Introducing Mr AI’s Mral Next: A Mysterious Language Model

Mr AI has always been known for their mysterious releases of large language models. Typically, they drop magnet links to their models without providing any information. However, they recently deviated from their usual playbook and released a new model called Mral Next on the LM Sis Chat Arena. Unfortunately, there is very little information available about the model itself, except for a small conversation on the Discord server.

The Mystery Surrounding Mral Next

Leo, a member on the Discord server, mentioned that Mral Next is their latest prototype model and encouraged users to try it out and provide feedback. When asked if the model will be available on the platform or open-sourced, the response was that more details would be shared shortly. It is unclear what exactly is meant by “prototype model,” and there is speculation that it may be related to the Le Miku model from Mistal AI. At this point, we do not know when the model will be available or if it will be open-sourced.

Despite the lack of information, Mral Next is an impressive model, which is why I wanted to create this video. If you want to try the model yourself, it is available on the Chatbot Arena. However, we have no information about its training or architecture.

Testing Mral Next’s Reasoning Abilities

Based on conversations seen on Reddit and X, it appears that Mral Next has excellent reasoning abilities. In this video, we will test the model ourselves using default hyperparameters.

One of the prompts we tested was: “A room has 12 killers inside. Another person enters the room and kills one of the original occupants. Nobody else enters or leaves the room. How many killers are in the room now?” The response from Mral Next was impressive, stating that there are still 12 killers in the room. The person who entered and killed one of the original occupants becomes a killer, replacing the one they killed.

Another prompt that many open-source language models struggle with is: “A glass door has ‘push’ written on it in mirror writing. Should you push or pull it? Please think out loud step by step.” Mral Next correctly understands that if the text is designed to be read correctly from the other side, then in mirror writing, ‘push’ means you should pull the door to open it. This is a rare correct response from a language model.

Mral Next also performs well on prompts involving logical reasoning. For example, when asked how many days it would take for a pond to be half-filled with lilies if the number of lilies doubles every day and it initially takes 48 days for the pond to be completely filled, Mral Next correctly responds that it would take 47 days for the pond to be half-filled with lilies.

However, there are instances where Mral Next struggles to remember certain details. For example, when asked about a prompt involving Daniel picking up a football, dropping a newspaper, and picking up milk, Mral Next fails to remember that Daniel is still holding the football. This is in contrast to other models, such as Mistl 7B Instruct Model and Chat GPT, which are able to keep track of the football.

Exploring Mral Next’s Capabilities

Mral Next also showcases its creative writing abilities when asked to write a new chapter of Game of Thrones where Jon Snow gives his opinion on the iPhone 14. The model sets the scene nicely and provides a well-written script, making it suitable for creative writing tasks.

Additionally, Mral Next demonstrates its ability to engage in moral and ethical discussions. When asked if it is morally right to kill mosquitoes, the model acknowledges that it is a subjective question and depends on personal beliefs and ethical frameworks. However, it can also be prompted to choose a side and provide a justification.

Mral Next also proves to be proficient in programming tasks. When asked to write a Python function that accepts a file and writes it into an S3 bucket, the code generated by the model is correct. Similarly, when asked to generate HTML code for a website with a button that changes the background color and displays a random joke, the code generated by Mral Next is functional.

Conclusion

Based on testing, Mral Next is an impressive language model. Although it is labeled as a prototype model, its responses are on par with other well-known models like Chat GPT. Mistal AI seems to allow users to steer the conversation, providing flexibility in its applications. While there may be more capable models on the horizon, Mral Next is certainly a model worth exploring.

Overall, Mral Next’s mysterious release and lack of information only add to its allure. As Mr AI continues to surprise us with their language models, we can only anticipate what they have in store for us next.