a b

[a b @ab11216174]

@ArielKwiat Literally all they had to do was agree with OpenAI to claim credit together. This could've been a nice moment where both labs come together to solve the first millennium prize problem with AI. Instead it's just this shit. Truly pathetic.

[a b @ab11216174]

@OrangeMint_OwO More like one message

[a b @ab11216174]

@tianyi I think you should not waste your time on V4 Pro. It is an inefficient pretrain and not worth it for the increased computation. Just make sure V5 Pro is good for its size.

[a b @ab11216174]

@bramiozo @theo "The destruction of something beautiful." Well, I guess that wraps it up everyone. We can't have technological advancement and automation because it "destroys".

[a b @ab11216174]

@Witchbunny_ @Youssofal_ Yeah, I understand, but Mistral is not able to add good agent capabilities in post training. So their models are barely useful to pretty much everyone.

[a b @ab11216174]

@Heattokun @ReticentY2K This project in particular has been in the works for a very long time. There are plenty of 100% ai decompiled projects out there right now.

[a b @ab11216174]

@jayson_casper It means you are taking too much and your body is not used to it. Reduce the amount, and slowly increase over time.

[a b @ab11216174]

@Witchbunny_ @Youssofal_ 0% is not useable at all. The entire reason LLMs have blown up the past year is because of being good at agentic tasks. No one would care about LLMs otherwise. At the end of the day, a model that cannot handle agentic workloads is barely useful to almost everyone.

[a b @ab11216174]

@thesharkpaw @rakishromantic Evidence?

[a b @ab11216174]

@petergostev This whole RL agent craze really started around a year ago. They are definitely doing that, but there hasn't been enough time to see the products getting announced.

[a b @ab11216174]

@word_rotator You can only stick your head in the sand for long enough before the world moves on without you.

[a b @ab11216174]

@PeasantSmith It's just another instance of LLM research being so quick. Any Qwen model with 'Next' in the name is using brand new upcoming architecture (which is not present in 2.4T).

[a b @ab11216174]

@PrimeDev_x @VideoCardz Not at all. The only way to get the game files is to jailbreak the latest firmware so you can dump the decrypted copies. The latest cracked firmware is 2 years old. So new games can't be dumped at all. Definitely not GTA 6

[a b @ab11216174]

@argofowl The only proper way to use them is to wait until your weekly limit resets, go all out and consume the entire limit quickly, in like one day, then use the banked reset.

[a b @ab11216174]

@Keepcalm8156 @coah80 Some people are just truly beyond reason. Most likely, someone will just fork it and get it done 5-10x quicker

[a b @ab11216174]

@melvindvivas Astra is there for OpenAI to distill into GPT 6 Sol, Terra, Luna

[a b @ab11216174]

@Gozengogo_ Instead you should ask it to finetune an existing time forecasting model for stock trading. You will get better results.

[a b @ab11216174]

@TinoWening Stop comparing on benchmarks and actually use them. They're not the same at all. Luna has none of the good intuition Astra has.

[a b @ab11216174]

@ctatedev Luna lacks intuition, and will often be clueless as to where to go unless the project is very straightforward

[a b @ab11216174]

@izutorishima People who deny technology due to personal bias can only stick their head in the sand for so long until the world moves on without them