![]() ![]() I think the way that GitHub have done it is to detect an unfinished block of code and the user waiting and not typing, to then take some code from before the cursor, and then to send an API request that asks the LLM to guess the next code tokens based on that context. You'd need to pay for the compute in silicon and electricity or outsource though, inference would need ~16GB of GPU RAM. ![]() I can't see why this wouldn't be possible, it's mostly a matter of data collection, training and plumbing. I would go even a step further, the best user interface in this case will be no interface at all! We should be able to issue commands in natural langauge using our voice, the LLMs can convert audio to text (using something like Whisper), feed it to ChatGPT/Copilot and execute the given code. I work as a consultant and like to use it to explore and validate or rule out theories among the team ad hoc. It's much easier to learn and use for ad hoc analyses python. I agree having Rstudio only without Copilot is not a bad problem to have. I don't want to conflate R with RStudio, but I do think it's safe to say that RStudio is one of the most popular IDEs for R, and thus provides a solid steer to the whole community. ![]() I think we might see a shift to where people think Python is easier to use. But if we can add code suggestions to Python and R falls behind because it's not widely implemented, and when it is, the predictions suck. One of R's strengths vs python is that it's easy to use, what w/ the tidyverse, better documentation/examples, less brittle in terms of R/package versions etc. If code suggestions "become a thing" that changes how we code forever (still to be seen, but possible), it behooves the R community to write a bunch of R code using copilot (and whatever else may come) to improve the prediction accuracy. Enabling copilot through RStudio would be great for the community, but there's another arguably more important strategic reason. ![]()
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |