albrewer

I mean, this is anecdata, but I've had to combat multiple hallucinations from Opus 5.5 _today_ about what a codebase I hand-wrote just last year does.

Qwen 3.8-27B being on par with Opus 4.6 from ~18 months earlier is a sort of case in point, right?

I have a hook that auto-denies when it sees 'python -c "', ' awk ', ' sed ', and '&&'

> the cost is included in your electricity bill

Acting like an extra $20 on my electric bill is equivalent to a $200/mo subscription is... a take.

A series of hot takes:

Benchmarks are useful but only on a log2 basis. One model performing at 50% and another at 75% is just as impressive as one model performing at 78% and another at 90%. Confoundingly, a benchmark becomes useless once a frontier model scores over ~95% on them.

If I get a better Gemma 26b-a4b MoE model out of it I'm all for it. Really wishing Qwen 3.8 would release a MoE variant of their 27b model size, but I wont complain if google beats them to it.

I think it'd be pretty easy to fix, too. Have a point on the line that represents "10 seconds from now" and orient your camera along the line between that and the point representation of your car, with slow, smooth panning so as not to get jerky transitions. If "10 seconds from now" is within a certain range of the car, it ought to just default to whatever its doing now.

Because I spent $15k in AI costs last month doing real work at my day job; I'd like to do the same thing for myself but I don't like lighting cash on fire.

It it has less than 64gb then probably not

The dream. Or a qwen-3.8-coder model.

Not OP, but your math is a bit off - I have solar panels :)

What problems are you running into? Ive been running multiple qwen coder next agents locally and it works fine.

> the battery often dies around the time most people would consider upgrading anyway

Most people I know would prefer to just use the same phone they're familiar with until its physically not possible. They don't want to learn new things. They dont care about the latest features. It's an appliance; a means to and end and nothing more.

My AI experience has varied wildly depending on the problem I'm working on. For web apps in Python, they're fantastic. For hacking on old engineering calculation code written in C/C++, it's an unmitigated disaster and an active hindrance.