4 ms·
I know all press is good press... but there are limits. If it feels like Grammarly does not respect your right to digital sovereignty, it is because it does no
by chilipepperhott 8mo ago
I know all press is good press... but there are limits.
If it feels like Grammarly does not respect your right to digital sovereignty, it is because it does not.
- mikeocool 8mo agoIt almost seems like this whole feature is designed to invite law suits. Seems pretty likely usage of Grammarly's core product has cratered in the past few years. Not totally hard to imagine one of the big AI labs paying their legal fees in exchange for putting this out there and kick starting the legal process on some of these issues.
- SoftTalker 8mo agoLLMs basically made Grammarly irrelevant as a product. Why have a tool to correct your grammar when you can just have it write the whole piece for you. And one things LLMs do well is construct grammatically correct text. So IMO they are just flinging things at the wall trying to find a way back.
- abirch 8mo agoAs Annie Duke said in her book Quit, "quitting on time usually feels like quitting too early." Grammarly was a great in the 2010s, but now it's too easily replaced. It reminds me of winzip.
- nxobject 8mo agoDepressingly enough, if Grammarly does throw in the hat, we'll lose an application of clear utility that could be run entirely locally.
- abirch 8mo agoIt seems like there are many apps that can be run locally that use LLMs. Although I haven't used this, I found it on reddit and it's made by a student. https://github.com/theJayTea/WritingTools https://github.com/theJayTea/WritingTools Seems like there could be others that are better.
- z3c0 8mo agoI believe the person you're replying to meant local inferencing. The tool you shared, like most (all?) LLM utilities, is wrapping API calls. https://github.com/theJayTea/WritingTools/blob/main/Windows_and_Linux/aiprovider.py https://github.com/theJayTea/WritingTools/blob/main/Windows_...
- nxobject 8mo agoIIRC, I think the core of Grammarly is a CL-based pattern-matching system, so it may be even simpler than that.
- BrtByte 8mo agoThe real issue seems more about transparency and consent around how the models are trained and how author personas are being used