Andy Halliday

But for really sort of personal household complex issues up to and including multi-stage issues.

efforts to apply uh reasoning to legal and or uh regulatory issues uh those kind of things the deep research that gen spark does and sort of the um the the mission orientation that the gen spark super agent has are really really good i'm not sure how speakly works or how it adds to that uh

I've always been able to prompt GenSpark by clicking on the microphone thing and speaking it or using WhisperFlow to do that.

So I'll find out.

I guess it's going to be that it's a really good respondent in voice.

So taking the output of the model.

Oh, here we go.

You can highlight text and tell it to translate or rewrite just by voice asking it to.

So you can give it voice instruction.

which is good.

But I hope that it can also eventually get to a really fluent dialogue with you as a personal agent.

Yeah.

All right.

I'm going to bop over to a new item here.

which is I want to talk about small language models and particularly those models that are being designed to operate at the edge on your mobile device.

And so I've been impressed by a newly released model, at least the press about it, which is from a company called Liquid.ai.

Now, Liquid.ai just released their next generation of on-device models, and it's called LFM 2.5.

And there's a range of them, you know, all in the one billion parameter class.

That one billion being the approximate number of parameters that are in the deep neural network that's there in that mathematical model.

So $1 billion compared to, you know, there are, I don't know, it's probably typically when you see in open source models, they'll show how many parameters are there.