Fixing Copilot suggestion quality in VS Code extension version 1.243.0?

[ryaninvents]

Select Topic Area

Question

Body

I upgraded my VS Code Copilot extension version to 1.243.0 yesterday, and the usefulness of Copilot took a major hit. It seems to be much more timid now, and has become less useful in several ways:

• Often refusing to suggest completions mid-token (for example: Object.assi begs the completion gn, but it frequently avoids doing so)
• Often bailing on useful suggestions, instead suggesting an // implementation goes here comment or similar
• Occasionally failing to suggest completions for very similar code (e.g. in a switch... case block). This is arguably one of the most useful features of Copilot, as it saves a lot of typing and has a low risk of the model suggesting bad code due to the established pattern.
• Assuming oddly-named or misspelled methods and variables exist, where the previous model would work as expected (e.g. suggesting getPolicys instead of getPolicies, where the getPolicies function exists and the previous model would have suggested it)
• Far more frequent suggestion of misspelled words in comments (e.g. "ancester" instead of "ancestor")
• Inserts redundant end-of-line comments (e.g. whitespace: true, // use whitespace when it would be obvious from context)

I'm sure there are others, but these are top of mind for me after ~6 hours of using the new release. I've downgraded to 1.242.0 in the meantime, and all of these problems have resolved themselves. Does anyone know if there's a new setting I can tweak to get the old suggestion model back in the new release?

[ryaninvents]

I tried version 1.243.0 again after a few days, and did not experience the same problems. I've since upgraded to 1.245.0 and it is working similarly smoothly. Seems like perhaps this was a backend issue.