Blowing ChatGPT out of the water: Google DeepMind's Gemini to be released soon
Demis Hassabis of Google’s DeepMind AI lab is claiming that their Gemini LLM will be more capable than GPT-4, thanks to their experience with reinforcement learning (via their old program, AlphaGo). And it should be built out within a few months.
What do you think — will this change the game or is it just going to be another flop?
My two cents? I don't think Google is capable of bleeding-edge innovation anymore, so it'll be a while before they catch up. ChatGPT will be bested before long, but not by Google.
I think exactly the opposite. I think google/apple/ and some social media giants are quite literally the biggest threat to producing the most powerful LLM's in existence. They have the world's data at their finger tips, ridiculous amounts of capital, and the best software engineers in the world. It's only a matter of time before the giants come out on top.
True true. And you may well be right. But traditionally, times of disruption and change (like the introduction of AI) are when the powerful are most likely to fall to the nimble.
Maybe that's just wishful thinking 😀
I agree. The Gafa have all the data, money, and skilled people to make the bests LLM.
And I guess that google has the biggest dataset ( web, mobile, and people usage!).
Google will be the number 1 in years, but they need to find asap a new way to make money.
True, they are the biggest threat, but not a David vs. Goliath situation. OpenAI is backed by Microsoft and has the money to pay great engineers (saw a SWE role offering ~900k total comp). Not to mention they've captured the inspiration of the world, the culture over there must be buzzing, great cradle of innovation.
Exciting to watch, that's for sure.
I donno... Google is throwing a LOT of money at AI. If OpenAI was able to make GPT-4 with $100MM, imagine what Google could make with all of their resources. I could see this beating OpenAI.
The thing is more money don't always guarantee a better product.
What's the purpose of Bard ?
Is Google trying to flood AI market with multiple models ?
Yeah - throwing money at a wall until they see what sticks would be my guess
It's difficult to predict the future with certainty, but advancements in learning and DeepMind's experience could potentially lead to significant improvements in AI capabilities.
Deepmind have a very good track record of making ground breaking innovations. Plus even if we discount the clear advantage google has over microsoft or openai (data, capital, researchers) Sundar Pichai is going all in on AI. And I believe gpt 4 is bound to be beaten given the sheer amount of progress we are seeing on this field daily, especially by Open source models and research.
Well, by definition, the best AI (generative, that is) is the AI using 1. a very smart LLM and 2. huge human-created data/datasets. This is where Facebook and Google exceel and will most probably overtake OpenAI.
All of the models are using the same technique to be trained though. Its just the matter of data that is being fed to them. I don't think it will change the game though. OpenAI and Google will be the top in the game still.
Interesting post ! Why do you say they they're not capable of innovation anymore?
Seems to me that they're bloated — the best they seem able to do is buy up the innovators who are nimble and fast (which is exactly what they did with DeepMind).
I think Google can still make SOTA models - for example T5 blew GPT3 out of the water when it was released. Also I have to give them a lot of respect as many of their most powerful models are open source.
Oh yeah, I remember when alphago was making headlines. That was pretty cool. It'll be interesting to see how they're able to incorporate it.