Anonymous
7/15/2025, 10:49:06 AM
No.105912840
[Report]
>>105912910
>With the cost of a year of claude max / chatgpt pro subscription you can get an m3 ultra and run the full deepseek r1 locally @ ~18tk/s plus you get the hardware as a bonus
What am I missing here? Why isn't this more common?
Because deepseek is a massive, smelly, stinking, blackened pile of dogshit since the Chinese can't create, only copy. There's a reason its cheaper anon
Anonymous
7/15/2025, 10:54:21 AM
No.105912873
[Report]
>>105912855
All I am hearing is I can't prooompt without my ai assistant each one of its messages with muh gay little emojis
Anonymous
7/15/2025, 10:56:44 AM
No.105912890
[Report]
>>105912911
What are you using it for? Its shit at vibe coding. Shit at roleplay.
Anonymous
7/15/2025, 10:59:38 AM
No.105912911
[Report]
>>105912890
I give it a semi detailed description of some problem I'm having along with the code and smoke a cigarette at my balcony while it works out the root cause. I don't use deepseek to write code from scratch, but it's certainly possible. Vibecoding from zero slows you down if you can actually program.
Anonymous
7/15/2025, 10:59:40 AM
No.105912912
[Report]
I do all my LLM inference for free on lmarena THOUGH, where I get access to pretty much every model in existence
sometimes I wonder what their business model is. I mean, sure, they're processing/selling my prompts, but is it really that valuable to cover the cost of all the latest and greatest models?
Anonymous
7/15/2025, 11:01:07 AM
No.105912920
[Report]
>>105912926
>>105912855
China is like 50 years ahead of the west in every single physical science and in math. The top US researchers are all Chinese nationals. You need to update your NPC programming.
