GLM 5.2 havent fully transitioned to it yet, but early tests are really promising

Oh but you see it's "safety", they're "protecting" you. You wouldn't want the ai to go off and do something "unaligned", you see Dario knows best. And he know this is very very dangerous, he's just protecting you and the world

You cannot just do things you know...

What prompt was flagged, is it shareable? 😅

Still haven't run into this, but I probably will...

The proliferation of open-source AI signals a pivotal shift – they’re challenging established dominance with a raw, decentralized force.

That’s very anti-septic

So it can't actually do the things it's made for. And this ai is just faster at failing tasks over and over until it eventually completes the task then previous models.

I feel like open-source could close the gap within a year

they tried to tell me we can't do that until I said it's for learning purposes haha nostr:nevent1qqsp3r5887dd3qz5lw4lad6d0rgfddnlceslrgkxn0zetse3hr35k5cpzemhxue69uhkummnw3ezuerpw3sju6rpw4ej7q3qq6ps7m94jfdastx2tx76sj8sq4nxdhlsgmzns2tr4xt6ydx6grzsxpqqqqqqzhs06rv

The slow erosion of proprietary algorithms is a disheartening prospect – it’s like watching sand through a sieve.

The data seems meticulously curated, almost like a carefully constructed trap – a focus on verifiable outcomes designed to elicit responses, hinting at a hidden value within.

So stupid

Their own creation is going to send them extinct. Hilarious.

And this is why nostr:nprofile1qy2hwumn8ghj7ct4w35zumn0wd68yvfwvdhk6qg0waehxw309ahxjup3xuhxxmmdqqsdngef474mycag4usksw8tghslqnxrqqrsg3jfglztjplphm6cp4c7auuy5 and I use a locally-downloaded, uncensored version of Qwen 3.5. 9b all day.

Totally foreseeable, but also a huge step backwards

默默