Applied AI Course
Notes: https://drive.google.com/file/d/1hY2y1U2vJ94QCgGw2HjJ2-0Eo9ogA7T6/view?usp=sharing
Similar Posts
12 thoughts on “GPT-3: A Quick Overview”
Comments are closed.
Applied AI Course
Notes: https://drive.google.com/file/d/1hY2y1U2vJ94QCgGw2HjJ2-0Eo9ogA7T6/view?usp=sharing
Comments are closed.
Today's language models ingest far more text than a human could read in a #lifetime reveals both the power of brute-force training and the #algorithms inefficiency at learning – @deeplearningai_
What's your comment on this, Sir?
Could you please provide good resources for the pre reqs mentioned in the video sir , so that we can understand this live session ?
For people wanting to see what the gpt – 3 model is capable of .. I would highly suggest this hackathon winning application made possible with the help of the gpt-3 model.
https://play.aidungeon.io/
Is it different from any search engine? Because we train on data from the internet, What's the difference between a search engine's reply and GPT-3 reply in case of "QNA". Please ignore in case if it is a stupid question. I am trying to understand the difference.
Training GPT-3 cost around 4.2M US Dollars for OpenAi. I think that even though the database size won't be a problem for companies like Google but getting an architecture of this level to train your network on will certainly be a huge challenge, even for them.
Thanks for the resources,will be prepared for the live sessions now! Looking forward to live sessions.
Hi Team, please suggest books for statistics, and ML.
Also is “oreilly hands on ML with scikit” a good book to go ?
As per My knowledge, Turing NLG had the highest number of parameters before GPT3 and it didn't get enough attention . Not just the parameters are a matter of high compute but the FLOPS involved in the rigorous decoder in GPT3 are in the order of 10^23 which makes it very computationally infeasible for anyone. I haven't heard of pruning on GPT and other NLG like models. This suggests in the coming future you might expect SOTA NLG models only with higher parameters and beyond the reach of individual researchers. The ongoing hype is just favoring this development of AI (and not democratizing AI without black boxes). Exciting times ahead but with very less control over it.
In which problem we use gpt 3
What I like most the way you explain the concepts in a simple manner. I really like your teaching style.
As far as I know the GPT-3 model requires API key access which is not easy to get ? am I correct ?
Why BERT is having only encoder and gpt have only decoder ?