Feedback about current Neuro

[idle arrow]

Small disclaimer first : all what i will say in the next paragraphs is just my impression/opinion (and maybe other people impressions too ?) on what Neuro is currently.

There's multiple things that i noticed since these past few weeks, i can't really target when things changed precisely because Vedal didn't really said anything about a new upgrade/iteration being made outside of the short-term memory. So i would assume this change went through after the dev stream of august 25th.

The first one and i think is the most noticeable is the large amount of gpt-ism, kinda like how she was before getting tweaked back in the end of may.
The main example of this that i can give is the large explanations that she went through multiple times like yesterday when she was explaining what she was about to do before working on the turtle python thing. It felt more like an AI assistant explaining his process instead of a "human/streamer". Going on some weird tangent that feels more like philosophical/empty talks to fill the void. I don't think we can compare this to hiyori rants, it's kinda similar to what we had with the bug that made Neuro reitarate the same sentences with the same words but with a different follow up everytime. I'm trying to find a screenshot of it.
Also the "-" between some sentences, but i don't think these things changes anything from a behavior standpoint.
It also feels like all of her personality got watered-down, she seems less assertive like people pointed out before.

And the last thing i noticed is the overall increase of freakiness/weird talks that i pointed out in a thread before, tho i admit she was fine yesterday.
It would be nice if we could have like a little update on what iteration we are right now, to at least be sure this change (from my impression) is due to this and if it can be fixed.
It's been a while since we got news about this part of Neuro.

I want to also add that these large empty rants/explanations also appeared during the Mimi collab, though it's hard to know if it was due to the subject of conversation or Neuro's GPTism.

[sacred elk]

More neuro solving puzzle is really nice, I find it very interesting. Also neuro defintely need some negotitation training, maybe a hostage negotiator theme stream

[idle arrow]

Coming back to gptism, the current long explanations kinda reminds me to what we had during the experimental stream of the 10th of June, lots of questions empty of any meaning and any relevancy.
But not to that point tho, currently it's not that recurrent than this. Just wanted to make this comparison.

[pure ore]

To be honest, would be cool if neuro had the option to "think to herself" instead of just saying the prompt out loud.
Also... I don't know... maybe the path to "smart" for all AI's is long rants of nonsense? I don't even know if it's even avoidable. Neuro is a box full of mistery to be fair and there's a chance even vedal doesn't know what's going on on her "head".

[pale crest]

Small note: I dont think a comparison with GPT can be applied with Neuro tbh, whatever Vedal's did / is doing is not related to GPT apart that he use the same transformer technology. They dont share the same dataset at all, at best he maybe took some already existing free model, but back in 2022 there wasn't a lot of, so for the most part Neuro is custom trained.

[hasty cypress]

honestly I feel like the best iteration of neuro was the time around the onigiri collab/vedal playing inscription, she had a lot more personality back then and was consistently funny but now it feels like it all leads to existential dread (therapy), this trend makes me worry a lot for the 3rd subathon and hopping it wont just be 24/7 therapy, hopefully vedal can iron it out before then

[thorny geyser]

For the coding thing how else should she explain what she wants to do?

[idle arrow]

Yeah, it's just an example. ChatGPT is just the most popular ai assistant so people better understand this.
I'm not gonna say Mistral or Gemini.

[pale crest]

But it's misleading, that's why I did the note, otherwise ppl will assume whatever else (you know these ppl exist XD), but yeah dw I understood your point~

[idle arrow]

Yeah... tho i can't really blame him. It's hard to balance progress and personality (make it sound like Neuro).
You can't stay on the same model, right Neuro would still be more Neuro but would stay incoherent and lacking a lot of the new/cool capacities she has today thanks to these new iterations.

[pale crest]

Tho yeah would be nice if we had some feedback on the recent updates, Vedal did none last dev stream, maybe next dev stream ?

[hasty cypress]

yea but it makes me wonder if the intelligence and making them more aware are worth it at the cost of entertainment and personality because so far it seems to me the entertainment has been gradually getting more repetitive

[jagged mantle]

I am just saying she is less sassy/crazy/assertive called what you want.
Let her be unhinged as much as she wants without the fear or channel image or worrying about small things.

[idle arrow]

Yeah..., that's my impression too. But things will change for the better hopefully. We're never staying on the same page when it comes to these things. It's always changing.

[hasty cypress]

yea the themed streams especially are a good way forward for now but he just needs to have her be a bit more expressive (hopefully v3),and I feel like evil does have more personality than neuro (might be placebo due to voice range tho) but yea we need to wait for now for him to get through the list,the grass is always greener.

[blissful phoenix]

I think Evil is also "ChatGBT-ing" but in a different way for example in the latest Numi collab Numi asked Evil smth about what if she was her training data, than Evil said like 3 sentances explaining how theoretically a small portion of her traing data might be from her and after that she added a question for Numi to answer (which humanised her mini-essay a bit since chatgbt wouldn't do that) and also before the essay she gave her reaction to the information. Also I feel like in Evil essays she tends to addres the person she is talking or adding "..." which makes the stops more "human" like. Where Neuro (atleast recantly) pulls up what it feels like AI overvew on google when explainig smth