How to build AI products people want | Hacker News Reader

How to build AI products people want

reforge.com

79 pointsnickwritesit37 comments

It's highlighted in this article briefly, particularly the section about integration: One cannot win the AI product arms race without a core solution to a real problem. This is why most "winners" currently, are either the ones with dominant control over access to AI (those who rent out hardware to use expensive AI, access to easy to use AI - basically google and microsoft for both categories lol). The next winners are those who control access to integral data needed for AI to be useful in that niche.

Everyone else is running around with a hammer, desperately looking for a 2x4. Applying it anywhere and everywhere hoping it does something, but I reckon for most of them if they were to do a cost analysis in the next ~5 years they'll have noticed how costly it was and how little it changed for their core solution.

The heavy hitters and potential big winners (outside of the above noted gatekeepers) are still yet to come thankfully.

Once the gold rush has hit the newspapers, don't go digging for gold. Sell shovels instead.

SpaceX has implemented last minute control of rocket launches and they land the booster rocket. Hard to see google and microsoft anywhere close to Tesla, Starlink, Twitter for AI potential even if you deliberately elide to mention SpaceX.

I’m playing with the little models at home on a local GPU. So far I’ve fine tuned one and gotten some of the better ones (WizardVicuna) to work as a complete drop-in for a use case I was testing with OpenAI. The surprising thing to me was, at least for my fairly simplistic requirements (Parsing an API and making decisions) it’s actually responding with less latency than the OpenAI API and the same degree of accuracy.

I think we are rapidly approaching the proliferation of good, local models. 99% of “AI” does not need a fleet of thousands of A100s.

It’s hard to think of anything but AI ideas these days (I happen to write an idea newsletter), but I think some of the older good advice still applies here:

The best ideas come from proximity to a given problem, and the network to distribute (and test) a solution.

Outside of that all you can do is try to validate fast and build fast.

I don’t think it’s just the accuracy problem. In general, if you can figure out a solution to the accuracy problem, the rest of the field will too —- AI is just that hot right now. The stuff that will endure seems to be AI products attached to companies that already have strong relationships with customers.

The tech is just too widely available to build a moat that can stand up to a larger company just yet.

> AI Use Case 3: Autonomous Driving

> And what about those self-driving cars we had such high hopes for? If Ford shutting down their autonomous driving technology is any indication, you’re still going to have to buckle into the driver’s seat for the foreseeable future.

Cruise and Waymo are both operating in major cities with fully autonomous cars driving real consumers around on their own...

I've seen no evidence that autonomous vehicle operations in cities is anything more than a novelty service

You will, once the lack of driver is priced in.

But will the upfront cost of creating the technology offset this for a long while?

May be not for the companies that have pioneered the technology, but for ones that go in their footsteps. As it often happens with innovation.

I think so, because the market is enormous. I can see these (mostly big) companies be patient while they establish market share.

Why wait? Uber succeeded by subsidizing the cost of rides until they had completely crowded out taxis for good

Go sit in one. It’s pretty mind blowing. This arrived way faster than I expected but yes it’s definitely here/quite close.

Self-driving cars haven't really failed. It looks like Cruise and Waymo are about to get permissions to operate 24/7 in SF.

I think the greater issue is that it is difficult to assess time-to-market. You to asses both what is useful, and how long it might take to build. Some things are just gonna need a lot R&D to work. AI-hype makes people underestimate how difficult it is to actually get things to work.

Agreed failed is too strong, but they are doing worse than than their 2016 hype levels would suggest today.

https://www.theverge.com/2016/9/18/12944506/lyft-self-drivin... <this was pretty absurd>

https://docplayer.net/59662295-The-2017-strategy-digital-aut... <Projected 4% of new cars in 2020 would be Level 3+ - it's approximately 0% today>

> As the Head of Product, Conversational Commerce, at Walmart, Aniket led the product market fit journey of Walmart’s AI-powered shopping assistant.

From what I read, Walmart’s Text To Shop feature that he led seems to be pretty buggy.

https://techcrunch.com/2023/01/26/hands-on-with-walmarts-new...

I can't bother with ideas to productize this generation of AIs. But holly molly, imagine what anybody could do with an army of tireless, cheap, highly competent intellectual workers. It will be the end of the world as we know it.

And I feel fine.

“Roughly every decade, technology makes a giant leap that erases the old rules and wipes out our assumptions. The Internet. Mobile. Video. Blockchain.”

You know you’re in the silicon valley echo chamber when you think blockchain is some sea change. So far it’s proven to be of basically no use at all except perhaps convincing your dumbest friends they’re now expert currency traders.

You know you are in the HN echo chamber when we have yet another comment telling us how useless crypto is.

True. Outside niche forums like HN, blockchain isn't even notable enough for people to brand it useless...

Who says it's useless? It's a godsend for grifters and criminals.

Crypto doesn't even hold a candle to the criminal activity that's going on in traditional banking. It's a public ledger after all, that's why more and more chain analysis startups are gaining traction and more criminals are brought to justice.

"The crypto community must be applauded for its focus on identifying and measuring – and mitigating – illicit crypto activity. And the fiat community has much to learn from the crypto community." - https://antimoneylaundering.wtf/f/financial-crime-estimates-...

Notice how you are downvoted without a single counter argument, in fact all the anti-crypto diatribes in this thread have 0 substance and are just toxic 1-liners which bring no benefit or substance to the discussion. They hijacked an AI-Products thread and derailed it from discussing AI-Products to purely anti-crypto flame-bait.

It wasn’t really an anti-crypto diatribe. It’s useful to know where authors are coming from when evaluating what they say, and somebody who puts Blockchain in the same league as mobile phones clearly isn’t thinking clearly. They’re obviously caught up in some echo chamber.

I don’t know, maybe Blockchain and/or crypto will be widely used one day, but it’s been around for quite a while now and so far it’s been nothing other than speculative gambling. It hasn’t really changed life for most people, other than maybe them having won or lost some money on it. It’s not exactly a ubiquitous technology that most people use every day. The internet and mobile phones impact all of our lives daily, even if we somehow opt out of them. NFTs don’t. It’s more akin to the advent of online poker than an iPhone currently, and hardly a once in a decade world changing tech.

And that’s sort of the point with LLMs too, it’s so new we don’t know if it’s going to be mobile devices or Bitcoin yet. I’m inclined to think the former too, but it’ll take time to tell.

You forget about drug trade, tax evasion, bypassing currency controls and many other illegal use cases.

All while assuming it was super private.

Can you provide sources for your claims and make good faith comparisons based on evidence? As I posted below: Crypto doesn't even hold a candle to the criminal activity that's going on in traditional banking. It's a public ledger after all, that's why more and more chain analysis startups are gaining traction and more criminals are brought to justice.

That is true. Criminals thought it was a safe way to pay for things anonymously online but it turned out to be the opposite.

> Can you provide sources for your claims and make good faith comparisons based on evidence?

Pick any illegal use case online at random. 99% probability there's an option to pay via bitcoin or other crypto.

Pick any legal use case at random. 95% probability online, 99.9% probability offline there's no option to pay in bitcoin.

What kind of evidence would you expect to see?

> convincing your dumbest friends they’re now expert currency traders.

Once you remove the geopolitics (that sovereign currency has), forex trading is just technical analysis which anyone can teach themselves

AI-enhanced smile seems to be a product in test here.

Business that operate in the physical world, like retail stores and healthcare clinics, stand to gain the greatest strategic advantage in an AI world because their data is the hardest to generate and access.

It’s a thought, but I’m not sure I agree. Isn’t this like saying that rural general stores have the greatest strategic advantage in internet commerce because their customers are less likely to have internet and it takes longer to deliver there?

It’s a bit of a buffer from disruption, but it’s hardly a strategic advantage. None of the e-commerce giants we know leveraged their position as tiny, remote corner stores to get there.

Arms race implies that there's an endgame for the spew of new tools, rather than acknowledging it for the bubble that it is.

AI is not crypto(currency) but it has all the same hallmarks of fad and bubble. I wouldn't be surprised if in 12-18 months the high wears off and jokers are left holding the bag on unprofitable companies that are quickly getting eclipsed by power players re-establishing market dominance.

You're gonna get trashed for flying in the face of the hype mill groupthink du jour but you aren't wrong. This pattern has been replicating itself in the tech industry since the run-up to the .com implosion. It's always the same. Some researchers come up with novel tech. Some fledgling attempts are made at seeing if this new tech might solve a few real-world problems, in the process generating some column inches in trade publications and academic journals. Critical mass is reached when awareness of this new tech bubbles up to the faceless horde of griftlords that perpetually turn the crank on the tech industry hype mill. Dumb is aggressively separated from money recursively until eventually the 2nd inflection point is reached, wherein the hype mill is no longer able to drown out the growing realization among late adopters that some (perhaps several) aspects of the situation are firmly rooted in bullshit. Then the bubble around that particular tech hype extravaganza implodes and the search for new grist for the hype mill begins anew.

Site somewhat hijacks the back button.

Reply on news.ycombinator.com