John Rodrigues introduces Inkling, a trillion-parameter model from Thinking Machines Lab built for easy fine-tuning on your own data. The accompanying tool Tinker lets businesses customize it with proprietary knowledge bases and deploy locally, moving away from generic models like ChatGPT toward tailored, on-device AI.
Chapters
Thinking Machines Lab launched one of the models called Inkling, and most of the models will tell they are the best model, but this model does say they are not the best model. They made very easy is ability to train this model based on your own data. For example, if you run your business, you use ChatGPT and Claude and all that stuff, all these models are generic world knowledge, right? You want something that connected to your, you know, knowledge base and your business and stuff though. So there's something called fine-tuning. They do that with the models. So Inkling models make it very easy to train the model based on your company data. Data. It's still very powerful, a trillion parameters and everything. Tool that they provide to fine-tune your model is called Tinker, so which gives a simple interface. You can go and customize that with your own data, so you can build like your own models and based on your own data and local, uh, which is definitely a trend that I'm seeing with AI, more open source on device and trained on their own data, because most of the businesses want models that understand their business better.