This is pretty neat. How complex can the tasks be? I'm also impressed by the model's capability in the demo considering it's only a 3B model. Curious to see what a larger local model like gpt-oss-20b would be able to accomplish, hosted via a local API on, say, Ollama.
This looks cool and I am also impressed with the tiny Qwen models.
I wonder if there is a malicious use of this similar to how people did bitcoin mining in visitors browsers. Maybe someone will create a botnet of compute slaves to run 470B models in a distributed way ;)
Super interesting! If small models are to get as better as they make them to be, maybe eventually you can have perplexity comet like browser agents running locally
This is pretty neat. How complex can the tasks be? I'm also impressed by the model's capability in the demo considering it's only a 3B model. Curious to see what a larger local model like gpt-oss-20b would be able to accomplish, hosted via a local API on, say, Ollama.
This looks cool and I am also impressed with the tiny Qwen models.
I wonder if there is a malicious use of this similar to how people did bitcoin mining in visitors browsers. Maybe someone will create a botnet of compute slaves to run 470B models in a distributed way ;)
Super interesting! If small models are to get as better as they make them to be, maybe eventually you can have perplexity comet like browser agents running locally
[dead]
[dead]