Article had another interesting tidbit: "advent of generative AI has made it easier for companies of all sizes to create try-on experiences for their customers and made it harder for us to differentiate our offering.”"
Some other large companies (can't mention names) have also done layoffs targetting genAI recently. I was surprised by this.
There’s a lot to unpack, this is from my perspective doing work in advertising innovations for these customers on social media.
Snap’s biggest problem is it’s a crummy place to work with too many… let’s say personalities that are not commensurate to what they deliver. Same could be said of Amazon, though they have a far higher revenue ad product than Snap ever had, so shows how much those two are related.
To this specific quote: Generative AI meaning image generators are anticipated to replace creative agencies for display ads.
They are not good ads right now. Mondelez and Unilever who are trying this for example have not found a big breakthrough. Part of this is that neither of the two latent diffusion models create sufficiently creative creatives. You need programmers with art backgrounds to master these and the brands have neither of these. They’re super rare generally, you cannot just pull them out of the professional recruiting pool, and they most definitely do not want to work for brands or Snap.
However the appeal is in the race to the bottom costs. Hard to compete with “free.”
These AR experiences still require opinionated creative development which Snap, its agencies and the brands lack. So that’s really what ARES in particular failed.
The product people want is “Ads people like.” That’s expensive to make, which is the opposite of the incentives for creative production in the ecosystem.
That said I have found success creating and supporting interactive, instant streaming games as branded destinations for Meta and TikTok. Try the one I made for Hyundai: https://appmana.com/watch/virtualtestdrive - tap drive and see for yourself. This isn’t meant to be self promotion so much to prove that I dogfood my opinions into a proven, factual reality. You can see more at AppMana.com for huge brands like Nike, VISA, etc.
The AR experiences people make are too “2D” IMO; also they are not that fun. They might hit 15s of engagement on average, and most of that is spent loading, so it’s kind of a fake number compared to that virtual test drive which has instant loading and average engagement measure in minutes at million-visitor scales. Video ads have a median zero watch time and mean of 2.1s so there’s lot of opportunities for “marginally” better. Ultimately it comes down to costs.
It's not really just startups - honestly I think 99% of those startups will fail.
The problem with generative AI as a business is basically what that leaked Google memo said, which was something along the lines of: "OpenAI doesn't really have a moat, but neither do we."
There are a few large companies who are creating these successful models (OpenAI, Microsoft, Google, Meta, perhaps Amazon with their Anthropic investment), but there are tons of other startups that are trying to come up with "individual products around AI", and I think it will be incredibly hard for them to differentiate. That is, for a lot of the "dedicated" early-stage products I've seen, I've thought "I can pretty much do this all with ChatGPT, maybe ChatGPT with plugins - why would I want to download your special app?"
I don't think all of these companies are doomed to fail (for example, I think Harvey AI was smart to focus on a very specific, profitable niche: lawyers have a lot of money to spend on this stuff, and they have specific needs e.g. around compliance, auditability, privacy, etc. that you could see a specific legal-focused AI tool addressing), but I think a lot of them are.
there are tons of other startups that are trying to come up with "individual products around AI", and I think it will be incredibly hard for them to differentiate.
Even more fundamentally, they're putting the cart before the horse. They're making the classic mistake: Taking a tool/technology, and then searching for a problem to solve with it, rather than starting with the problem. I agree a lot of these companies are doomed to fail for the same reason companies founded to "Figure out what to do with blockchain" were doomed to fail. They'll no doubt suck down plenty of funding while failing though.
Meta released a lot of high quality AI research and models (Llama) and tools (PyTorch) as open source. So instead of paying OpenAI for access to their large language model, you can just download a LLM from Meta and use it for free. So in effect, they are flooding the market with free alternatives to paid Microsoft/Google/OpenAI offerings.
I'm not sure this is quite relevant here, although I do love the link.
Meta's open source play I think comes back to regaining public sentiment about them after scandals/metaverse, and also making it difficult for established competitors to have much moat.
Given how much Snap pays their employees it was probably the right call.
Well, which is it? Is it so trivial that anyone can compete, or is it too costly to build it?
I think that's the wrong "either/or" framing in this case. It's not that building competing tech would be trivially cheap, but it's something around which Snap would not likely have a moat or any sort of competitive advantage. It's also in a business area that's pretty unrelated to their core offering.
They tried something, and it failed, and they don't want to say that.
The CEO's exact quote was "Leading in augmented reality means that sometimes we will fail, and I am proud that our team dared to build this business even if we did not succeed." So I'm not sure why you think "they don't want to say that."
And they didn't know these things six months ago?
Given that the initiative was originally announced in March, I'm assuming the whole project was greenlit at least 3-6 months prior to that. And ChatGPT was only just released on Nov 30, 2022, and since then it hardly feels like a week goes by where I'm not mindblown by some new AI announcement. So yeah, I think it's entirely reasonable to think that the landscape when this ARES project was first started is very different than the one we find ourselves in today.
Comments
Article had another interesting tidbit: "advent of generative AI has made it easier for companies of all sizes to create try-on experiences for their customers and made it harder for us to differentiate our offering.”"
Some other large companies (can't mention names) have also done layoffs targetting genAI recently. I was surprised by this.
Is it really startups eating their lunch???
There’s a lot to unpack, this is from my perspective doing work in advertising innovations for these customers on social media.
Snap’s biggest problem is it’s a crummy place to work with too many… let’s say personalities that are not commensurate to what they deliver. Same could be said of Amazon, though they have a far higher revenue ad product than Snap ever had, so shows how much those two are related.
To this specific quote: Generative AI meaning image generators are anticipated to replace creative agencies for display ads.
They are not good ads right now. Mondelez and Unilever who are trying this for example have not found a big breakthrough. Part of this is that neither of the two latent diffusion models create sufficiently creative creatives. You need programmers with art backgrounds to master these and the brands have neither of these. They’re super rare generally, you cannot just pull them out of the professional recruiting pool, and they most definitely do not want to work for brands or Snap.
However the appeal is in the race to the bottom costs. Hard to compete with “free.”
These AR experiences still require opinionated creative development which Snap, its agencies and the brands lack. So that’s really what ARES in particular failed.
The product people want is “Ads people like.” That’s expensive to make, which is the opposite of the incentives for creative production in the ecosystem.
That said I have found success creating and supporting interactive, instant streaming games as branded destinations for Meta and TikTok. Try the one I made for Hyundai: https://appmana.com/watch/virtualtestdrive - tap drive and see for yourself. This isn’t meant to be self promotion so much to prove that I dogfood my opinions into a proven, factual reality. You can see more at AppMana.com for huge brands like Nike, VISA, etc.
The AR experiences people make are too “2D” IMO; also they are not that fun. They might hit 15s of engagement on average, and most of that is spent loading, so it’s kind of a fake number compared to that virtual test drive which has instant loading and average engagement measure in minutes at million-visitor scales. Video ads have a median zero watch time and mean of 2.1s so there’s lot of opportunities for “marginally” better. Ultimately it comes down to costs.
It's not really just startups - honestly I think 99% of those startups will fail.
The problem with generative AI as a business is basically what that leaked Google memo said, which was something along the lines of: "OpenAI doesn't really have a moat, but neither do we."
There are a few large companies who are creating these successful models (OpenAI, Microsoft, Google, Meta, perhaps Amazon with their Anthropic investment), but there are tons of other startups that are trying to come up with "individual products around AI", and I think it will be incredibly hard for them to differentiate. That is, for a lot of the "dedicated" early-stage products I've seen, I've thought "I can pretty much do this all with ChatGPT, maybe ChatGPT with plugins - why would I want to download your special app?"
I don't think all of these companies are doomed to fail (for example, I think Harvey AI was smart to focus on a very specific, profitable niche: lawyers have a lot of money to spend on this stuff, and they have specific needs e.g. around compliance, auditability, privacy, etc. that you could see a specific legal-focused AI tool addressing), but I think a lot of them are.
Even more fundamentally, they're putting the cart before the horse. They're making the classic mistake: Taking a tool/technology, and then searching for a problem to solve with it, rather than starting with the problem. I agree a lot of these companies are doomed to fail for the same reason companies founded to "Figure out what to do with blockchain" were doomed to fail. They'll no doubt suck down plenty of funding while failing though.
Harvey AI! Guess those founders were fans of Suits (the tv series).
Also could be Harvey Birdman: Attorney at Law :)
Meta open source
Exactly. Facebook is brilliantly using open source to neuter any potential new entrants.
Could you please elaborate on this?
Meta released a lot of high quality AI research and models (Llama) and tools (PyTorch) as open source. So instead of paying OpenAI for access to their large language model, you can just download a LLM from Meta and use it for free. So in effect, they are flooding the market with free alternatives to paid Microsoft/Google/OpenAI offerings.
Commoditize their complements: https://www.joelonsoftware.com/2002/06/12/strategy-letter-v/
I'm not sure this is quite relevant here, although I do love the link.
Meta's open source play I think comes back to regaining public sentiment about them after scandals/metaverse, and also making it difficult for established competitors to have much moat.
Given how much Snap pays their employees it was probably the right call.
Probably that Meta would rather lift small players to crush established companies (Snap has 330M+ DAU) than allow them to carve into Meta's profits.
Will be interested to see down the line how any startups nowadays will fare with Meta.
Its just an excuse to avoid admitting that it was a failure. In the next paragraph it says:
Well, which is it? Is it so trivial that anyone can compete, or is it too costly to build it?
I think that's the wrong "either/or" framing in this case. It's not that building competing tech would be trivially cheap, but it's something around which Snap would not likely have a moat or any sort of competitive advantage. It's also in a business area that's pretty unrelated to their core offering.
And they didn't know these things six months ago? No. They tried something, and it failed, and they don't want to say that.
The CEO's exact quote was "Leading in augmented reality means that sometimes we will fail, and I am proud that our team dared to build this business even if we did not succeed." So I'm not sure why you think "they don't want to say that."
Given that the initiative was originally announced in March, I'm assuming the whole project was greenlit at least 3-6 months prior to that. And ChatGPT was only just released on Nov 30, 2022, and since then it hardly feels like a week goes by where I'm not mindblown by some new AI announcement. So yeah, I think it's entirely reasonable to think that the landscape when this ARES project was first started is very different than the one we find ourselves in today.