Mycroft was an open source alternative to Alexa.
Qwen3 and Gemma models are fairly capable, they are slow-ish (a few tokens per second) but will run.
You can start building with cheap hardware and simple models, and use something more capable once you're more confident on the use case.
It’s something I’m exploring - stay tuned :)
I suggest checking Arvind Narayanan's tinkering with his own children. That being said not every parent is a Princeton professor working on AI so tread with caution.
Also on a pragmatic front if you want to tinker on the domain something potentially easy to setup yet relatively open https://www.home-assistant.io/voice-pe/
I’m looking forward to a future where everyone could have their own personality model that is actually fine tuned at the weight level. Plus if we come up with better ways to do lifelong learning, personality could emerge from the robot’s experiences
https://github.com/skorokithakis/havpe-server
It'll just send the recognized speech to your own API endpoint, and speak whatever is returned.
Another benefit is if your hands are full. For example if you're cooking or driving.
Another benefit can be speed. If you're doing something in your house near the smart speaker, it's probably faster to ask it a question than to pull your phone out of your pocket, unlock it (I only have a password, not fingerprint/face ID), and type in the query. For people who are slow at typing, this benefit is larger.
Yeah this shows severe misunderstanding of how sociopaths up there are shaping the future, while masses flock for next dopamine kick to wherever the carrot goes.
And I dont mean the word sociopaths as a slur but rather a factual description of basically all people in power - sillicon valley (or redmont or whatever) moguls, politics, pentagon and so on.
I am not generally a negative person and have rather positive outlook on future overall, but such naivety... no clue where it comes from. Even EU with its open source initiative on one side is visibly a control freak aparatus when looking at other actions. And apart from few small havens like Switzerland EU is still behaving light years better than US where most of these inventions come from. Then there is China. No.
1 is the group you discuss, the ones that want to control the models to do what they want so they can control the future.
Group 2 is the ones that want models to be able to do anything, which is the group you seem to belong to. Of course in your desire to do anything yourself you find yourself unable to imagine the hellscape of a world that will be created when you release billions of agents that have little desire to fit in the rules of human society.
Unfortunately... yeah, they're all stuck in the useless Esper management still if you factory default them away from whatever weird factory test app they're in. Care to throw us a little bone? ;)
