Rasa
In this video, Rasa Developer Advocate Rachael will talk about what GPT-3 is, what it can be used to do and some of its drawback and limitations.
– “Rasa Reading Group: GPT3 (Part 1)” https://youtu.be/I8IHla4cXHk
– “Rasa Reading Group: GPT3 (Part 2)” https://www.youtube.com/watch?v=9sL08TCh7T8
– “GPT-3: Careful First Impressions” by Vincent Warmerdam https://blog.rasa.com/gpt-3-careful-first-impressions/
If you want to reach out to us, feel free to leave a message at our community forum; https://forum.rasa.com/.
Gpt3 can code. I would call this more than just "word autocompletion"
Much awaited video from Rasa❤
Excellent explanation of GPT3 that touches possible undesired outcomes such as racist and homophobic output.
Wow this is great. I just saw Mady's talk at pybay yesterday so this is really great!
Hi Rasa, can i use gpt-3 for machine translation by fine tunining it. Help
Hello, I disagree with your analysis about the "intelligence" of the thing (but do not worry I am not either consider it like skynet :P). Yes the task it has been trained on is to predict the next word, but the whole point of this kind of research is to see which internal model of the input the AI will naturally infer to complete its task, what will emerge from this single task it is trained on. And for that it is obviously very good, it "understands" structurally the input at least to some extent. It is very well able to identify what refers to what even in long sequences, meaning relationships, logical repetitions in the input (like what it is doing when used in few shots learning), it has basically learned to learn, at least partially. This is by definition intelligence. The training task nature cannot be used to distinguish intelligence from non intelligence, me and you have never been trained to be intelligent, we are just handling a data feed and a huge part of what our brain tries to optimize is just it's prediction of this data feed. Our intelligence is hugely also a "side effect of our main task" it does not mean we do not have any.
You are also talking about factual knowledge just after and this is not a measure of intelligence either, it's just data. Intelligence is really defined by the internal model of the input data, it's power of generalisation, interpolation, extrapolation etc… Intelligence is more about how we internally do things rather than what we produce. By example if GPT3 were just searching in the training data the exact match of what is inputed, to complete it, it would be the less intelligent way to perform it's task. But most of the time it's not doing that, it is indeed, doing a very complex structural process using the internal model of its "world".
Maybe you wanted to talk about consciousness? for that GPT3 is some kind of a stateless process and is not reflexive at all (it cannot use it's own previous internal states to do anything, it has no access to it as far as I understand) so without memory and reflexivity, yes it cannot be conscious, not even a little.
But it's a great video you made! 🙂 I come in peace I just wanted to add this precision/point of view.
Thanks for the clear explanation! Good vid =]
Our education system need drastic change. Why I never encountered any good teacher like you?
🤔
Glad i found this playlist