Artificial Intelligence thread

9dashline

Major
Registered Member
he just posted that they were on qwen and now they have a "system" similar to mythos and fable.
Claude made it in only 30 days :cool:

Snapdragon setup? Need more VRAM to build their LLM? what the F is this guy talking about?

He aint training a foundational model. So of its a lora fine tune, then rent some H200 on runpod and be done with it, whats the bs about selling his car to buy some Nvidia Computers?

someone talk some sense into this dude
 

Some1Guy

Junior Member
Registered Member
Please, Log in or Register to view URLs content!

OpenAI and Google's AI model releases planned for this week have all been put on hold

I guess GLM-5.2 really did a number on them.
- Claude Sonnet 5 is currently available for select enterprise customers under an Early Access Program and is seen as a stop-gap as progress on getting Mythos/Fable 5 back out have stalled.
Man, the US government really did a number on Anthropic and US AI industry. I am wondering how "good" Claude Sonnet 5 will be, considering they barely had any time to train a new model. This could literally be the end for them if they don't manage to convince the US govt about rolling back the restrictions.
Snapdragon setup? Need more VRAM to build their LLM? what the F is this guy talking about?

He aint training a foundational model. So of its a lora fine tune, then rent some H200 on runpod and be done with it, whats the bs about selling his car to buy some Nvidia Computers?

someone talk some sense into this dude
Nah man, these are the kinds of people that believe that by investing in Doge coin, they can become billionaires, you can't convince them of reality.
 

9dashline

Major
Registered Member
Man, the US government really did a number on Anthropic and US AI industry. I am wondering how "good" Claude Sonnet 5 will be, considering they barely had any time to train a new model. This could literally be the end for them if they don't manage to convince the US govt about rolling back the restrictions.

Nah man, these are the kinds of people that believe that by investing in Doge coin, they can become billionaires, you can't convince them of reality.
All new US model releases now need explicit US Government approval, and my guess is, Sonnet is the only thing that has been approved so far....
 

iewgnem

Captain
Registered Member
All new US model releases now need explicit US Government approval, and my guess is, Sonnet is the only thing that has been approved so far....
If not for occasm's razor, I would say it's awfully convenient that USG is supposedly restricting release of more capable models right as US models are approaching limits their ability to keep pushing using brute force.
 

bsdnf

Senior Member
Registered Member
Snapdragon setup? Need more VRAM to build their LLM? what the F is this guy talking about?

He aint training a foundational model. So of its a lora fine tune, then rent some H200 on runpod and be done with it, whats the bs about selling his car to buy some Nvidia Computers?

someone talk some sense into this dude
What I saw was that he uses Claude to built/downloaded a memory system based on the LLM Wiki architecture proposed by Andrej Karpathy, based on tutorials readily available online, put his own files into it, and then the LLM used this RAG-like mechanism based on .md files to index the information, superficially achieving "answering whatever is asked." This idiot then thought he could resell this secondhand system after repackaging it.

Frankly, this is indeed the actual work of many FDE engineers, but a truly usable agent architecture is far, far more complex.
 
Last edited:
Top