ethicalabs.ai

This paper is making the rounds: arxiv.org/abs/2506.21734

A tiny (27M) brain-inspired model trained just on 1000 samples outperforming o3-mini-high on reasoning tasks.

#MLSky 🧠🤖

GitHub - ethicalabs-ai/completionist: Command-line tools for Synthetic Datasets Generation Command-line tools for Synthetic Datasets Generation - ethicalabs-ai/completionist

Introducing Completionist, an open-source command-line tool that automates synthetic text dataset generation.

👉 Check out Completionist on #GitHub: github.com/ethicalabs-a...

#LLMs #GenerativeAI #DataEngineering #FineTuning #OpenSource #Python #SyntheticData #RAG

Building and Sharing a Multimodal ViT Model for Skin Lesion Analysis From Proof of Concept to Huggingface App 🤗

Building and Sharing a Multimodal ViT Model for Skin Lesion Analysis: From Proof of Concept to Hugging Face App 🤗 #huggingface #MLSky #opensource hashtag#python #vit #transformers #medicalai #visionmodel #skincancer

medium.com/@massimo.sca...

Kurtis-E1.1: Supervised Fine-tuning of Qwen2.5-3B-Instruct with Flower.ai & Hugging Face A Blog post by Massimo Roberto Scamarcia on Hugging Face

Rather than chasing benchmark supremacy or scaling wars, Kurtis E1.1 focuses on understanding, sustainability, and practical impact, especially in areas like mental health support and safer human-AI interaction

huggingface.co/blog/mrs83/k...

#MLSky #EthicalAI #LLM

To developers: Build opt-in systems.
To policymakers: Legislate data transparency.
To artists: Unionize.
To users: Demand ethical tools.

#EthicalAI #MLSky

YouTube video by Massimo Scamarcia Kurtis-E1-MLX-Voice-Agent

Just built an offline voice assistant for macOS:
🎤 Whisper STT (MLX)
🧠 LLM via #Ollama
🗣️ XTTSv2 TTS
🌍 Optional translation
No cloud. No tracking. No vibe coding — all handcrafted.
Demo here 🎥 www.youtube.com/watch?v=8-1P...

#OnDeviceAI #LLM #Privacy #TTS #STT

Testing Kurtis E1 Can a Small, Fine-Tuned Model Generalize Beyond Its Training Scope?

Testing Kurtis E1 beyond its training scope—AI ethics, decentralization, even philosophy. No hallucinations, just structured reasoning. Is this emergent? You decide.

Read more: medium.com/@massimo.sca... #LLM #AI #EthicalAI

AI can now model and design the genetic code for all domains of life with Evo 2 | Arc Institute Arc Institute develops the largest AI model for biology to date in collaboration with NVIDIA, bringing together Stanford University, UC Berkeley, and UC San Francisco researchers

🧪 The Arc Institute's Evo2 models DNA like an LLM models language, predicting mutations, gene function, and evolutionary signals. With 40B parameters trained on 128K genomes, it hints at AI-driven biological discovery. 🧬💻 #MLSky
Link to the paper: https://arcinstitute.org/manuscripts/Evo2

🌀 Ouroboros: Recursive LLM Refinement with Small Models From Synthetic Data to Recursive Self-Refinement: An On-Device Experiment

🌀 Ouroboros: Small models drive recursive #LLM self-refinement for synthetic datasets generation. On-device AI shaping smarter futures! #EdgeAI #OpenSourceWeek #DeepSeekR1 #Ollama medium.com/@massimo.sca...

🚀 NVIDIA Minitron: Efficient LLM Compression!

arxiv.org/pdf/2408.11796

Minitron uses pruning + distillation to create smaller, high-performance models

🔑 Highlights:
- Teacher Correction: Adapts models to new data
- Structured #Pruning: 2.7x faster inference
- #Distillation: Uses 40x fewer tokens

OpenAI scrubs diversity commitment web page from its site

OpenAI has eliminated a page on its website that used to express its commitment to diversity, equity, and inclusion. The URL “https://openai.com/commitment-to-dei/” now redirects to “https://openai.com/building-dynamic-te…

#ai #news #openai

This Nashville Startup Is Protecting Kids Online with Smarter, Safer AI - Hypepotamus Nashville-based Angel Q (previously known as Angel Kids AI), got its start building a safer browser option for kids to access the internet. But browsers are not the only way of searching for […]

Arcee AI and AngelQ just launched KidRails for hashtag #LLMs — an open-source framework for safe, age-appropriate #AI responses for children

hypepotamus.com/startup-news...

Setting new standards in security, transparency, and responsibility 🌸 #EthicalAI #ML

They Said It Couldn’t Be Done A Blog post by Pierre-Carl Langlais on Hugging Face

Pleias is a large language model trained exclusively on open data. It was developed using the Common Corpus, a dataset that addresses the need for high-quality compliant training data in AI development. huggingface.co/blog/Pclangl...

#opensourcellm #opendata #commoncorpus #llm #ai #ml

Active Inheritance, A Smarter Way to Train Models with Synthetic Data The practice of fine-tuning models on synthetic data is becoming well established. But synthetic training data, even if it represents the training...

A naive way to generate synthetic fine-tuning data is to feed prompts to a model, collect its output, and use that as the fine-tuning set. Synthetic data is cheap, so we can afford to be more choosy. By generating responses to each prompt, we can select the one that best suits our purposes. #AI #ML