Kurt Uwe Stoll

@Fchatban When GTA 6?

🤖❤️

@dankvr Let’s talk I’m dabbling around in a similar domain

Chatrooms move too fast for people to keep up with. To solve this common issue, I'm working with a couple peeps today to use LLMs to summarize daily chatlogs per channel / working group / server to extract insights about: - FAQ - whose helping who - action items - daily progress

@AnthropicAI Generate a playlist that can be imported to spotify for a 18 month german toddler with relaxing space jazzy classic music, 20 tracks. 🤘

@alexocheema @exolabs The main question is the optimal Mac mini model. I don’t think maxed out will be best.

M4 Mac Mini AI Cluster Uses @exolabs with Thunderbolt 5 interconnect (80Gbps) to run LLMs distributed across 4 M4 Pro Mac Minis. The cluster is small (iPhone for reference). It’s running Nemotron 70B at 8 tok/sec and scales to Llama 405B (benchmarks soon).

@itsandrewgao @perplexity_ai It’s a bit lame that we now do with system two artificial intelligence and a lot of GPU power what we could do like 20 years ago with the Semantic Web and knowledge graph technology.

i haven't tweeted about @perplexity_ai, mainly bc i wasn't that impressed with the technology. it seemed to be just taking my search query, googling it with SERP, & then summarizing the results (SERP+LLM wrapper) i decided to give it another try yesterday since I got pplx pro free through uber eats and @AravSrinivas announced new reasoning . i was really impressed. i think these screenshots capture the importance of reasoning in REsearch. old perplexity was basically a glorified summarizer. new perplexity can actually do helpful research that saves me time. a query such as the one in the screenshots naturally requires several branching steps. you can't just google the query and read from the top articles because there is no article about it. LEFT: Perplexity regular RIGHT: Perplexity PRO kudos to aravind, @denisyarats, and team!

@gdb @nvidia These wannabes obviously do not have any clue how to use a computer. They forgot the power plug, the network and the cooling whatsoever. Complete losers.

First @NVIDIA DGX H200 in the world, hand-delivered to OpenAI and dedicated by Jensen "to advance AI, computing, and humanity":

@paulg So you are arguing for an intermediary scalping monopoly.

There is a missing startup here. Restaurants should be making this money, not scalpers.

@Mascobot @simonw @runpod Seconding

@simonw Probably @runpod

What's the cheapest option right now for me to spin up a Linux server somewhere for an hour with enough GPU to run the latest Mixtral model?

@ClementDelangue Data problem.

Who will be the open-source Suno?

@burkov They flipped to prepaid credits API. Must be tight

There are only 2 possibilities: 1. GPT-4 is a 2T model and OpenAI uses an entire node of 8xH100 (that costs $400,000) to serve the inference just for you for $20/month. or 2. GPT-4 is a model that is 10 times smaller (it cannot be smaller than 200B) and OpenAI uses one H100 (that costs $50,000) to serve the inference just for you for $20/month. In either case, the numbers don't look good for OpenAI.

@michael_nielsen Agree. Was too tunneled in LLMs. There is still other relevant software.

@ustoll I don't know. I think a lot of high-profile users - including many companies - would happily pay a compute price to use a HE Google Docs.

Homomorphic encryption has come a very long way since the first scheme was announced by Craig Gentry back in ~2009: en.wikipedia.org/wiki/Homomorph…

@HamelHusain @axolotl_ai @winglian @johnowhitaker Axolotl rocks big time!

FDSP + QLoRA from is now merged into @axolotl_ai 🔥 GPU Poor -> GPU Rich💰 @winglian fast on the integration per usual 🏇 w/ @johnowhitaker for behind the scenes support This is how to use it: 1. Upgrade axolotl per the README 2. Set `adapter: qlora` in the config 3. fdsp needs to be enabled 3. model types supported are currently llama, mistral, or mixtral Example config: github.com/OpenAccess-AI-…

@jeremyphoward @Tim_Dettmers @huggingface @Mobius_Labs Great, is it easily scalable to multiple nodes?

Today, with @Tim_Dettmers, @huggingface, & @mobius_labs, we're releasing FSDP/QLoRA, a new project that lets you efficiently train very large (70b) models on a home computer with consumer gaming GPUs. 1/🧵 answer.ai/posts/2024-03-…

@randhindi @zama @multicoin @protocollabs Congratulations!

@huggingface 🤗

We're having some infra issues; we're working on it. Please send hugs! 🤗 In the meantime, import os os.environ['HF_HUB_OFFLINE']=1

Not selling #GPU anymore and going fully vertical integrated #AI #monopoly would be the obvious @sama move for #NVIDIA

The elegance of ML is the elegance of biology, not the elegance of math or physics. Simple gradient descent creates mind-boggling structure and behavior, just as evolution creates the awe inspiring complexity of nature.

🏡 I asked GPT4 to imagine if nomadlist.com would start coworking spaces + nomad bases around the world and how it'd look: 1) A renovated villa in Portugal transformed into a coworking space, overlooking the Atlantic Ocean. 2) A coworking space in Bali that blends traditional Balinese architecture with contemporary design, surrounded by rice fields. 3) An urban high-rise coworking space in Bangkok with a rooftop area. 4) An illustration showcasing a sustainable coworking space with solar panels, rainwater harvesting systems, and a community area surrounded by greenery.