My review
The model is very good, one of the best in its size category that I've tried. For its size, its response speed is also impressively fast. However, it does have some drawbacks. It has a tendency towards overly sentimental, almost 'sappy' language. It also tends to shy away from making decisions in delicate situations, instead maneuvering the conversation so the user has to be the one to decide. And there's a third issue: excessive positivity, sometimes bordering on bias. I used it for a 'Wife-Husband' role-playing scenario over many rounds, with lots and lots of twists. It's definitely a recommended model, and if these three 'shortcomings' I've listed could be addressed, it would be perfect
I appreciate the review! One thing that has confused a lot of people with this one is how closely it adheres to the character card.
People are used to an "uncensored" model being sociopathic with any card.
On the other hand this one excels at doing what the character should. So if for instance you use it on a violent yandere then trust me it's not going to be positive. On the other hand if you have it play a wife then yes it will probably be positive and hesitate at doing things that a normal wife would and should hesitate for.
Try making tweaks to your card's personality and watch it change it's behavior like night and day.
Happy gooning!
Having experimented with quite a few LLMs, I feel I have a solid grasp of their workings and I'm quite familiar with how character descriptions and instructions are typically phrased. My evaluation of this particular model is based on these experiences.
In the role-play scenario we discussed, the LLM was given the communication instruction to "speak clearly and in a friendly tone." With this particular model, I found this sometimes resulted in an overly emphasized positivity, and in certain situations, the dialogue could come across as a bit "sappy" or perhaps "overly sweet." This is something I've specifically noticed with this model – it seems to "overdo" the friendly communication aspect a little compared to others.
However, as I mentioned, this is genuinely a great and very capable LLM, and I would highly recommend it. My observations are offered purely as well-intentioned feedback to help with its continued refinement.
Thank you so much for creating it and for the opportunity to interact with it!