people invest in them despite the ??.


> There was no Pets.com business model that would have worked, with any technology shy of a free energy transporter.

The Pets.com business model is essentially the same as Chewy, which does work. As far as I know, Chewy has not discovered a free energy transporter.


Chewy is Pets.com + being owned by one of two major pet related brick-and-mortars + diversification into vet and pharmacy services.


Chewy was already a multi-billion dollar company by the time it was purchased by PetSmart. Chewy is now publicly traded at a value of over 16B.

Not sure what your point is regarding vet and pharmacy services. Businesses expand. Pets.com, had they been successful, would have likely done the same thing. To use your Amazon example: obviously most of Amazon's revenue is no longer generated by selling books.


Rationalize all you want but Amazon's retail arm still turns a loss, minus platform ads. Bezos realized early on that the bread and butter of Amazon is AWS (and Amazon software in general), and that's what's keeping the retail arm stable.

From a revenue perspective though, your argument is prudent. Maybe targeting books rather than pets helped it lessen its relative burn rate.


That whole internet bubble, and where's the internet now, pfft


Timing is everything in investing. You can get it right in principle but wrong timingwise and still end up losing your shirt.


I develop a few interesting features using OpenAI in an existing product.

After a wow-effects of the first days, few users are still using it. If I talk to users they say features are awesome but they are not actually using it. Really funny.

Do you have experience of OpenAI API-based product that has high retention ?


> Do you have experience of OpenAI API-based product that has high retention?

perplexity.ai comes to mind. I introduced it to my students at uni a few weeks ago and they now use it extensively over Bard/ChatGPT. There are others too, like copy-writing tools (jasper.ai, copy.ai) who seem to be doing well. Besides, from experience, GitHub Copilot is so good, I can barely program without it.


Yes I developed a SaaS and the functionality is solely based on output from OpenAI. We have several thousand paying users already with ARR above $1.5M


Making $900k in the first 3 week it means monetize on the wow-effects.

Is it https://prompthero.com/ ?


Don't know why they're pretending it's not, it's literally in their Twitter profile: https://news.ycombinator.com/item?id=36197024


I've got some crazy news for you: you can actually post other people's tweets on HN! Crazy, right?

By the way the OP of the Twitter thread commented in that thread you linked. He has the same HN username as Twitter.


My apologies, I see the OP's comment you're referring to.


No it is not Prompt Hero (that isn't even a SaaS is it?)


May I ask what your profit margins are? Why can't someone just use ChatGPT instead?


Around 85% profit. It wouldn't be possible to get the same result from ChatGPT without a ton of prompt engineering and manually retrieving data and inserting that into your prompts. Plus, we don't just do a single API call and call it a day. A single button press in our app can set off a process which calls the OpenAI API over a dozen times plus retrieves data from several other APIs to help with the prompting


out of curiosity, are you using openai's base models or are you fine tuning them? and what tech stack is your app using?


NextJS, CF Workers, CF Durable Objects

We use the base models but once fine tuning becomes available on 3.5/4 we will be looking into that


Makes sense! Thanks for your answer :)


What's the service?


Big mistake imo to hop on trendy grifts. Please focus on the founders as before.


They will still do that, but within the context of having an application that takes advantage of recent developments in AI. Whether that's smart or not is an open question, time will tell. But, given that YC is a seed level investor (with some follow on capacity) it makes sense because the 'downstream' VCs are handing out money like candy to start-ups that can plausibly claim to be AI startups. The idea is that some of these will go on to corner their respective markets and that by getting in early you guarantee yourself a place at the table later on.

Whether that is smart is also an open question.


When you say handing out money like candy, it makes me think I should try to apply to something. I am spending most of my time doing low-paying contracts that integrate generative AI for other companies or startups. And then I use that money to build out my own project for a few weeks before I run out of money and have to get another contract.

If it's actually easy to get money? How would I find a VC who would be impressed by my actual technical ability as opposed to something like marketing or networking prowess? I tend to not spend a lot of time on making slick web pages and promoting myself but focus on implementing features.

Right now my own thing is a ChatGPT plugin that generates web pages including appropriate images using a template and stable diffusion. I am working on building in a CRUD system that GPT-4 will have concise docs for, as well as a user login system and an API proxy and secrets management. This way the system should be able to create fully functional websites.

The previous version integrates GPT-4 with fly.io VMs so that it directly executes commands in a loop to accomplish tasks.

But the plugin is on hold to work for some other startup so I can make rent and stuff. I am building out a simple generative fill inpainting/outpainting editor inspired by Playground.ai. Previously for this other startup I created a Dreambooth tool and a live editing scribble image generation thing with Controlnet.

And before that I was working on a Slack bot where you can ask it questions about PDF or other documents, and another that used GPT to write SQL for querying a specific table and optionally creating charts on the fly.

I guess I am just suspecting that maybe working on all of these other startups to pay the bills isn't necessary if I ask the right person for money. Which I never really seriously considered as an option.


You have to be a bit of a charlatan to really get the money. If you’re not willing to do that then maybe it’s not for you.


Well.. I guess I have had the impression that people who get investments or grants are often better at convincing people to give them money than actually solving problems.

But can you elaborate on why you put it like that, "a bit of a charlatan"?


You have to be willing to tell them what they want to hear and hit the right notes, even if you know it’s not necessarily true or accurate.


Not necessarily. Not all founders and not all VCs are cut from the same cloth.


It's easy, but you have to wonder if it is really what you want. Your chances of seeing it through to the end are probably the same as every other start-up and taking VC money comes with a ton of strings attached.

More about this:

https://jacquesmattheij.com/three-roads-to-the-top-of-the-mo...


Thanks for the link to the article. It's a great article. My interpretation of your response is that you would suggest I do not try to get any investment and rather continue to do the contracts, but otherwise following the specific advice on the "third road" from the article.

I think I have been ineffectually following something like the "third road" for ten years already. Although in my case maybe instead of a road up a stately mountain, it's more like getting lost in a ghetto, because I do not seem to progress beyond the low-paying contracts or get any significant interest in my startup attempts. Probably I have just gotten used to them.


It depends on a lot of factors. Personality, life goals, circumstances. I took on investment for a start-up at some point and ended up paying it all back because I realized I was headed straight for a burn out with all this pressure on me and it helped me a lot. And - controversially! - I ended up as a small time investor and LP at a bunch of funds and I do a lot of tech DD, which is still in the service of the VC world. It was quite a series of coincidences that led to that but the end result is that I'm pretty happy with how things turned out, even though several attempts to make this company independent of me have failed and I'm pretty much reconciled to running it at a lower level of ambition.


Unless VCs hope to get their profits from the "bigger fool"?


I'm curious whether AI aligns with startups. IMHO it does not well. It aligns with academia where time flow differently.

Business then takes what an academia produces -- i.e. papers / methods / algorithms and implements it to generate value for customers.

It doesn't align with startups as well in case if you have a method and require a lot of compute to produce a model -- you need capital for compute and that's not a typical garage startup.

I'm really curious whether there is any opportunity left for YC-sized startups in AI.

I lot of really cool developments happen on github -- but again that's open source, not ventures to generate money.


I thought YC backed individual founders and ideas rather than general technologies or sectors.


YC has a long tradition of soliciting applications in particular areas—see the links below. Of course, founders with other ideas are entirely welcome to apply.

Startup Ideas We'd Like to Fund - https://news.ycombinator.com/item?id=250704 - July 2008 (380 comments)

A New Experiment: The RFS - https://news.ycombinator.com/item?id=766703 - Aug 2009 (57 comments)

YCRFS 1: The Future of Journalism - https://news.ycombinator.com/item?id=766796 - Aug 2009 (56 comments)

YCRFS 2: New Paths Through Product Space - https://news.ycombinator.com/item?id=768425 - Aug 2009 (21 comments)

YCRFS 3: Things Built on Twitter - https://news.ycombinator.com/item?id=904193 - Oct 2009 (67 comments)

YCRFS 4: New Ways to Use Live Video - https://news.ycombinator.com/item?id=904195 - Oct 2009 (19 comments)

RFS 5: Development on Handhelds - https://news.ycombinator.com/item?id=950781 - Nov 2009 (44 comments)

YCRFS 6: iPad Applications - https://news.ycombinator.com/item?id=1160140 - March 2010 (106 comments)

YCRFS 7: Applications of Facebook Instant Personalization - https://news.ycombinator.com/item?id=1638004 - Aug 2010 (29 comments)

Request for Startups: Kill Hollywood. - https://news.ycombinator.com/item?id=3491542 - Jan 2012 (664 comments)

New Requests for Startups - https://news.ycombinator.com/item?id=8308666 - Sept 2014 (390 comments)

Requests for Startups Refreshed - https://news.ycombinator.com/item?id=12592619 - Sept 2016 (124 comments)

Request for Startups: Water - https://news.ycombinator.com/item?id=13311257 - Jan 2017 (224 comments)


Thanks for the links, some of product ideas in the 08-10 threads make for interesting reads.


A lot of these links in the original submissions are broken. Would be nice if they could be fixed or archived. I.e:

https://www.ycombinator.com/rfs5.html


Ok, I've pinned archive links to the top of those threads.


Hi Dang, how about an archive mirror link next to flag for links older than, say, 30d? https://news.ycombinator.com/item?id=21766547


How can it be automated in a reliable way? Archive links don't necessarily take you to a readable copy of the article. Someone has to check that, no?


Not sure if the Wayback Machine supports API-based on-demand archival, but they do partner with firms like Cloudflare to auto archive websites: https://archive.is/O9SAO

There's availability API but it won't tell us if the archived webpage does indeed display content (say, if it was behind a paywall or cookie banner): https://archive.org/help/wayback_api.php

archive.is, which breaks past some of the paywalls, apparently (haven't tried it myself) supports the Memento protocol: https://github.com/palewire/archiveis/blob/master/archiveis/...


Thanks, that's interesting!

> There's availability API but it won't tell us if the archived webpage does indeed display content (say, if it was behind a paywall or cookie banner

Right - that's the problem. It takes a human to tell if the archive link is worth anything.

I do want to do something better with this someday (someyear?) - most likely a way for community members to submit and vet archive links, and then (if so) an "archive" button in the detail bar to point to it.


Thanks!


It’s common during a bubble to lose fundamental principles.


Anyone had any interview offers?


the new generation LLM are just better Siri/Alexa. why such hype about them


> companies building AI developer tools.

There it is, that's where you want to be - selling shovels during the gold rush.


The only one selling shovels is nvidia.


They are selling wood and metal to make shovels, which is substrate.

There are shovel sellers which are making tools.

Then there are application makers that end users see that are made with shovels.


Isn’t this the same logic the crypto bulls said during their mania?


"Eschew flamebait. Avoid generic tangents."

https://news.ycombinator.com/newsguidelines.html

We detached this subthread from https://news.ycombinator.com/item?id=36734710.


The difference is crypto is useless and AI is useful.

That’s a pretty important difference.


Please don't take HN threads further into repetitive flamewar hell. Nothing interesting is going to come out of repeating this argument for the n-thousandth time. Worse, when discussions cease to be interesting, they turn nasty.

https://news.ycombinator.com/newsguidelines.html


> The difference is crypto is useless and AI is useful.

Is that why Stripe, TicketMaster, Moneygram, Checkout.com, TransferGo, VISA, etc and many other companies continue to use it?

> That’s a pretty important difference.

Nope. The important difference is that crypto has energy efficient alternative methods of consensus available today such as proof-of-stake, byzantine fault tolerance (BFT), etc and even a former PoW blockchain (Ethereum) cut their emissions by 99.99% [0] by switching to a proof-of-stake consensus method. For crypto it is possible to switch to greener methods.

AI (Deep Learning) continues to destroy the planet since it has no efficient alternative methods available for training, fine-tuning and inference. This means their methods waste tons of resources, water and electricity. [1] [2]

It is a serious issue in the field after decades of its existence all to produce unexplainable black-box AI models which cannot reason and have to be retrained, fine-tuned again once it gets confused by a single pixel or regurgitates nonsense output.

The AI hype is another grift that the VCs are taking advantage of just like they did with thousands of crypto companies. The reality is, 90% of these AI / crypto companies will fail and the remaining 10% will still be around that are useful.

[0] https://consensys.net/blog/press-release/ethereum-blockchain...

[1] https://gizmodo.com/chatgpt-ai-water-185000-gallons-training...

[2] https://www.independent.co.uk/tech/chatgpt-data-centre-water...


> Is that why Stripe, TicketMaster, Moneygram, Checkout.com, TransferGo, VISA, etc and many other companies continue to use it?

“Use” is inaccurate. Accept and immediately sell. It persists in being a pot of unsophisticated money, so yes, it will be accepted.


> “Use” is inaccurate.

Nope. Inaccurate to you. They would NOT use it if it was "useless".

The truth is, they continue to use crypto in their services.


> would NOT use it if it was "useless"

Of course they would. It’s useless, not worthless. If it were useful they’d actually use it.


> If it were useful they’d actually use it.

It is being used and businesses are using it in their products today.

Companies like Moneygram would not use Stellar if it was useless for them and their customers and would instead announce that they are shutting it down after trialling it. They seem to have seen utility in it for their use-case. The same can be said for the other companies I mentioned.

Again, they are using it and that is the truth.


> Companies like Moneygram would not use Stellar if it was useless for them

Useless, not worthless. I happen to be familiar with that case. It’s great for sales. Useless for performance. (It might become useful one day.) Hence, marginal-enough adoption and a lot of marketing.

A couple years ago I counselled a friend in banking to do something with Ripple. It helped him win deals with crypto firms, and likely helped their wealth managers, too. Worthful. But useless.


> Useless, not worthless. I happen to be familiar with that case. It’s great for sales. Useless for performance. (It might become useful one day.) Hence, marginal-enough adoption and a lot of marketing.

Nope. Useful today for people using Moneygram worldwide (In the example I gave). Moneygram used Stellar for their own use-case and its customers use it today for that same utility, but this time with no bank accounts for the end user.

> A couple years ago I counselled a friend in banking to do something with Ripple. It helped him win deals with crypto firms, and likely helped their wealth managers, too. Worthful. But useless.

Now with Stellar today, it is used by many businesses like Moneygram, and aid programmes [0] are trailing the network [1] for the same use-case. Again, it would not be used by such organizations or even trailed or piloted in the first place if it was 'Useless for performance.' or had no utility.

[0] https://stellar.org/aidassist

[1] https://stellar.org/press-releases/unhcr-launches-pilot-cash...


> but this time with no bank accounts for the end user

Moneygram originated with cash or money order in, cash or money order out. No bank account needed.

> with Stellar today, it is used by many businesses like Moneygram, and aid programmes

You’re linking to a pilot. As it happens, I also worked with UNDP years ago. We similarly launched pilots to try and find uses for novel technologies. (Mine was in mobility data from cell phones.)

That’s what’s happening here. They are looking for a use. That should be encouraged. Looking for something doesn’t mean it is found.


> Moneygram originated with cash or money order in, cash or money order out. No bank account needed.

Even better, meaning that the smart decision was to add another rails (crypto) which after choosing something like Stellar and trailing it, they found it was useful for them and their customers, to continue using it. Otherwise it would have been scrapped years ago.

> That’s what’s happening here. They are looking for a use. That should be encouraged. Looking for something doesn’t mean it is found.

Indeed, it is a pilot which means more interest in the use-case which was found AFTER Moneygram trialled it and are using it right now today.

The same is said about the other companies I mentioned. Far from the first broad claim of it all being 'useless' which we all know that is blatantly false.


> they found it was useful for them and their customers, to continue using it. Otherwise it would have been scrapped years ago

You’re not seeing how something performative can be both profitable and useless?

Maybe it's semantic. I'm not saying crypto is useless as a moral judgement. The bumblebee keychain on my backpack is functionally useless. But I like it, so it's worth something to me. But it has never carried a key, it never well, and its artistic value is objectively passable as best. It's useless but with worth.

If we say MoneyGram's crypto integrations are useful as marketing, I'll agree. But again, as someone familiar with that particular example more than most, if those customers would have given business for money stained with mayo, MoneyGram would be slathering it on.


> You’re not seeing how something performative can be both profitable and useless?

Moneygram and their customers with other organizations using the Stellar network is not performative. It exists and works today, right now and it was that 'something' that was used, tried and tested first before making a decision and it turns out that the use-case was valid for them (and many others).

Henceforth, it is not 'useless' going by the OP's original broad claim.

> Maybe it's semantic. I'm not saying crypto is useless as a moral judgement. The bumblebee keychain on my backpack is functionally useless...

Again, the main use of the Stellar network was serious enough for MoneyGram which out of the thousands of other alternatives used to get on another rails (crypto) for their worldwide customers after assessing that use-case.

But you've only made it clear that to you it is "useless" and no user or business would even bother to use it, if it was found to be as such in the first place.


[flagged]


> 90% of AI companies will thrive.

I would expect the failure rate of AI startups to be ~ equal to the failure rate of all start-ups. AI isn't something magical that will cause your startup to have better chances in the marketplace, all of the other caveats still apply (timing, luck, execution).

So lets say between 1 and 10% of AI companies will survive and between 1 and 3% will be home-runs.


> Difference is that 100% of the Crypto companies will fail.

You do realize what you just said is complete absolutist nonsense?

> and 90% of AI companies will thrive.

I think you know that this is also nonsense and you have really confirmed that we are indeed in another AI bubble. A hype cycle reinforced by VC capital inflating expectations once again.

The reality is this: 10% of these AI companies will thrive with the remaining 90% shutting down.

The major gatekeepers (OpenAI, Google, Meta) who own the AI models and the heavily VC backed ones are part of that 10%. The 90% of late comers and pre-revenue copycats are climbing an uphill battle which will end in a merger or mostly them end up shutting down as their credits and VC capital being depleted after generating little to no revenue for years.


[flagged]


So, business as usual then?


I can't wait to see this bubble bursting spectacularly.


A great opportunity to sell then during the rush in capital from the VCs throwing money at any company filled with AI buzzwords before it all collapses.


I have been working on the same idea since the 90's. I find humorous that I could describe the idea I have been working on, using the buzzwords of every funding excitement of wave.


Early? The sector is quite mature by now, they’re basically late to the game.


Early by YC schedule

>early interview slots for the YC W24 batch

The general deadline for W24 is September 11

But this early round for AI companies is available until July 17


YC.org literally backed OpenAI in 2015, Cruise a year before that, and ScaleAI a year after that.

Per YC's publicly available records, the earliest they funded a ML/AI business was in 2008. They aren't late, but this AI specific Early Decision may be in reaction to funds like AIGrant.org, SouthParkCommons, Neo et al.

https://www.ycombinator.com/companies/?tags=AI&tags=Machine%...


Exactly, the chances of getting funding for AI companies are up considerably and if YC doesn't fund them someone else will. To get a seat at the table they need to move faster compared to their normal schedule.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: