GPT-4 Powered GitHub Copilot for Docs Is So Good!

preview_player
Показать описание

Hello everybody I'm Nick and in this video, I will take the brand new GitHub Copilot for Docs feature for a test drive. It is a much-anticipated feature because it can greatly reduce the time developers spend looking into documentation trying to find information about the tech they are using.

Don't forget to comment, like and subscribe :)

Social Media:

#openai #githubcopilotx #chatgpt
Рекомендации по теме
Комментарии
Автор

Definingly will be using it! Even current GPT3 saves me a ton of time searching for the docs, when this thing comes out it would be so much better since it is specifically trained on this. Also, managing knowledge even within a team is a challenge, people typically prefer to ask rather than search. With Copilot I expected more people will be asking it instead of the co-worker. Which to me means less distractions and more learning and engagment from the collogues.

andriisnihyr
Автор

Great video Nick,
I was at Dublin Tech Summit 2023 this morning and one of the talks was on “How Smart is GPT-4 Really” and Adam Cheyer was one of the panel members. And they spoke about how fast, even in the last few months its growing, not using years as time scale but months.

annraoi
Автор

I wasn't surprised it could do a comparison because I know Azure has topics that compare it's services to AWS, just like AWS docs have topics that compare their service to Azure. But in either case, the comparison is probably biased in some way based on the source.

pilotboba
Автор

To use copilot for docs do i need to have enterprise subscription?

pranavbilurkar
Автор

Hm. I believe you do actually need to care about physical partitioning of CosmosDb.
1. You are limited to 10kRUs per physical partition or 50 GB of data (30 GB for Cassandra).
2. Your throughput will be split evenly between those partitions, so you need to carefully consider the partition key.

In your example, every 5 months of 10GB writes/month you should trigger a provisioning of another partition that should take 4-6 hours.
Say your partition key had only 2 values: APPLES and ORANGES and they are now they are mapping to those different partitions. If you now try to load your 5.5kRUs on APPLES, you will hit the actual limit of 3kRUs that should result in throttling, failed requests and the like.

Admittedly, this is a simplified example, but it illustrate a potential issue with throughput splitting and it's omission from the provided answer. Wouldn't blame ChatGPT too much though, Microsoft doesn't do a good job explaining this themselves. This information is buried in the middle of 2 different articles.

tldr: you should consider partitioning before scaling.

Lammot
Автор

I do not think I would use it every day, but if Ihad a couple, lets say for the company, .Net, Algorithms, and Design practices, I would alternate daily between them...

kgnet
Автор

Plan your budget based on AI outputs, what could go wrong?

Cristian-ekxy
Автор

I need this. I am stunned that this tech happened in my life. Please let me see a useful quantum computer, fusion and the return of zepplins.

davidjsutherland
Автор

Can it be thrown on, say, a confluence space and learn it and then reason about it like this?

AvenDonn
Автор

Hi Nick, can you please ask in Azure if Service Fabric is still alive? 🙂

NextProgrammer
Автор

I guess the headache of comparing and researching the cost of an application based on workload with different cloud providers was almost over.

aathifmahir
Автор

Do you plan to make a video on Github Copilot Chat?

diklccu
Автор

Copilot is not free, right? Please correct me if I am wrong.

RuturajPatki
Автор

"And if you're coming from GCP well... I'm sure they have one as well." LOL Burn.

bnm
Автор

I do not think that MS will allow the users to look at the prices of other cloud providers, because no one will use their Azure in many cases 🤣

EarthCitizen
Автор

My company tried to use ChatGPT and train it with our confluence, the result was I would say... mixed :D sometimes it returns responses in foreign languages, and from time to time it was just worthless... I think the reason was that company docs was at first not always the best source and also there was too less information so chat model can be trained well

yogipl
Автор

At 8:43 there is 1250*0.00065$*24*31=612.75$ which is wrong. Its 604.5$. And the other calculation equals 120.9$, not 122.5$. That's really the most frustating part aboout it all. There is no way to tell how accurate a result is.

acbd
Автор

If its anything like Github Copilot Chat (part of x), it will be completely garbage, it’s literally worse than even ChatGPT 3.5 despite Microsoft claiming its GPT4. I have a suspicion that MS silently downgrades to a much simpler model way too often when they claim GPT4 everywhere in their new Copilot products, during their latest build conference they even mentioned using many different large language models, so would take “GPT4” based with a huge grain of salt.

But maybe they actually used GPT4 for this one for docs.

The Cosmos answer is another example, not everything has RU cost of 1:1, depending on how you actually use it some operations cost more than others, it didn’t even mention that.

michaelnjensen
Автор

So basically this is just chatGPT but with a more specific training source instead of the entire web.

Sounds interesting for this scope of usage.

Assgier
Автор

Sorry, no sense in this video. Sorry.

stefanotorelli