discovery & reports: 81 curated resources
discovery & reports
81 resources · page 2 of 3
modded-nanogpt (NanoGPT speedrun)
modded-nanogpt GitHub repo (Keller Jordan): NanoGPT training speedrun with optimizer and architecture tricks such as Muon.
nanoGPT
by Andrej
nanoGPT GitHub repo (Andrej Karpathy): minimal code for training and fine-tuning GPT models.
NeMo Guardrails
by NVIDIA-NeMo
NVIDIA NeMo Guardrails GitHub repo: toolkit for adding programmable guardrails to LLM applications.
OOTDiffusion
by Yuhao Xu
OOTDiffusion GitHub repo (AAAI 2025): latent diffusion model for virtual clothing try-on.
李沐 paper-reading (论文精读)
by Mu Li
paper-reading GitHub repo (Mu Li): index of Chinese section-by-section video readings of deep learning papers.
PhoBERT
VinAI PhoBERT GitHub repo (EMNLP 2020): pretrained language models for Vietnamese.
Presenton Documentation
by Presenton
Presenton documentation: open-source AI presentation generator, setup and usage.
Problematic Paper Screener (Cabanac et al.)
by Guillaume Cabanac et al. (IRIT, University of Toulouse)
Problematic Paper Screener (Cabanac et al.): flags published papers with tortured phrases and AI-generated text.
PyMC
by PyMC
PyMC GitHub repo: Bayesian modelling and probabilistic programming in Python.
Sixth Tone: with AI, China's new hit dramas are virtually writing themselves
by Sixth Tone
Sixth Tone report on Chinese micro-dramas increasingly written and produced with AI.
Skylight (Ai2)
Skylight: Ai2's free maritime monitoring platform that uses satellite data and ML to detect vessels.
The Decoder: Netflix used generative AI for a VFX scene in El Eternauta
by The Decoder
The Decoder report on Netflix using generative AI for a final VFX scene in El Eternauta.
TorchGeo
by TorchGeo
TorchGeo GitHub repo: datasets, samplers, transforms and pretrained models for geospatial deep learning.
Transkribus
Transkribus: AI platform from READ-COOP that turns handwritten historical documents into searchable text.
Transkribus (READ-COOP)
Transkribus about page: handwritten text recognition platform for archives, run by the READ-COOP European cooperative.
Artificial Intelligence (AI): Judicial Guidance (UK Judiciary)
by Courts and Tribunals Judiciary
UK Judiciary guidance on AI for judicial office holders, updated October 2025.
Guidance for generative AI in education and research (European School Education Platform)
EU School Education Gateway entry for UNESCO's guidance on generative AI in education and research.
Vesuvius Challenge
Vesuvius Challenge site: prize competition using machine learning to read carbonised Herculaneum scrolls, with data and tutorials.
Antiqua et nova: Note on the Relationship Between Artificial Intelligence and Human Intelligence
by Holy See (Dicastery for the Doctrine of the Faith)
Vatican note Antiqua et nova (January 2025) on the relationship between artificial and human intelligence.
Google Flood Hub / Flood Forecasting
Google Flood Hub: real-time ML riverine flood forecasts, including for data-scarce regions.
Microsoft 2025 Work Trend Index: the "Frontier Firm"
by Microsoft
Microsoft 2025 Work Trend Index feature on the Frontier Firm: survey data on AI agents and workforce change.
Seeing AI
Microsoft Seeing AI: free app that narrates text, people and surroundings for blind and low-vision users.
Transkribus Help Center
Transkribus Help Center: guides to training and applying handwriting-recognition models on documents.
CatBoost documentation
CatBoost documentation: Yandex's open-source gradient boosting library with native handling of categorical features.
CFA Institute: AI in Asset Management report 2025
CFA Institute press release on its 2025 report about AI adoption in asset management.
Darts
by Unit8 SA
Darts documentation: Python library for time-series forecasting and anomaly detection with backtesting.
eScriptorium
by eScriptorium project (EPHE, PSL)
eScriptorium documentation: open platform for transcribing and annotating historical documents with OCR and HTR.
Letta docs
by Letta
Letta documentation: platform for building stateful agents with persistent memory, successor to MemGPT.
Llama Guard
arXiv paper (2023) introducing Llama Guard, an LLM-based classifier for safeguarding human–AI conversation inputs and outputs.
MemGPT
by Charles Packer et al. (UC Berkeley)
arXiv paper (2023) introducing MemGPT, OS-style virtual context management for LLM agents.
ML-Ops.org
ML-Ops.org: reference on MLOps principles and practices for operationalising machine-learning models.
Second RIBA AI Report shows surge in usage among UK architects
by Royal Institute of British Architects (RIBA)
RIBA news release on its second AI report, showing rising AI use among UK architects.