https://ift.tt/3ea5SDf Its that time of the year and while I do not like predictions, I think there is one which I want to talk about It...
Its that time of the year and while I do not like predictions, I think there is one which I want to talk about
It’s a bit specific so needs some context
From an AI standpoint, 2021 has been the year of large language models like GPT-3
And that has led to near-magical functionality for AI
GPT-3 can translate language, write essays, generate computer code, and more — all with limited to no supervision.
So far so good but as HAI Stanford points out - How Large Language Models Will Transform Science, Society, and AI
- As language models grow, their capabilities change in unexpected ways
- GPT-3’s uses and their downstream effects on the economy are unknown
- Is GPT-3 intelligent, and does it matter?
- Future models won’t be restricted to learning just from language
- Furthermore, some workshop participants also felt future models should be embodied
- Disinformation is a real concern, but several unknowns remain
- Future models won’t merely reflect the data — they will reflect our chosen values
Also, the capabilities of GPT-3 have already been surpassed
DeepMind came out with Gopher, a 280 billion parameter transformer language model and Google has introduced the Generalist Language Model (GLaM) – a trillion weight model
All this us well understood and debated – but, for most part, GPT-3 and other such models are still not accessible to mainstream users
That might change in 2022
But what does this mean for AI applications?
So, let’s think of the analogy of Client server
Think of large language models as the sever and then increasing capacity for the server could see 'thinner clients'
Hence, AI applications could be simpler and would take advantage of 'prebuilt' large language models
I see this happening in 2022 since both the pieces (low code and large language models) are in place
This leads to richer AI apps which are easier to develop along with the drawbacks of large language models(costs, environmental costs, and competitive advantages with a few providers)
It could also lead to a shift in the balance of power i.e. whoever has the resources to build these large language models, could gain a competitive advantage. In any case, we will see a whole new class of applications that we have not seen before driven by large language models
Image source - pixabay
from Featured Blog Posts - Data Science Central https://ift.tt/3EdhiAX
via RiYo Analytics
No comments