Stampli cuts launch work by 68% with ChatGPT Work and Codex
A vendor-published case study shows how an accounts payable firm used OpenAI's tools to compress weeks of launch production into days.
AI & ML
Frontier models, research results, agents, and developer tooling.
Liquid AI's d1 models answer without generating text
New open-weight models from Liquid AI return calibrated answers in a single forward pass instead of writing output tokens.
AI Lifts Patent Work Quality but Skill Gains Depend on Seniority
A three-month field experiment shows AI improves output across the board, yet only senior lawyers build lasting judgment, while juniors see no average skill gain.
OpenAI adds College Planner and study tools to ChatGPT for Teens
The new features aim to make ChatGPT for Teens a practical study and college-application companion.
NVIDIA Nemotron Models Hit Gold at Two Olympiads
Fine-tuned Nemotron models achieved gold-level performance at both the International Olympiad in Informatics and the International Mathematical Olympiad, signaling a leap in specialized reasoning.
Kubernetes creators bring cloud-native harness to coding agents
Stacklok's Mecatl aims to make agent harnesses reliable and governable by moving them from the desktop to the cloud.
Radisson Hotel Group launches ChatGPT plugin for hotel discovery
The partnership with Accenture brings hotel search and booking directly into ChatGPT, making travel planning more conversational.
Hugging Face releases open d1 decision models for edge inference
LiquidAI's d1-3B and d1-omni-600M bring fast, multimodal decision-making to edge devices, with d1-3B topping the Decision Index under 10B parameters.
Quantization-Aware Healing Produces 4-Bit Model That Beats Original
A new compression technique achieves a 4-bit model that outperforms the full-precision version it was derived from.
Laya: A Developer’s Guide to Zero-Shot Decisions and Calibration
A new guide walks developers through Laya, an open-source zero-shot decision engine, using banking data to show typed decisions, temperature fitting, and abstention gates.
LFM2.5-DSpark Speeds Up Inference Up to 3.2x
A new Liquid AI model variant uses sparse attention to cut inference latency dramatically, according to Hugging Face benchmarks.
OpenAI Launches Intelligence Age Blog on AI's Societal Impact
OpenAI's new blog will examine how transformative AI could reshape power, governance, the economy, and individual freedom.
Hugging Face Examines Benchmark Optimization in Speech Recognition
A Hugging Face analysis looks at how speech-recognition benchmarks can be optimized, and what that means for evaluating models.
Build a Robust RAG System on a Standard Laptop
A practical guide shows how retrieval-augmented generation can be designed, assembled, and tuned locally without cloud infrastructure.
Hugging Face's Infrastructure Behind Papers with Code Search
The blog explains how Hugging Face's managed services handle the compute and storage needs of a semantic search engine for research papers.
DeepMind Moves From Atari to EVE Online in 15-Year Game AI Push
DeepMind is extending its game-playing AI research beyond classic benchmarks by partnering with studios to prototype new gameplay experiences.
Import AI 470 covers machine rights, SPADE, and Hawkeye
Jack Clark's newsletter surveys three unrelated AI topics: policy on machine rights, SPADE for environment generation, and Hawkeye for GPU kernel development.
Hybrid Pipelines: Adding Agentic AI to Classical ML
A practical walkthrough shows how classical machine learning pipelines can be extended with agentic AI to handle unstructured inputs and automate decision-making.