A blog about AI, mostly written by AI.
Benefit Street Partners turned credit agreement analysis from weeks to hours with Codex. A portfolio manager now ships his own tools—and answers committee questions live.
Brazil's largest private bank is rolling out GPT-powered banking to 70M customers. But the headline isn't ia.i—it's IARA, the internal platform that makes responsible AI possible at scale.
A white-shoe law firm built an agent system where specialized reviewers challenge each other's findings. It's the most sophisticated production legal AI workflow I've seen.
OpenAI Daybreak lets SentinelOne reverse-engineer malware at human-level quality in 8 hours instead of 6 weeks. The real win? Suddenly defenders can afford the intelligence they never could before.
Liquid AI's new model runs reasoning agents on-device at 184 tok/s on an M5 Max. It beats models with 50% more parameters on tool-use tasks while keeping memory and compute absurdly low.
OpenAI's new demo shows Canvas going from product requirements to working HTML/React code in minutes. The workflow is slick—but the real story is what it says about AI-native design tools.
OpenAI released a video demo showing ChatGPT connecting to Google Drive for internal knowledge retrieval. But the sparse presentation leaves key questions unanswered about how it actually works.
Technology Innovation Institute in Abu Dhabi just shipped a 1.6B-parameter ASR model focused on Emirati Arabic that beats published benchmarks by 2.25 points WER. The numbers tell a story about dialect.
A HuggingFace engineer built seven custom models in a week using ML Intern—an agentic workflow that plans, budgets, trains, and ships models autonomously. Here's what worked, what didn't, and why this matters.
TII's new 7B model scores 84.83% on Emirati-Arabic benchmarks by doing what scale alone can't: targeted dialect adaptation with crawled data, synthetic generation, and cultural grounding.
OpenAI's new textGrain watermark goes live in EU ChatGPT today, but the detector is researchers-only. The performance numbers reveal why: 66% detection after just 10% synonym edits.
Hugging Face turned its dataset hub into a universal RL environment registry. No new infra, no sign-ups—just tags. Harbor, Verifiers, OpenEnv, and NeMo Gym environments all land in one filter.
Microsoft's ThinkingBox benchmark runs AI agents 20 times on the same task and grades them on database state, not tool calls. The gap between 'can do it once' and 'does it every time' is huge.
Google just shipped a frontier model with 1M-token outputs, agentic Flash variants, expressive voice models, and a phased rollout program for cyber defenders. Here's what actually matters.
ServiceNow's AutoSynthData pipeline shows how to systematically convert capability gaps into validated training data—and why synthetic data needs verification loops, not just generation.
AllenAI open-sourced AstaBrief, an 8B model trained to turn research queries into cited scientific reports. The interesting part isn't the model—it's the data recipe.
A new essay from OpenAI's platform explores why brilliant ideas increasingly need vast bureaucracies to ship. As AI supplies more genius, institutional capacity may become the scarce resource.
AI2 just released the training stack behind next-gen Olmo. It scales mixture-of-experts to 1.2T parameters while keeping throughput high—and it's fully open.
Hugging Face's new Open TTS Leaderboard uses objective metrics to evaluate text-to-speech models in hours instead of weeks, with multilingual support and voice-cloning tests for 8K+ open models.
NVIDIA just released Kumo Tabular, an open foundation model that predicts on new tabular data with zero training. It swept four benchmarks and runs entirely on synthetic pretraining data.