4 ms·
> and have access to far more compute than anyone else can offer to have the biggest chance at state of the art results What models do they have to show after
by visarga 3y ago
> and have access to far more compute than anyone else can offer to have the biggest chance at state of the art results
What models do they have to show after 15 years of AI research? AlphaGo, AlphaFold, and that's about it. Search is mediocre, translation the same, not even their OCR APIs are very good.
On pure research they invented the transformer in 2017 (Vaswani) and word embeddings in 2012 (Mikolov). But Microsoft invented residual connections (Kaiming He 2015) which underline transformers and modern CNNs. CNNs and LSTMs were invented longe before, by Yann LeCun (1989) and Jürgen Schmidhuber (1997).
And yes, OpenAI also invented something of their own - the Adam optimiser (Kingma 2015), that trains 99% of the models today.
Hinton invented Dropout (2014) and backpropagation (1986).
- londons_explore 3y agoThey're good at research. They're bad at turning the research into a product. Specifically, most AI based products need to be trained on user data, and Google is too scared to make any real use of much user data in AI models for fear of getting sued into oblivion if, for example, gmail can be tricked into writing something in an email which reveals some other gmail users private data.
- mrkramer 3y ago>They're good at research. They're bad at turning the research into a product. Tbh, Google Lens is one of the best computer vision products in the market. Generally speaking, it's hard to crate a product around something that is completely new.
- tempusalaria 3y agoOpenAI didn’t even exist when Adam was invented! The contact email on the paper is from 2017. OpenAI has made 1 major contribution - the auto regressive decoder. You could argue also their productisation of RL for LLMs has been highly influential even though they strictly didn’t invent it.
- londons_explore 3y agoPlus all the things they haven't yet published... there must be something, because nobody has yet come close to the performance of GPT-4 8 months later, despite many people trying.
- tempusalaria 3y agoThe likely architecture of GPT-4 was invented at Google - MoE. OpenAI poached the team who developed it. OpenAI is ahead on the RL data side. To me that’s the likely biggest advantage they have in GPT-4.
- danenania 3y agoWouldn’t Google be able to catch up easily if that was really OpenAI’s only advantage? Google has a lot more resources to throw at RL. It seems like there must be some other secret sauce for OpenAI to continue maintaining such a huge lead.
- visarga 3y agoI believe they are using much more synthetic data than others. That's the secret sauce besides their larger training set and model. Synthetic data makes the model more consistent.
- washadjeffmad 3y agoFor all of the novel interest in the tech and service side, I can't understand why so many overlook the value of the datasets. I've been promoting them as potential products to license, similar to textbook publishing, and a bit reminiscent of how Google used its search appliances. Automating methods of creation will be an immensely valuable service. I've already introduced methods of instruct formatting within our org, because as the architectures change, what we learn from performing the work to organize and account for our internal knowledge will continuously pay off.
- famouswaffles 3y ago
- logicchains 3y ago>On pure research they invented the transformer in 2017 (Vaswani) Transformers were invented in the early 90s: https://people.idsia.ch/~juergen/fast-weight-programmer-1991-transformer.html https://people.idsia.ch/~juergen/fast-weight-programmer-1991... Google were just the first with the compute to scale them up.
- blovescoffee 3y agofwiw Schmidhuber is famous for claiming to have come up with virtually every architecture that's now useful/important.
- blovescoffee 3y agoGoogle researchers publish leading/important research monthly. They publish impressive work in vision, nlp, and RL. Have you tried reading papers or following research out of google?
- tester756 3y ago>translation the same What? between what languages?