0 ★ 4 ↺
From what I understand #deepseek works differently, it seems more efficient and is a type of model called Reasoning Model whereas LLMs are text creators. This thing being open source needs to be studied in depth. I am interested but also doubtful and concerned. For example you lower the bar on the resources needed to implement #AI on devices. Apple puts it on iPhone 16 and on PCs with powerful processors not accessibile to the big part of the market. Here we see a future of low-cost devices with AI bloatware of various and dangerous nature.
...
mario boosted
@mario Please, allow me to point out some inaccuracies in your post.
Reasoning models are still LLMs, the technique and implementation is also called CoT, Chain of Thought.
Models are more accurately open weight. The code for training them is open source if available.
Phones and laptops are not capable to run state of the art models yet. Deepseek did not lower the cost of inference at all, only training.
Apple already put AI bloatware into the iPhone. It's too late for this concern to be heeded.