Open weight models released
Apple: 145
@spacexai: 0
🚨 Alert: Apple just dropped FastFrames on Hugging Face: "generate videos at interactive latency".
huggingface.co/apple/FastFra…
1
15
Replying to @krika2810
they get stuck when the context gets stale, because f(a)=b
have a cheap model watch the trace and inject relevant context to guide the model towards the desired output
1
1
10
considering
teapawt.net/broadfield_dev/status/…
Need $100 million to train and serve custom models to 5000 enterprise customers. Revenue projected to be $5 billion per year.
3
Replying to @brian_armstrong
what
are you running a neocloud serving hundreds of custom finetuned models?
12
Replying to @clif_high
it would be quantum mechanical
first you need to find the right entangled particles, and decode them in the correct order
84
Replying to @BillyM2k
prediction
teapawt.net/broadfield_dev/status/…
Replying to @edzitron
Prices will drop if supply outpaces demand.
They'll dump the GPU's into the ocean if they have to.
3
Replying to @NoemiTitarenco
My agent only has write access to its sandbox. I trust LLM's about as much as a random number generator.
18
Replying to @MicrosoftAI
Local is good, but you can have secure cloud inference on any H100+ GPU using a Trusted Execution Environment.
427
Replying to @TinfoilTricorn
awesome, I'm liking the idea of the looping transformer to save RAM without cutting flops. bonus points if relevant data can be injected into the loop too. Should help get more out of modest hardware.
1
6
Replying to @nikilravi
I run at least 2 async agents every turn, one manages the context, builds the wiki, updates recursive learning, and the output from it only reaches the main agent dynamically injected into the prompt when relevant
24





