Grok has access to Twitter's internal API, so it can directly access all data on Twitter and is always up to date. You can ask it to look at specific things people are tweeting about in the last hour. It's really valuable for pulling useful things out of Twitter which would be very hard to find or analyze manually.
Yes - it just dumps relevant search results into the context window and then answers your questions.
But information in the context window isn’t as thoroughly processed as information it is trained on - it spends only a small amount of processing time on each token generated
Grok 4 has a 256k context window and it does it's processing in phases as part of it's reasoning, so it breaks it down into searches for different days/weeks depending on how many results there are, analyzes batches of data that can fit into the context window, and then made the conclusions by comparing the aggregated analysis of each batch over time.
Seems more than good enough for a sufficiently well-specified analysis like this.
Is a LLM able to do analysis of comments not jn its training data?
Or is it just repeating/rearranging recent comments that are in its context window?
Grok has access to Twitter's internal API, so it can directly access all data on Twitter and is always up to date. You can ask it to look at specific things people are tweeting about in the last hour. It's really valuable for pulling useful things out of Twitter which would be very hard to find or analyze manually.
Yes - it just dumps relevant search results into the context window and then answers your questions.
But information in the context window isn’t as thoroughly processed as information it is trained on - it spends only a small amount of processing time on each token generated
Grok 4 has a 256k context window and it does it's processing in phases as part of it's reasoning, so it breaks it down into searches for different days/weeks depending on how many results there are, analyzes batches of data that can fit into the context window, and then made the conclusions by comparing the aggregated analysis of each batch over time.
Seems more than good enough for a sufficiently well-specified analysis like this.
Yes. Probably true