holy shit this thing is pure fucking SLOP AHAHAHHAHAHAHAHAHAHAgpt 5.6 did better on the same prompt earlier todayreal use case in a real project not one shot promptkek megaflop from openai. probably gonna switch to claude
>>109882304>no proof/thread
>>109882304I unsubscribed from Claude
It's astra!
>>109882345i dont want to be too specific, but basically asked it to write some relatively trivial code just to collect some statistics in a c++ codebase. code it wrote does not match the style of the codebase AT ALL. completely RL fried. With that kinda output I would have believed you if you had told me it's some benchmaxxed chinese model KEK it's so bad it's funny
>>109882353>>109882304So how does one get reliable quality slop machine these days? Clearly these subscritptions are a scam. But buying hardware is nearly impossible, makes no sense for most use cases.Solutions?
>>109882544full google ecosystem lock-in, don't trust the shills
>>109882544Stop being a cheapskate and buy the frontier model
>>109882553Google is very far from being good. Not that their slop machines and models are much worse, no. But their software in general is now below Microslop tier and I'm not even joking. Everything is ugly, convoluted, riddled with legacy parts that are probably full of vulnerabilities. I would not be surprised if later turns out it was infiltrated by NK coolhax0rs and they've been feeling like home for 10 years inside goggle infrastructure, with full access to everything.
>>109882544Realistic local models today outperform frontier models from 1 year ago. Who knows how the pace of development will proceed, and how local models will be attacked with legislation, but I am confident the future is local models for those with the hardware. For now I am just juggling whichever of the two slopcorps gives me a better deal. The question I have been debating is whether the same effective hardware will be more expensive a year from now (i.e invest in rtx6000 now or wait for future releases)
>>109882636google or 5 codex plans
i used it, its cheaper than before and it has much better talking style than 5.6 and it did great work on my projects. It worked longer and harder and did more technically sophisticated and coherent work like I wanted
>>109882304that's because GPT-6 "Sol" is actually Terra, Astra is the GPT-6 Sol-class
