I don't trust your cloud. Your APIs are dogshit. Fuck your TOS. I'll run it myself. I will not ask permission to utilize intelligence.

The boy is half crawling and half standing on one leg and it looks like he's The Hunchback of Notre Dame lumbering across the rooftops.

And just like that we have transformers

Hey @grok, based on ALL my tweets, I am:.Which dictator?.Which philosopher?.Which fictional character?.Which politician?.Which religious figure?.Which historical figure?.Which artist?.

Yup. That happened.

It's almost perfect

Yeah I'm happy with this.

:D

The beginning

RT @BigDickBarclay: The G boys strike again

RT @theodorvaryag: Most programmers are capable of articulating and sustaining luxury beliefs in stark opposition to reality because they'v….

Where are all the LLM powered robots?.

Most of you retards would have been one-shotted by an Eliza chatbot on IRC if you grew up when I did (and could figure out how to get on IRC).

ARE YOU TIRED OF YOUR GPU JUST SITTING THERE JUDGING YOU? ARE YOU SICK OF MODELS THAT HALLUCINATE TEXT WHEN YOU COULD HAVE MODELS THAT HALLUCINATE EXISTENCE ITSELF? INTRODUCING NULLLLM AND SIMPLEMLP. THESE AREN’T MODELS. THEY’RE BENCHMARK-OPTIMIZED SPEED DEMONS THAT DO ABSOLUTELY

Now we're fucking talking. Just need to get this shit running in the browser and see what WebGPU vs native Vulkan looks like.

I write code, I write books (based off my code) and I write checks (also based off my code, nigga.

Fuck it I'm rewriting it from scratch. Was a fun toy but now it's time to treat it like the whole-ass runtime it wants to be.

After actually building an inference pipeline it's easy to see there's no actual "intelligence" here.

:D I now own the entire inference pipeline. Next up, big boy models and even more throughput chasing.