2 posts tagged with "llms"

Introduction​

A popular use case for LLMs is as a writing assistant, for example for editing text as a grammar and spelling checker. Students may need help with writing essays, while businesses use it for drafting and editing emails, marketing copy, reports, etc.

An early adopter of language models for writing assistance was Google within the Gmail product. With broader availability of LLM platforms for lower costs, major companies have integrated LLMs into their products for writing assistance, such as Apple and Microsoft.

However, many of these systems present a popup and do not display suggestions inline.

There has been a gap between the ability of LLMs to generate text and the tight integration of LLMs within text editors. We consider why this is the case, taking the current most popular frontier LLM, ChatGPT-4o mini, as an example, and explain why you should consider using a task-specific LLM for grammar checking (such as the one offered by Sapling's API).

Introduction​

Since the release of OpenAI's ChatGPT in November of 2022, hundreds of millions of users have used it to generate billions of tokens per day.

ChatGPT, like other large language models (LLMs), has a distinctive style, and in the initial months of its release would frequently use the phrase "as a large language model" as a precursor to its responses.

Since then, this tic appears to have been reduced, but led us to wonder: what other phrases are strongly associated with ChatGPT's style, but not with the writing of humans? Beyond these statistics, what does this tell us about its training?