Effective prompting of large language models (LLMs) involves providing detailed context, guiding the model's thought process, and embracing an iterative approach to refine outputs.
Mind Map
Нажмите, чтобы развернуть
Нажмите, чтобы открыть полную интерактивную карту
I'd like to share you some tips for
prompting launch language models if
you're using the web user interface of a
LM provider hopefully these tips will be
useful to you right away and it turns
out that similar tips are also useful
for if you're ever involved in building
a software application that uses LMS
let's dive in in this video we'll go
through three main tips for prompting
first is be detailed and specific second
is guide the model to think through his
answer and third is EXP experiment and
iterate this starts with be detailed and
specific using the fresh college grad
analogy I would often think about how to
make sure the OM has sufficient context
or sufficient background information to
complete the task so for example if you
were to ask it help me write an email
asking to be assigned to the legal
documents project well given only a
prompt like this and on doesn't really
know how to write a compelling case to
advocate for you to be assigned to that
project but if you give it additional
context such as I'm apply for a job in
the legal documents project we check
legal documents Vamp for experience on
prompting Els to get after text a
professional tone then this gives the LM
more relevant context to write that
email to help you ask to be assigned to
the project further if you can also
describe the desired task in detail so
if you tell it instead of saying help me
write an email if you ask it write a
paragraph of text explaining why my
background makes me a strong candidate
on this project an Africa from candidacy
then this type of prompts would not only
give the own sufficient context but also
tell it quite clearly what you wanted to
do and this is more likely to get you
the result that you want second tip is
to guide the model to think through his
answer so if you were to tell it
brainstorm five names for a new cat toy
it actually could do pretty well but if
say you have in mind you want a rhyming
cat toy name with a relevant emoji this
is what I might try I might tell it
brain Stone five names and tell it step
one come with five joyful words ready to
cats that for each work come with a
rhyming name and finally for each toy
name at a fun relevant emoji and with a
prompt like this you might get result
like this where the LM follows your
instructions to first come with pearl
whisker and so on and then Pearl 12
whisk whisper feline beine with fun
emojis add it to the end so if you
already have in mind a process by which
you think the own could get to the
answer that you want giving a clear
step-by-step instructions to follow
could be quite effective finally there
have been a bunch of articles that I've
seen on social media that say things
like 20 problems that everyone must know
or 17 problems that will help you grow
your career I don't think there a
perfect prompt for everyone instead I
find it more useful to have a process by
which you can write the prompt that
generate the result for you so so when
I'm prompting on myself I will often
experiment and iterate and try something
like my S off say help me rewrite this
and if I don't like the result I might
clarify and I say correct any
grammatical and spelling errors in this
and if it still doesn't give me exactly
result I want I might clarify even
further to say correcly grammatic spring
errors and rewrite in the tone
appropriate for a professional resume so
very frequently the process of prompting
is not about starting off for the right
prompts is about starting off with
something and then seeing if the results
are satisfactory and knowing how to
adjust the prompt to get it closer to
the answer that you want I think of the
process of prompting as like this you
start up of an idea of what you want the
LM to do and you just Express that in a
prompt and then based on the prompt the
LM will give a response and it may may
not be what you want if it is then great
you're done but if it isn't satisfactory
then that initial response wants helps
you shape your idea and modify the
prompt and iterate maybe a few times
before you get to the results that you
want so I think of the prompting process
as when I start off I try to be
reasonably clear and specific but to
save time I'll often start off with a
short prompt that may be as frankly less
specific than this desire but I just
want to get going quickly after you get
a result back if it's not what you want
then think about why the result isn't
the desired outputs and based on that
refine your prompt to clarify your
instructions and keep on repeating until
hopefully you get the El response that
you want one tip I want to share is I've
seen some people overthink the initial
prompt I think it's better to usually
just try something quickly and if it
doesn't give you the result you want
it's fine go ahead and approve it over
time you will not break the Internet by
just accidentally having a slightly
incorrectly worded prompt so go ahead
and try what you want two important
cards first if you're in possession of
Highly confidential information I would
make sure I understand how a large
language model provider does or does not
use or keep that information
confidential before copy pasting highly
confidential information into the web
user interface of an LM and second as we
saw in the last video with the lawyer
they got into trouble submitting court
filings with facts made up by ANM before
you count to L's result it may be worth
double-checking and deciding for
yourself whether or not you can trust
and act on tlm's output but with these
two caveat when prompting I will often
just jump in and try something and see
it not work but then use the initial
result to decide how to refine the
prompt to get a better result and that's
why we say prompting is a highly
iterative process sometimes you have to
try a few things before you get the
result you want so that's it for tips on
prompting I hope that you go to some of
the web user interfaces of the large
language water providers and try out
some of these ideas yourself and in this
course we provide some links some of the
popular om providers um and I hope you
go play with them and have fun with them
that brings us to the end of the main
set of videos for this week um there's
one optional video to follow where I'll
talk a little bit about image generation
or diffusion models so take a look at
that if you want and then look forward
to seeing you back next week where we'll
talk more about how to build projects
using lunch language models look forward
Нажмите на любой текст или временную метку, чтобы перейти к этому моменту видео
Поделиться:
Большинство транскрипций готово менее чем за 5 секунд
Копировать одним кликом125+ языковПоиск по текстуПерейти к временным меткам
Вставьте ссылку на YouTube
Введите ссылку на любое YouTube-видео, чтобы получить полную транскрипцию
Форма извлечения транскрипции
Большинство транскрипций готово менее чем за 5 секунд
Установите расширение для Chrome
Получайте транскрипции прямо на YouTube, не переходя на другие сайты. Установите наше расширение и открывайте текст любого видео в один клик — прямо на странице просмотра.
