Youtube wasn't the first video streaming service but it was one of the first for the DSL era when people could watch video without lengthy waits.
AI companies repeatedly failed until enough things, specifically data and compute were at enough scale to deliver.
Advancements in battery technology made electric cars practical bucking the trend of decades of failed EV car companies.
So many things - contactless payment, touchscreens, even LCD panels, these were lousy and impractical for decades.
Attempts at mass adoption of handheld computers, now called smartphones, started in the 1980s. Without high speed mobile networks, high density color LCD screens, reliable geolocation, these things were necessary to make the handheld pocket computer something that everybody has.
Even online grocery delivery services, now common place, had its start in the catastrophic collapse of WebVan in the 1990s. Cell phones, the gig economy, mature e-payments, these were all needed.
You always need to look for the context change and how that can untar some tarpits.
Smart people spend time on this problem/solution.
Solutions appear but fall short of expectations.
The technology or more commonly that application of it is stigmatised.
Sometimes the whole field becomes tainted.
The problem/solution complex is declared a “dead end” or “false dawn”.
Interest cools. Nobody invests for a while. The wreckage of the previous cycle rusts away. The craters erode. This takes 20-30 years.
During this period, some very small companies, academics, and individuals continue to guard the flame, but lack funding or new talent to advance.
Go to step 1, invent new buzzwords/framing and repeat.
Ignore much of what was learned during the previous cycle.
[1] https://blog.eutopian.io/the-next-big-thing-go-back-to-the-f...
Then 30 years on Oculus was founded and everyone who'd never used one of the old VR systems was super excited. To be fair, the technology was a step better - much cheaper and more accessible, low motion input latencies, better resolutions. But ultimately it's still not really quite good enough and it seems that the hard reality is it's not going to make its way into mainstream consumer everyday use this time either.
I can't wait for round 3 in 2040 or so.
I'm not sure what use cases you've tried it, but I'm "playing" a bunch of flight simulators, and after getting used using a HP Reverb 2 for all my simming, it's basically impossible to move back to "flat" screens again. It gives you a completely different depth-perception that is as vital when you fly as when you race, so basically any simming is a lot easier and more fun with VR. But again, if you make the plunge into VR simming, it's short of impossible to go back to "normal" afterwards.
> hard reality is it's not going to make its way into mainstream consumer everyday use this time either.
Yeah, simultaneously I agree with this. VR-in-motion (so not sitting still) is still pretty bad, and the setups you need for good performance are pretty expensive, so it's unlikely to break into mainstream unless some breakthrough is being made. With that said, there are niches that are very well served by VR and personally I guess I hope it'll be enough when the mainstream hype dies off.
It's been in the closet for a few years. Beat Saber is fun too, but.. I guess if you're the kind of person who has a sim setup in a dedicated room in your house it's still appealing but for anyone remotely casual it's just not worth it
And then a bit later, there were 3D glasses, ones that synchronized with the high-rate monitor to show each eye its respective right and left frame. The demo for it at the time was Rogue Squadron and I thought the effect was amazing.
I don't want to see sports teams playing on my coffee table. I don't want to see recipes dancing in front of me while I wear glasses in front of a stove, the humidity of boiling water vapor sticking to plastic lenses. I don't want people dicking around with headwear while they're supposed to be driving. I don't want to see contact and bio information hovering above my friends. And I certainly don't want to see ad overlays throughout daily life.
I want to escape life and enter fantasy worlds. I want to be transported. I want to see the Matrix unfold in front of me. That's about as far from AR or XR as I can imagine.
You just have to have a colossal inventory, and a reasonably good algorithm.
People want what they’re providing, companies want to sell ads, and increase people’s tolerance of ads. Until the next platform without ads comes out (note - tiktok didn’t have ads almost at all for the first year)
Just like Pandora which was also really great for a while.
That's not how I remember it at all. The biggest features (that at least made me and my friends use it) was that no matter what player you used, it probably had a "Scrobble to last.fm" features (which sadly, seems Spotify at least removed), and then you'd use Last.fm to find new songs to play via your own player.
I don't think I remember anyone using the Last.fm radios/playlists, but instead just as a data-browser to find new artists/albums/songs, then play those somewhere else.
But this was around 2005 sometime in Sweden, we basically just had Spotify and maybe Grooveshark available for streaming, maybe things typically worked differently elsewhere.
Sustainability and climate are not products or businesses in themselves.
If you're trying to sell sustainabiity, what is your business? What are you selling? If it's advice on how people can be sustainable, or carbon credits or something, sure, those are things that may be tarpits.
But electric vehicles were probably a tarpit idea until they went from being advanced golf carts to being real cars.
Environmentally friendly packaging is doing very well. Etc etc.
Tell that to the funding programs.
I learned this when I was put on a team for a modeling competition in college. You had like 72 hours to solve the problem and write the report. It was really stressful but a LOT of fun.
There were a range of topics that you could choose from. Some were really obvious how to apply mathematics to, and some... weren't.
I was the math talent on the team... but my team members talked me out of going for one of the problems that were easy to apply math to. We instead picked the problem where it was LEAST obvious. And... we ended up winning the competition against a field of 10,000+.
I think that lesson applies in business all over the place. There's actually a lot of good comfortable money to be made in unglamorous industries.
I spent 2024 building an awesome TV series recommendation platform. It worked by matching you to professional critics who shared your tastes, by basically crawling Rotten Tomatoes and getting an LLM to grade the reviews out of ten. The recommendations were awesome, and having a personalized Rotten Tomatoes where you could read about and research the show using reviews by people who felt the same way as you did about stuff was freakin' cool.
However, getting people to actually sign up and use the app without a massive marketing budget was very, very difficult. The stickiness to get people to go back to it is difficult. Asking people to input their preferences in the first place is hard. People also simply didn't believe the recommendations, and wouldn't take chances on shows; the computer can recommend The Detectorists to as many people as it wants, but there's a high number of people who would love the show but will dismiss it looking at the cover image and having a quick read of the synopsis.
The recommendation part isn't super hard, the getting people to use a B2C app is super hard.
Even if people wanted your standalone app, they're not going to sit and enter the kind of rich data a decent recommendation engine needs. It really has to be a tool that gathers data about you as a side-effect of you using it.
This has severely fallen out of fashion since the 2000s, but it used to be not uncommon that when one web app wanted to do actions on your behalf on another web app, it would just take your username and password and log in as you. According to Cory Doctorow (I wasn't there) Facebook did this to MySpace.
For Netflix in particular, logging in from your server would probably trigger anti-account-sharing, but you could avoid that by making the requests you need from the user's app on their device, not from your server.
I think the industry feels like it's illegal now, but I don't think it's actually illegal? since there's no criminal intent. I don't think it's the same, legally, as when a criminal steals your login details and logs in as you. But I'm not a lawyer and this is not legal advice. But my evidence is that there are apps (e.g. POLi) that do this with bank accounts and still don't seem to be in any trouble. Even the banks don't seem to be locking it out as that would hurt the customer's relationship with the bank.
Now, if someone made a "Recommendation-Engine-in-a-Box", where someone who wanted to make a recommendation app for themselves would supply the content and could tweak the algorithm and the design, I could see that being successful in this market :)
I guess SaaS aimed primarily at founders makes it a meta startup? The snake is eating its tail.
Would probably be worth it even if just to have a consistent UI across services.
Isn’t this what Netflix used to be
You need like critical mass of early adopters so that people would see „hey this is useful, maybe I can use it too”.
Hn is a discovery/recommendation site as is Reddit. Amazon makes a lot of margin on theirs and arguable it's part of the major value add for Spotify and Netflix.
Almost everybody looks at food and accommodation reviews and people bring up IMDb and rotten tomatoes when considering whether to watch a movie.
Search engines and llms make decisions on what to surface, those are a kind of recommendation as well.
So although I understand the sentiment, it's not really a great example - there's plenty of successful executions beyond the dreaded "for you recommendations" engagement bait slop on social media feeds. You're using the successful executions dozens of times a day without noticing it.
Nope, HN is just an online forum. I can't tailor what I see on HN to my tastes, and there's a subset of posters who get preferential treatment on the frontpage (YC companies), so nope, HN is not a recommendation site.
> If your idea has been tried before, do your research and understand why it didn’t work. Assume the founders who tried before were very smart, very determined people; what’s different now?
If you can't answer that question, don't try again.
At least from my read of the definition tar pit ideas are not just ideas that have been tried and failed but they are supposed to seem easy. Things like the restaurant discovery example are technically very easy to build but the limiting factor to people enjoying buying from restaurants is not a tech platform.
No other streaming site would have got away with that. Napster was also a bit of a demonstration of how it probably wouldn't work, so it wasn't a low risk strategy.
It probably won’t be different this time unless something has changed. “I’m just that good, I will out execute everyone before me” is probably BS. The people before you were probably not lazy or dumb, it just wasn’t time.
We have known how to do it for a while now. There are just not too many applications where its strengths are more important than its drawbacks.
What's this mean?
I mean, I live rural and even I have fiber now, but that's new.
Every developed country has a set of professional standards for teachers, and teachers who don't live up to those standards are pushed out, sometimes by having their teaching accreditation revoked. In Australia, for instance, there's a set number of hours of 'professional development' that teachers have to do every few years, and if you don't complete them, you lose accreditation and have to find a new career. The professional development activities and courses that meet the requirements are audited by the Department of Education, and have to draw on the latest research in educational psychology: keeping up with the latest research is the entire point of that professional standard.
When I did my teacher training, the first thing we were told in the first lecture was to never cite any research older than ten years, because it would be out of date. Now, if you've trained in the sciences - I was a physicist - you should be troubled with this, because a discipline can't really accumulate knowledge about the world if it throws everything out after ten years. That's why, when I broke the rules and searched through the databases of academic literature going back more than ten years, I saw the same ideas being reinvented under different names in different decades.
So there seems to be a bit of a trend for people to build flashcard-type tools at the moment, probably because someone's seen a paper on spaced repetition. That's nice, but you can't build a business around this. It doesn't matter if all the thought leaders are all in on spaced repetition this year, because next year they'll have moved on to something else, because they need to have something new to talk about. In Australia and the UK at least (I don't know the figures for other countries), half of all teachers leave the profession within five years of joining it, so most of your user base is overenthusiastic twenty-somethings with no life experience (yes, I was one of these) who will do whatever The Research tells them, and the ones who stay long enough to gain leadership positions tend not to grow out of this, so the classroom side of EdTech is basically a bunch of fads, so it's impossible to build a stable business in this space.
If you want to sell software to schools, go and work in a bunch of them, find some obscure administrative problem, and solve that.
1. Spaced Repetition is not a fad. It's the most consistent and reliable way we we know for rote memorization (conditions apply). And it's not a new thing either. It's been around since the late 1800s. It just wasnt practical until the advent of computers and mobile devices. So I'm skeptical that there is another "something else" to move on to that is as impactful as SRS.
2. Not sure what the state of education in Australia these days but speaking strictly from my school days in India (1980s-1990s), something like spaced repetition would have been a godsend for every single student. And I'm 100% sure a vast majority of schools and teachers still havent heard of it.
3. I've been learning German for the past few years from some of the top private institutes in Vienna, Austria and let me tell you that neither the teacher, nor the students have any idea about spaced repetition.
That said, you're probably right about the business-viability of such ventures because of the difficulty of selling to the decision-makers, I just strongly disagree about Spaced Repetition being a "flavor of the year"
> There is no measure of 'best practice'; a lot of these shifts are driven by personal preference.
Again, you're probably right but using the example of SR threw me off because it's the one thing where I think the data is so clear that it's easily justifiable.
We saw elements of this with “new math”, Singapore math, and common core math, each label of fairly similar concepts promising to improve kids’ facility with math. Test scores haven’t leapt though.
Spaced repetition on the other hand is a cross-disciplinary technique that just needs to be introduced and kept there. There's nothing else out there to substitute it with. If the young staff hype it up one year and then it becomes part of the curriculum and then they move on to other fancy edtech things, then there's nothing wrong with it.
> So there seems to be a bit of a trend for people to build flashcard-type tools at the moment, [...] so it's impossible to build a stable business in this space.
I am of the radical idea that lots of things should not be for-profit businesses (doesn't mean that it can't be profitable — just not exorbitantly so), and that economist's mistaken goal of exponential growth expectations is criminally separated from the sigmoid limits to reality.
So, therefore, while I agree that EdTech is a bunch of fads, I think the fact that EdTech is a thing is wrong.
And I agree with your main point that we should be chasing accumulation and refinement of knowledge, and not doing some sort of spring-cleaning every 10 years.
It would have to be funded by adventure capitalists (e.g., retired techies having fun building stuff) for awhile, but it could easily take over once it got traction.
I think this principle generalizes to: If you want to sell software to X, go and work in a bunch of X, find some obscure administrative problem, and solve that.
Although some people seemingly have a talent for selling products into industries they have no idea about. I always assume that means a highly motivated buyer.
So as a founder how can you tell if you are about to jump into a tarpit?
1) do a lot of research on the problem and see what has been done in the space in the past and who is working on the problem now. If you find lots of failure - dig in and try to understand what the core failure modes were. 2) work on something that people will pay you for, even a very ugly early product. Income is a strong validation. 3) reconsider your idea if it requires the incineration of mountains of cash to get people’s attention.
But at the end of the day Tarpit is really a descriptive heuristic that VCs can find to be useful but not absolute.
So it’s exactly like you say: you keep your distance and you look for evidence.
I do think the point of the metaphor is that sometimes a tarpit is just what it is. I.e. there is no value under that shiny surface and you’re only getting stuck staring at it.
YC has a rare opportunity and it squanders it. It is a hub that gathers most problems and approaches to them in each discipline and many many failures on them at least once a quarter and all of that goes down the drain, instead of being published and explored publicly. The energy of bright new founders is not spent re-hashing the old but exploring the new. YC can still evolve into a science hub for things people want with much more impact than it does now. New founders want to protect IP and hold back competition, so publish the failed ideas and approaches - make it a competition. Show the full length and breadth of tarpit zones and any time they may be cracking. This way new energy goes towards better VC returns instead of falling into old cracks. Build a Yelp of things people want that need to be built or solved.
Many typical tarpit ideas (to do apps, habit trackers, note taking etc.) can be great businesses for a couple of people building software together but not have venture-scale outcomes.
I do agree that as soon as you get network effects (recommendations, marketplaces etc.), SOOOO much is tarpit.
The biggest reason I think founders are going for B2C is because they have zero clue about how to network and sell to other businesses. It's easy to set up a shopify account. It's hard to cold call your first prospect. Do you even have any prospects? Do you know how to find them?
The advantage of B2B is that once you figure it out for the first customer, you are on an exponential path to happiness. You can practically cancel your marketing budget at that point. B2C requires an ongoing assault on the dopamine economy. Unless you can get someone on a subscription and program them to forget about it, you're gonna get steamrolled by TikTok & friends.
The bigger the "B" you're trying to sell to, the bigger the desire to say yes. This is a very dangerous path.
If you're DHH, you can say no. If you're responding to RFPs for a living... It becomes complicated. Not impossible of course, but it's a different path altogether.
Yes.
> The bigger the "B" you're trying to sell to, the bigger the desire to say yes.
Absolutely.
> This is a very dangerous path.
You lost me here. What you term "a very dangerous path" I call "a roadmap".
The bigger the customer, the more they can pay. Also, just because it's 'B' and not 'C' doesn't mean there aren't humans involved. You can go a long way (often, all the way) with a simple 1:1 conversation with the most senior person responsible for your project on the customer's side.
It's dangerous because the wishes of the next customer might contradict the wishes of the previous customer ; or worse, the wishes of the next next customer, etc...
Building your roadmap becomes a balancing act of "staying on course", "changing course when you realize there are other opportunities there", "guessing the opportunity costs of everything", etc...
I'm not in a marketing position, but I suspect there is a slight difference between "we're hearing it from customers, and it would be a good idea to work on X, Y" vs "X is item n°2345 in a 5000 items RFP, and if we don't do it we're not getting the contract at all".
> The bigger the customer, the more they can pay.
Also, the more they can "not pay" ;) Meaning that if, for some reason, you can not fully implement all their wishes (you're going to try to avoid that, but it can be hard), they might be in a possible to just pull the plug altogether.
Sure, you might deem they demands unreasonable, but then it comes down to:
- realizing that they might know better than you do (by definition, they've been in the business longer than you do)
- the old adage: "can you stay solvent longer than your customers stay unreasonable ?"
etc...
How do you learn this skill? Any resources or books you recommend?
This is what being a "non-technical" founder is all about - building bridges between people and organizations.
Location estimation (figuring out where you are) based on indoor WiFi / BLE is one example. Compared to 15 years ago, we have (IIRC - I don't work in this space) super-precise timing API from the modem, and there has been work on the reflections issue (the two big problematic things that non-RF people typically miss).
Participate in a transmitter hunt (also called a foxhunt or a t-hunt) where the organizers or the people hiding the transmitter know their stuff. Reflections and multipath can lead you miles away from a transmitter location.
Anyhow, someone asked me if I knew how to do this without consent once; that is, if I knew how to track people in a building without them knowing. This was 8 years ago or so. I had hoped saying "that's not possible" would dissuade them, but instead they just never spoke to me again.
I specifically remember one of them being "facebook for dogs" and another being for restaurant professionals.
Some of these ideas could maybe be done better now that we have genAI but the question might would it work as a standalone app or is it just a feature?
For instance, some animals who survived the tarpit were naturally selected for evolution.
Philosophically, idea of building a startup is a tarpit too. You need lot of perseverance and courage to navigate out of it, survive and eventually win.
I think it would be great if YC turned discussions like this into well edited written articles. I know there’s talk about producing more text content to help startups.
"The restaurant doesn't exist" is an important axiom.
It's why recommendation engines are useless, Netflix has nothing more. Smart users will see TikTok doesn't really have a good recommendation engine, just good content, bite sized so lots can be produced.
> the world seems limitless but for these physical things it's actually fairly limited
This is a really good quote, it also applies to digital.
Anyway, a list of tarpit ideas would be useful. The axiom's are too hard, like software complexity and getting money out of educational institutions.
This. Nobody has invented a Recommendation engines that really works, the ones that seem to work just have lots of content they can throw at you. I really think tiktok would be no less succesful with an engine that simply threw a random video at you. Recommendations is a really hard problem and anyone that solves it had essentially solved AGI
Those features are very common on every educational video. I do agree that this one focus on an audience that is supposed to be too advanced in relation to the video depth of knowledge, so it shouldn't need it. But still it is normally pushed for every video, and the risks of getting the pacing wrong are very asymmetric.