Hi, how did you set chattr to background job? I can set it to background job, but it will pop up as a new window in my browser. How to keep it in the viewer pane?
Discussion of {chatter} mentioned the prompt is enriched with ‘path to data files.’ Does this include paths to files stored locally on my laptop? Does the model ‘remember’ the data? Is this a potential security issue?
In an attempt to coach Copilot by using ‘expressive’ names to identify functions (and possibly other data objects), is there a possibility of running into a conflict with generally accepted naming/coding conventions? If so, how should this be resolved?
You can always rename them after the fact. I often start with more verbose names and then dial back to adhere to my normal standards during development.
Live demos are hard, especially with generative AI. But at 37:46 to about 41:00, the problem is the user, not copilot. The suggested completion correctly prescribes "mpg" with quotes, which would have given the expected result given the function that was just generated (which obviously didn't have the embracing operator). The following minutes were also unimpressive, unnecessarily convoluting things and refusing to just add the {{ }} manually. This is exactly how not to use copilot, by refusing to code and instead deleting and regenerating until you get something that seemingly works, but is trash. If the user programs at this level, any tests they generate and are happy with when the test "passes" are similarly harmful. These tests proclaim to test the na.rm functionality, yet all the test values are the same regardless of the status of the flag? Okay. This demo could have gone better.
Howdy! The goal was to try and guide Copilot to give the answer without me doing anything else on my end, and very quickly move through a number of examples. Doing live demos with a non-deterministic tool is definitely one of the harder things I've done, even after giving dozens of webinars. Ultimately when using a GenerativeAI tool, it SHOULD be used to enhance your existing knowledge/skills and not used as a replacement. As far as the tests, yep - very very basic and in some situations not even testing what I wanted, but boilerplate was mostly ok.
I definitely agree that blindly accepting Copilot responses is bad. We want to always validate that the output is what we want, and it's often faster to edit a response than to try and get a full, complete example all at once.
This just shows how frustrating it is to use chatGPT 3.5 when it spits code that is almost good but isn't and actually doesn't work as well and you have to tweak it.
Great overall presentation and intro to using Copilot with R. Anxious to try.
Wonderful content and presentation. Thank you!!
Hi, how did you set chattr to background job? I can set it to background job, but it will pop up as a new window in my browser. How to keep it in the viewer pane?
chattr::chattr_app(as_job = TRUE) where as_job = TRUE will be set elsewhere as the default
Discussion of {chatter} mentioned the prompt is enriched with ‘path to data files.’ Does this include paths to files stored locally on my laptop? Does the model ‘remember’ the data? Is this a potential security issue?
great one, so it has access to my packages and their manuals? Or do i still have to post snipets of examples from those manuals for it to understand?
Great!, chattr requires a paid plan? With my free subscription is not working. Thanks.
In an attempt to coach Copilot by using ‘expressive’ names to identify functions (and possibly other data objects), is there a possibility of running into a conflict with generally accepted naming/coding conventions? If so, how should this be resolved?
You can always rename them after the fact. I often start with more verbose names and then dial back to adhere to my normal standards during development.
Wow... finally some good fucking food. Really excited to see what I can do with it. Other languages had their new toys so fast R felt forgotten :(
Live demos are hard, especially with generative AI. But at 37:46 to about 41:00, the problem is the user, not copilot. The suggested completion correctly prescribes "mpg" with quotes, which would have given the expected result given the function that was just generated (which obviously didn't have the embracing operator). The following minutes were also unimpressive, unnecessarily convoluting things and refusing to just add the {{ }} manually. This is exactly how not to use copilot, by refusing to code and instead deleting and regenerating until you get something that seemingly works, but is trash.
If the user programs at this level, any tests they generate and are happy with when the test "passes" are similarly harmful. These tests proclaim to test the na.rm functionality, yet all the test values are the same regardless of the status of the flag? Okay. This demo could have gone better.
Howdy! The goal was to try and guide Copilot to give the answer without me doing anything else on my end, and very quickly move through a number of examples. Doing live demos with a non-deterministic tool is definitely one of the harder things I've done, even after giving dozens of webinars. Ultimately when using a GenerativeAI tool, it SHOULD be used to enhance your existing knowledge/skills and not used as a replacement. As far as the tests, yep - very very basic and in some situations not even testing what I wanted, but boilerplate was mostly ok.
I definitely agree that blindly accepting Copilot responses is bad. We want to always validate that the output is what we want, and it's often faster to edit a response than to try and get a full, complete example all at once.
This just shows how frustrating it is to use chatGPT 3.5 when it spits code that is almost good but isn't and actually doesn't work as well and you have to tweak it.