[0:00] When you ask artificial
intelligence, like chat GPT a
[0:03] question, like, what is the
best way to slice bread It
[0:07] provides an answer that
seems like a human expert.
[0:11] The AI appears to know about
types of knives and that bread
[0:15] is heated when cooked.
[0:17] Does AI really know these things? How
can it know about so many topics?
[0:23] Can we trust the answers?
[0:25] To answer those questions,
[0:27] let's look at the source of
the information used by tools like ChatGPT.
[0:32] Large language models.
[0:35] Imagine the task of scanning
every word on the internet.
[0:38] That's every blog, website,
research paper, book,
[0:43] newspaper, computer
program, and more.
[0:47] That's the goal of massive
computers that power tools like
[0:51] chat GPT to capture
everything it can on the web.
[0:56] This snapshot of over a
trillion words creates the
[1:00] foundation for large
language models.
[1:03] By analyzing the words in the
model, chat GPT can answer our
[1:07] questions and requests
with human like quality.
[1:11] What sounds like expertise
[1:13] is really a math problem
for artificial intelligence.
[1:16] It works because powerful
computers are trained to look
[1:20] for common patterns
between words and phrases,
[1:23] like tomorrow morning,
[1:26] cup of coffee, and
suppose that you.
[1:30] Large language models
can also detect context.
[1:34] It can tell based on words
that appear together, if you mean
[1:38] bat or bat,
[1:41] every word becomes
a math problem.
[1:43] That uses artificial
intelligence to find what word
[1:46] should come next in a sentence.
[1:49] So when we ask AI, how to
slice bread It doesn't know
[1:53] about bread or knives like
you and me. It only knows the
[1:57] words in the large
language model.
[2:00] Because it was trained on
a trillion words and their
[2:03] context, it can assemble the words
that most often answer that question.
[2:09] Keep in mind,
[2:10] that AI responses are based
on what's published on the
[2:13] internet and like the internet, they
can be biased, misleading, and inaccurate.
[2:21] Evaluation and use them
with skepticism and care.
[2:24] With experimentation
and practice,
[2:27] you can use these tools
to easily access the world's
[2:30] knowledge and find the
answers you need quickly.