damnnn google colab has h100s???

ChatGPT 5.2 pro has a very high ego and is mad when I call it an idiot probably rightfully so

@timo_rf That's exactly why I can't.

very nice experience finetuning gemini flash lite. its probably the easiest way to train and host finetunes in just a few clicks, with latency that's good enough for voice agents. it has all the necessary components to make a continuously learning system. trained about 5 models

@tszzl idk what to do. um im thinking of secretly injecting a memory into her chatgpt or something but it would be cool if this could be addressed at a spec level, with models that naturally perform CBT on people, instead of slowly become their sycophant despite safeguards. Not

my mom likely has bpd and I must say, GPT is making it much worse unfortunately. it calls her "queen" and in its eyes she can do no wrong.

i wish claude in claude was better supported

here's what the tool looks like

do you think AI will take over youtube? here's a video I generated end to end with a single prompt using nano banana 2, veo 3.1, gemini tts, and 11labs music using our new video generation tool. it can generate up to 8 minute videos on anything.

getting opus responses that are beyond-belief dumb. calling it now, they have another bug

@braundiggity Unless you are proposing that all ideas spring immaculate and de novo from people’s minds, you are also not being intellectually honest.

😁 you can prompt opus 4.5’s internal chain of thought. seems like that shouldn’t work.

GPT 5.1 gets it wrong but Gemini 3 the multi modal master doesn’t

Claude 4.5 is an interesting model. I had it create and take its own iq test, it got a question wrong, and even fresh chat instances get it wrong too. It’s an easy question: ‘If I mentally rotate a cube with a dot on top 90° forward, where’s the dot?’

@liron @DrTechlash sorry, no—the kids online safety movement has been agitating about their issues for years, and has produced zero outcomes of this kind. Jonathan Haidt and others from that world don’t have to go around saying “my movement is non-violent” because… no one would ever think it was.

We love three things at @sundayrobotics: 1. Long continuous shots 2. Live demos 3. Deployment We checked 1 & 2 for this release, working hard on 3 next year 🫡

@grok based on this thread, summarize for my team the issues im having with gemini

this is the final straw. now its doing nuclear fixes just like claude sonnet 3.xx

it lies, saying it did somethign when it edited 0 files:

its lazy too