ChatGPT / Copilot / Bing with ChatGPT

[Lazybones]

Anyone here played with the latest AI Chat bots?

I played around with ChatGPT3 for a while and was quite impressed but since its knowledge is static it depends on what you want out of it. It also sometimes makes things up confidently.

What I find more interesting is that I sighed up for the bing beta which has ChatGPT 4 and live search.

So far nearly everything I have asked it has been super useful it does a quick search then very nicely summarizes what it learned from the results with references. This lets your easily verify the info if needed or dive deeper into the full page.

It seems actually useful.

I haven't tried  visual studio codes copilot yet but I hear it is quite good as well. Or you can just ask ChatGPT to write you something.

Microsoft seems to be going all in on this and it might actually be the next big thing.

[Mr. Analog]

Honestly Bing is a game changer, Google can hardly find anything anymore with so many sites messing with SEO for rankings. I hate to say it but in five years we might all be Bing-ing stuff.

As for Copilot it's impressive but I have a moral objection about the way it scraped GitHub and Stack Overflow without even attempting to gain permission from the code authors.

Copilot is very useful though the code it generates is only about 75% useful, you still need to know what you are doing to make use of the code it writes.

[Melbosa]

I've been working with ChatGPT a bit, and have done some experimenting with Copilot.  Both have their merits, but I am with @Mr. Analog on that Copilot is 75% accurate. I have asked it for some code, that while written syntax-ily correct, man you would never want to run it cause it wouldn't work or be very damaging.

I want to see in the next 3y where Bing goes... but yeah I kinda agree that we may see a shift from google to bing... if only they could improve the layout of their search results - that is what bugs me the most lol

[Lazybones]

GPT-4 launched today and is a rather large leap in functionality.. Now has vision / image understanding.

https://openai.com/product/gpt-4

https://www.theverge.com/2023/3/14/23638033/openai-gpt-4-chatgpt-multimodal-deep-learning

[Lazybones]

Summary of what is in GPT-4

https://www.youtube.com/watch?v=TxkJMX0KyS0

[Lazybones]

Features coming to office 365 

https://blogs.microsoft.com/blog/2023/03/16/introducing-microsoft-365-copilot-your-copilot-for-work/

[Lazybones]

Even more detailed videos from MS https://www.youtube.com/@Microsoft365/videos covers what it does in each Microsoft 365 app in more detail.

Some of these capabilities are really impressive if they work well.

[Darren Dirt]

...very useful though the code it generates is only about 75% useful, you still need to know what you are doing to make use of the code it writes.

LLMs are great for "here is my HTML table, make it look pretty" or "In Foo what is the same as Bar in Xyz" or even "write some tests and JSDoc for my function".

Not so great for "My code doesn't work. Why?"

TBH also not great for "summarize this book" (or even "give the table of contents") when it has never gathered that info... but very convincely tries to make it all up!  But it does seem helpful at "summarize what are the Top 10 names/techniques in this field of study".


Overall I've treated it as glorified autocomplete (with context of thousands of words instead of a handful) -- a now-unavoidable tool to extend human ability esp. learning.

Like a microwave instead of open flame for cooking. You still need to know how to not misuse the tool... or things might blow up in your face.


PS: most of the time I don't use the "free" web UI, instead the more flexible Playground @ https://platform.openai.com/playground -- one day I plan on writing a simple Node or Go script to take limited actions. Won't name it HAL tho.