title
stringlengths 12
112
| published
stringlengths 19
23
| url
stringlengths 28
28
| video_id
stringlengths 11
11
| channel_id
stringclasses 5
values | id
stringlengths 16
31
| text
stringlengths 0
596
| start
float64 0
37.8k
| end
float64 2.18
37.8k
|
|---|---|---|---|---|---|---|---|---|
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t197.0
|
Because as well as knowing where everything is,
| 197 | 203 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t200.0
|
this person also speaks the same way that we do.
| 200 | 205 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t203.0
|
We can ask them a question like,
| 203 | 210 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t205.0
|
okay, guide, where are those marble-like things?
| 205 | 215 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t210.0
|
And with that, they will hopefully be able to guide you in the right direction.
| 210 | 218 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t215.0
|
And maybe they'll be able to guide you right to that product
| 215 | 223 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t218.0
|
or at least in this area that they are found in.
| 218 | 227 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t223.0
|
So that's the difference between traditional search
| 223 | 230 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t227.0
|
and a question and answering search.
| 227 | 232 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t230.0
|
Now, don't get me wrong.
| 230 | 235 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t232.0
|
There are places where you do want to keep traditional search.
| 232 | 240 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t235.0
|
But particularly for unstructured text data or as we saw earlier,
| 235 | 245 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t240.0
|
video and audio data, this sort of Q&A approach can be really powerful.
| 240 | 250 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t245.0
|
So that leads us on to the second question or first question I asked,
| 245 | 253 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t250.0
|
which is how does Google do that?
| 250 | 256 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t253.0
|
Now, Google is pretty complex.
| 253 | 261 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t256.0
|
But at the core of what we saw there was something called
| 256 | 264 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t261.0
|
Open Domain Question and Answering or ODQA.
| 261 | 273 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t264.0
|
Now, ODQA is a set of language models and technologies
| 264 | 278 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t273.0
|
all piled together into a open domain question answering pipeline.
| 273 | 285 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t278.0
|
Now, that pipeline at its simplest will look something like this.
| 278 | 290 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t285.0
|
So we have our, let's say at the top, we have our question.
| 285 | 295 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t290.0
|
That question is going to come down here and it's going to hit
| 290 | 302 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t295.0
|
what is a retrieval model, which is what we will train or fine tune in this video.
| 295 | 310 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t302.0
|
Now, this retrieval model, that will handle taking our query
| 302 | 313 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t310.0
|
and converting it into a vector.
| 310 | 318 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t313.0
|
Now, we convert it into a vector because then we compare it to other chunks of text
| 313 | 325 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t318.0
|
and we can encode the semantics and meaning behind that text
| 318 | 327 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t325.0
|
rather than just the keywords.
| 325 | 330 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t327.0
|
So that's why we can search for concepts and meaning
| 327 | 333 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t330.0
|
rather than just keywords, as I mentioned earlier.
| 330 | 336 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t333.0
|
So we have this retrieval model.
| 333 | 344 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t336.0
|
Now, the retrieval model creates a vector, but then we need something that allows us to search.
| 336 | 348 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t344.0
|
So we need other vectors to compare to.
| 344 | 350 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t348.0
|
Now, where do we store those vectors?
| 348 | 352 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t350.0
|
Well, we have a vector database.
| 350 | 357 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t352.0
|
Now, a vector database, let's bring that over here.
| 352 | 364 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t357.0
|
This is going to contain context vectors.
| 357 | 367 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t364.0
|
So you remember earlier on with those Google searches, we had the question,
| 364 | 369 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t367.0
|
we had the context and the answer.
| 367 | 371 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t369.0
|
This is where that is relevant.
| 369 | 375 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t371.0
|
So in here, we have loads of context.
| 371 | 380 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t375.0
|
So just put C, but they've all been converted into vectors
| 375 | 383 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t380.0
|
using the same model we have up here, the retrieval model.
| 380 | 387 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t383.0
|
We just did it before we started searching.
| 383 | 392 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t387.0
|
So we index those contexts into our vector database.
| 387 | 397 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t392.0
|
Now, at search time, we convert our question into a vector.
| 392 | 403 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t397.0
|
That comes down into the database and the vector database will compare
| 397 | 406 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t403.0
|
that question vector to all of these context vectors
| 403 | 408 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t406.0
|
and it will return the ones that are most similar.
| 406 | 412 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t408.0
|
So maybe we want to return the top five most similar contexts.
| 408 | 418 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t412.0
|
At this point, if we are just using a retrieval model and a vector database,
| 412 | 420 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t418.0
|
we can return those.
| 418 | 424 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t420.0
|
We had our question and we can return these contexts to the user.
| 420 | 428 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t424.0
|
So that would be like in our earlier examples, we ask a question
| 424 | 433 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t428.0
|
and Google returns the page or it just returns a paragraph to you
| 428 | 437 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t433.0
|
rather than highlighting the specific answer.
| 433 | 442 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t437.0
|
And these are the components that we're going to cover today.
| 437 | 448 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t442.0
|
And in a future video and article, what we are also going to include
| 442 | 449 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t448.0
|
is a reader model.
| 448 | 454 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t449.0
|
So a reader model is added onto this Open Domain Q&A stack
| 449 | 457 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t454.0
|
or it's the last component of the stack.
| 454 | 464 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t457.0
|
And what this does is it takes each of your context vectors
| 457 | 468 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t464.0
|
and it reads them so it has a look at the context vectors
| 464 | 473 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t468.0
|
and we have this long text here and it says,
| 468 | 481 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t473.0
|
okay, given the question, which we also feed into our reader model,
| 473 | 487 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t481.0
|
I think the answer to that question is right here.
| 481 | 496 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t487.0
|
So it allows us to extract a very specific answer from a longer chunk of text, the context.
| 487 | 504 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t496.0
|
So that's the Open Domain Q&A structure or pipeline.
| 496 | 509 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t504.0
|
And now I think we should move on to actually fine tuning the retrieval component of that pipeline.
| 504 | 513 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t509.0
|
The first thing we need to think about when fine tuning our retrieval model
| 509 | 517 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t513.0
|
is whether or not it actually needs to be fine tuned
| 513 | 523 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t517.0
|
because there are retrieval models out there already that we can just download and use.
| 517 | 529 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t523.0
|
And there's this really good concept that I saw from in one of Nils Reimer's YouTube videos
| 523 | 536 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t529.0
|
where he's going to talk and he talks about the long tail of semantic relatedness.
| 529 | 543 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t536.0
|
And the basic gist of it is that you have common knowledge that everyone,
| 536 | 550 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t543.0
|
pretty much everyone knows about and you have a lot of data and benchmarks in that area.
| 543 | 553 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t550.0
|
So that would be our cat versus dog example up here.
| 550 | 557 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t553.0
|
So imagine you're on the street, you walk up to a stranger, you ask them a question,
| 553 | 560 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t557.0
|
what's the difference between cat and dog? They're probably going to know the answer.
| 557 | 564 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t560.0
|
Everyone knows that. Then you get more specific.
| 560 | 568 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t564.0
|
So you ask them, OK, what's the difference between C and Java?
| 564 | 573 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t568.0
|
Some people know, some people will not. And then we get even more specific.
| 568 | 576 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t573.0
|
PyTorch versus TensorFlow. And then we get more specific.
| 573 | 581 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t576.0
|
Roberta versus Dberta and then TSA versus Mirabert.
| 576 | 587 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t581.0
|
As we get more specific, less and less people know what you're talking about.
| 581 | 592 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t587.0
|
And with that, there are less data sets and there are less benchmarks.
| 587 | 597 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t592.0
|
But at the same time, that's where most of the interesting use cases exist.
| 592 | 605 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t597.0
|
Now, whether your use case exists within the common knowledge area or within the long tail
| 597 | 613 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t605.0
|
is really how you can hazard a guess at whether you need to fine tune a retriever model or not.
| 605 | 617 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t613.0
|
So if we just modify that chart a little bit, we get this.
| 613 | 621 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t617.0
|
So we have the same thing, common knowledge on the y-axis.
| 617 | 624 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t621.0
|
And what we have on the right, I've just renamed it.
| 621 | 629 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t624.0
|
So the ease of finding a model and or data.
| 624 | 637 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t629.0
|
OK, so based on how niche your use case is, the harder it is going to be to find data for that niche.
| 629 | 644 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t637.0
|
And the less data there is out there, the less likely someone else is already trained or pre-trained a model.
| 637 | 655 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t644.0
|
So most of the pre-trained models out there are trained on a very generic, broad range of concepts.
| 644 | 658 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t655.0
|
They're trained on Wikipedia pages or something like that.
| 655 | 660 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t658.0
|
So that's fine.
| 658 | 664 |
How to build a Q&A AI in Python (Open-domain Question-Answering)
|
2022-01-11 14:00:20 UTC
|
https://youtu.be/w1dMEWm7jBc
|
w1dMEWm7jBc
|
UCv83tO5cePwHMt1952IVVHw
|
w1dMEWm7jBc-t660.0
|
You know, if you compare cats versus dogs or even seed versus Java,
| 660 | 669 |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.