#How up to date is Neuro's memory/knowledge? (Or does how does it work in general?)

1 messages · Page 1 of 1 (latest)

orchid vector
#

So, I'm not stating Neuro is ChatGPT or anything like that but I was screwing around with ChatGPT earlier and when asking it certain questions it kept referring to lacking an understanding since it's last "Knowledge Update" was in September 2021 so I googled it to try and find out more information.

I actually wasn't aware until now that ChatGPT had something of a cap on its information. So I went back to ChatGPT and tried asking it about something extremely recent ||(I apologise but the only thing I could think to ask was "What is Skibidi Toilet" as it was the most recent trend I could think of.)|| to which it again spat out that it has no knowledge of what that is, since it's last update was 2021.

It made me think of Neuro and how, I swear she has a habit of making statements like claiming it's currently 2021.
Of course that could just be Neuro screwing around as usual but I couldn't help but ask if Neuro's data is "living in the past" or something.

Again I don't presume that Neuro runs off of ChatGPT but it was something that got me questioning things.

image.png
glass axle
#

From what I’ve noticed neuro is stuck in 2021 but she also refers to 2019 a lot she works on set data so she doesn’t have access to current information

orchid vector
#

I know in the past she has spoke of Warframe which was surprising after the most recent TennoCon and I thought it was related at the time, but it could have just been pure coincidence looking at it now

agile roost
#

The majority of custom LLMs use the same base models. Those base models are trained on datasets that can only be collected to the point of training. Retraining is quite costly, though you can fine-tune a LLM somewhat easily. Fine-tuning is limited in the infornation you can provide. There is no good system currently in place to update LLMs on broad recent information beyond when the training occured or datasets updated. Which for most models is 2021.

orchid vector
#

I see.

#

In a way I feel kinda saddened but I guess there's not really a whole lot that can be done, more so since chatGPT is being sued and stuff for data scraping.

agile roost
#

OpenAI kinda hit gold by scraping a ton of data before anyone noticed but now everyone is wary

#

Course being Microsoft or Google gets you access to data pretty easily.

orchid vector
#

Feel kinda like I came across AI Dementia.

#

Regret looking into this topic now lol.

hard ruin
agile roost
#

LLMs are smoke and mirrors at the moment. But it may be very soon that they won't be