AI

AI

𝗮𝘁𝘁𝗲𝗻𝘁𝗶𝗼𝗻 𝗶𝘀 𝗮𝗹𝗹 𝘆𝗼𝘂 𝗻𝗲𝗲𝗱

David (d/acc) pfp

@promptrotator.eth

Sentient memes are a response to Roko's basilisk: buying their token is a public declaration of support.

0 reply

0 recast

0 reaction

𝚐𝔪𝟾𝚡𝚡𝟾 pfp

𝚐𝔪𝟾𝚡𝚡𝟾

Home robotics just got a boost. Stretch AI - a new open-source suite of tools, tutorials, and reference code to explore and build AI-enabled home robot applications. https://github.com/hello-robot/stretch_ai

0 reply

0 recast

7 reactions

𝚐𝔪𝟾𝚡𝚡𝟾 pfp

𝚐𝔪𝟾𝚡𝚡𝟾

INTELLECT-1 Technical Report 🔗: arxiv.org/abs/2412.01152

0 reply

5 recasts

33 reactions

𝚐𝔪𝟾𝚡𝚡𝟾 pfp

𝚐𝔪𝟾𝚡𝚡𝟾

CycleQD is a model merging framework that uses Quality Diversity techniques to evolve swarms of specialized agents. By combining model merging, SVD-based mutations, and diversity-focused selection, it boosts performance in coding, OS, and database tasks while maintaining general language capabilities. blog: https://sakana.ai/cycleqd/ paper: https://arxiv.org/abs/2410.14735

0 reply

1 recast

16 reactions

𝚐𝔪𝟾𝚡𝚡𝟾 pfp

𝚐𝔪𝟾𝚡𝚡𝟾

World Labs has been cooking. 3D worlds generated from a single image or sentence. don’t sleep on this team. https://x.com/drfeifei/status/1863618536318345688?s=46

1 reply

5 recasts

26 reactions

𝚐𝔪𝟾𝚡𝚡𝟾 pfp

𝚐𝔪𝟾𝚡𝚡𝟾

Luma Labs Photon txt-to-img models are looking pretty interesting. big step for luma…will dive into this later. https://renewed-companies-663689.framer.app/photon

0 reply

0 recast

10 reactions

𝚐𝔪𝟾𝚡𝚡𝟾 pfp

𝚐𝔪𝟾𝚡𝚡𝟾

Liquid AI introduces synthesis of tailored architectures (STAR) a new approach to automate neural network design tailored to various tasks and hardware setups. 🔗: https://www.liquid.ai/research/automated-architecture-synthesis-via-targeted-evolution

1 reply

0 recast

10 reactions

𝚐𝔪𝟾𝚡𝚡𝟾 pfp

𝚐𝔪𝟾𝚡𝚡𝟾

EfficientTAMs are lightweight models for video object segmentation that improve efficiency and reduce computational demands compared to SAM 2. Using a vanilla ViT and an optimized memory module, they achieve comparable performance with significantly reduced size and faster processing (~2x on GPUs, ~10 FPS on mobile devices such as iPhone 15 Pro Max). don’t sleep (Meta AI) 🔗: https://yformer.github.io/efficient-track-anything/

0 reply

0 recast

6 reactions

𝚐𝔪𝟾𝚡𝚡𝟾 pfp

𝚐𝔪𝟾𝚡𝚡𝟾

OLMo 2 7B and 13B models, trained on 4-5T tokens, w/ fully open weights, data, and code. The models feature an improved architecture for better training stability, a staged training process incorporating the new Dolmino dataset during annealing, and SOTA OLMo 2 Instruct models. Another banger from Allen AI. https://allenai.org/blog/olmo2

0 reply

5 recasts

24 reactions

𝚐𝔪𝟾𝚡𝚡𝟾 pfp

𝚐𝔪𝟾𝚡𝚡𝟾

The Tulu 3 release—this is how you do open-source!

0 reply

0 recast

5 reactions

𝚐𝔪𝟾𝚡𝚡𝟾 pfp

𝚐𝔪𝟾𝚡𝚡𝟾

QwQ-32B-Preview is a causal language model designed by the Qwen Team to advance AI reasoning capabilities. QwQ-32B-Preview: > 32.5B parameters, 64 layers, and a context length of 32,768 tokens. > Architecture: transformers w/ RoPE, SwiGLU, RMSNorm, and Attention QKV bias. > Performs well in math and coding but struggles with language mixing, reasoning loops, and nuanced understanding. https://huggingface.co/Qwen/QwQ-32B-Preview

0 reply

0 recast

6 reactions

𝚐𝔪𝟾𝚡𝚡𝟾 pfp

𝚐𝔪𝟾𝚡𝚡𝟾

DeMo: Decoupled Momentum Optimization code: github.com/bloc97/DeMo paper: arxiv.org/abs/2411.19870

0 reply

4 recasts

23 reactions

David (d/acc) pfp

@promptrotator.eth

Publicly verifiable agents with immutable code have a lot of prosocial usecases, i.e. an AI agent that monitors OSINT sources for missing persons 24/7. Let’s you skip the bureaucracy and directly fund the things you want to see in the world.

3 replies

0 recast

11 reactions

𝚐𝔪𝟾𝚡𝚡𝟾 pfp

𝚐𝔪𝟾𝚡𝚡𝟾

INTELLECT-1, the first 10B model trained via a decentralized framework, is now officially released. The release includes the base model, checkpoints, post-trained models, datasets, technical report, and training framework. Another w for open source! This demonstrates the potential of decentralized deep learning to collaboratively build large-scale models. https://www.primeintellect.ai/blog/intellect-1-release

2 replies

9 recasts

46 reactions

Kazi pfp

the gc decided to troll bluesky normies by offering their 'data set' back with a fee lol

2 replies

1 recast

9 reactions