9 ms·
I think you're missing a fundamental reason: adding AI functionality into products is simply easier. That is, these companies are largely not doing the hard pa
by hn_throwaway_99 4y ago
I think you're missing a fundamental reason: adding AI functionality into products is simply easier.
That is, these companies are largely not doing the hard part, which is creating and training these models in the first place. The examples you gave of of cloud and iPhone have both huge capital barriers to entry, and in iPhones case other phone companies just didn't have the unique design talent combination of Jobs and Ive.
- Vt71fcAqt7 4y agoAdobe in particular, however, has been more twords the forefront of AI research. I'm pretty sure they aren't just using SD here. They might not even be using transformers at all. See https://news.ycombinator.com/item?id=35089661 https://news.ycombinator.com/item?id=35089661
- capableweb 4y agoThey also have the resources to build a huge training set, together with people who willingly upload their art and photos to them, which they can use to make the training set better than publicly available data.
- lelandfe 4y agoOne step further, they already have a huge training set. Stock libraries have the luxury of the hard part already being done: labeling. As of today, that's >313M labeled images they can use with no fear of legal woes: https://stock.adobe.com/search/images?filters%5Bcontent_type%3Aphoto%5D=1&filters%5Bcontent_type%3Aillustration%5D=1&filters%5Bcontent_type%3Azip_vector%5D=1&filters%5Bcontent_type%3Avideo%5D=0&filters%5Bcontent_type%3Atemplate%5D=0&filters%5Bcontent_type%3A3d%5D=0&filters%5Bcontent_type%3Aaudio%5D=0&filters%5Binclude_stock_enterprise%5D=0&filters%5Bis_editorial%5D=0&filters%5Bfree_collection%5D=0&filters%5Bcontent_type%3Aimage%5D=1&order=relevance&safe_search=1&search_type=remove-breadcrumb&search_page=1&get_facets=0 https://stock.adobe.com/search/images?filters%5Bcontent_type... Stable Diffusion was trained on billions of images, of course. But having explored some of LAION-2B, it's clear that Adobe Stock has far better source images and labels.
- mesh 4y agoJust to be really clear what we do and do not train on: Firefly was trained on Adobe Stock images, openly licensed content and public domain content, where copyright has expired. https://firefly.adobe.com/faq https://firefly.adobe.com/faq We do not train on user content. More info here: https://helpx.adobe.com/manage-account/using/machine-learning-faq.html https://helpx.adobe.com/manage-account/using/machine-learnin...
- Bedon292 4y agoIs it also trained on all of the Adobe Stock images which were created with Generative AIs in the first place? I think that could get into a messy spot too. It looks like there are already millions on there, not all of which is actually labeled as generative. If you go to https://stock.adobe.com/ https://stock.adobe.com/ and search for `-generative` you can see it filters out ~3M of them, but then go to the "undiscovered content" filter, and it appears to me like the vast majority of it is generative too.
- boplicity 4y agoThey also know most of their business customers already have GPUs, and often have high-end GPUs, so they're able to tailor solutions to the hardware their customers already have. For example, the speech to text feature in Adobe Premeire runs on local hardware, and is actually pretty good. Hopefully they'll continue to push the potential for locally run models.
- exac 4y agoI don't think they missed it, their second point is that the DX (developer experience) is good.
- usrusr 4y ago> That is, these companies are largely not doing the hard part, Are they the hard parts though? The short time it took from the first waves of public excitment around DALL-E to stable diffusion being the well established baseline looks more like the class quantity of problems that can be reliably solved by shoving enough resources at it. What I consider hard problems are those full of non-obvious dead ends are where no matter how much you invest, you might just dig yourself in deeper.
- mesh 4y ago>That is, these companies are largely not doing the hard part, which is creating and training these models in the first place. fyi, for Adobe Firefly, we are training or models. From the FAQ: "What was the training data for Firefly? Firefly was trained on Adobe Stock images, openly licensed content and public domain content, where copyright has expired." https://firefly.adobe.com/faq https://firefly.adobe.com/faq (I work for Adobe)
- hn_throwaway_99 4y agoThanks for correcting my bad assumption, appreciated.
- samstave 4y agoHa, I didnt even need to read the article to assume this! I instantly thought of how bitchin' their library of images must be. Can you tell us how many images/size of set?
- egypturnash 4y agoI am definitely glad to see attention being paid to ethical sourcing of training images but I am curious: did the people who made all those stock images get paid for their work being used for training? Did they check a box that explicitly said "Adobe can train an AI on my work"? Or is there a little clause lurking in the Adobe Stock agreement that says this can be done without even a single purchase happening?
- anigbrowl 4y agoI'm sure that's a no. When you license a stock image you license it for any use whatsoever. You don't get to complain if it becomes the background to a porn movie or an advert for a product or person you despise. Songs can licensed on a case-by-case basis but images are so plentiful as to be a commodity.
- ghaff 4y agoNot quite. For example, this is one thing Adobe says in their FAQ: Images showing models can't be used in a manner that the models could perceive as offensive. (For example, avoid using images with models on the cover of a steamy romance novel or a book about politics or religion, etc.) There are also a few other more Adobe-specific restrictions.
- samstave 4y agoI think you are both correct But this phrase from GP is pretty darn salient: >>"They are aggressively integrating it into their products which (1) provides a relatively cheap step function upgrade and (2) keeps the barrier high for startups to use AI as their wedge."
- quitit 4y agoAlso missed: All these big tech companies were already invested in AI and using it in their products: it just happens that the latest batch of AI tools are far more impressive than their internal efforts.
- joe_the_user 4y agoThat is, these companies are largely not doing the hard part, which is creating and training these models in the first place. No, there is no real "hard part" to AI current. Training is simply "the expensive part". It seems "the bitter lesson" has gone from reflection to paradigm[1] and with that as paradigm, the primary barrier to entering AI is cash for cpu cycles, other things matter but recipes is relatively simple and relatively available. [1] http://www.incompleteideas.net/IncIdeas/BitterLesson.html http://www.incompleteideas.net/IncIdeas/BitterLesson.html
- zwaps 4y agoI get your point, but please do read the training logs for Meta's OPT, it's some Greek drama I tell ya
- shostack 4y agoWhich checks the box for great timing. All these companies with billions in the bank needing a new thing to prove they still warrant their growth valuation will dump obscene sums into this and cause things to evolve at a staggering pace, much like we've seen with AR and VR.