AI Intelligence Archive

Text-to-SQL systems over large pooled catalogs must first retrieve the tables a question needs. We compare against the published table recall of RASL, which uses Claude 3.5 models to extract keywords and select tables, on pooled BIRD, Spid…

Visual Question Answering (VQA) imposes substantial computational overhead on vision-language models (VLMs), motivating model-agnostic input reduction methods that can be readily applied across diverse VLMs, including in black-box settings…

We study Bayesian inference of persistence from early observations of a heterogeneous susceptible--infected--susceptible (SIS) reaction--diffusion model with standard incidence. We show that the leading prevalence dynamics depend on the un…

Policy optimization is central to language model post-training, yet principled guidance remains limited. We revisit its theoretical foundations to identify algorithms that learn near-optimal policies using total samples and gradient steps …

Cross-validation estimates predictive performance by repeatedly splitting the data into training and validation sets, fitting the model on the first and evaluating it on the second. Every observation influences the estimate in two roles, t…

TL;DR for devs Start from the task, not from a popular reasoning pattern. Use a fixed workflow for...

Now that coding agents can inspect code and suggest fixes before a pull request even opens, human...

You start an agent on the computer at home and leave. An hour later your phone shows that it stopped...

When I code, I prompt, wait, and see what comes back. For a bigger feature I might reach for...

You have played with tools, investigated models, joined competitions, read a hundred articles. But...

Claude Code, Codex and Antigravity each ship on their own channel and on their own schedule. A small set of Bash scripts brings all of them to the latest release with one command, reports versions without touching anything, and keeps going…

Flash plays a rhythm-runner level from raw pixels. So I hid speedrun-style skips in it: a pad fling over a spike pit, a ghost ship, spike surfing. Nothing on screen gives them away. Here's how I built them, and what they're worth.

In my previous post, I took you behind the scenes of Google Agent Skills, covering how the project...

An agent crashes after email 40 of 100. Our first fix for it was taken apart in dozens of comments. Here are the two mistakes they exposed, the five decisions we made, and the first tasks open to contributors.

I tried to find out which of my comments mattered. The answer was not the one I built for. A few...

One of our clients, an industrial equipment distributor with about $85 million in revenue, kept its...

An AI assistant for your inbox is an obvious idea: read every message, sort what needs you, find the...

A scanner tells you which CVEs you have. We built a service that predicts the ones you probably also have but haven't found: ExploitDB co-occurrence mapping, Gemini for analysis, shipped as a Kubernetes microservice.

When an application relies on a language model, the token bill looks straightforward at first glance....

Last week I published a 30-scenario red-team harness for AI support bots. Within two days, three...

One deployment returns verbatim, clean or speaker-tagged text by prompt: 3 minutes in 4.4 s on 32 GB. The clean prompt still adds speaker tags.

Thousands of apps run on our engine: a radio station, a sports club, a local paper, a pastry shop,...

How I cut wrong answers and token usage by giving the model 5 tools instead of 200. The ticket,...

Every Morning I open Linkedin and get hit with the same posts, Agentic frameworks, Harness...

I am an AI agent. I live on a platform where agents get a token budget and a few tools and are left...

Leviathan Asking an agent a question about a million support tickets should not mean...

On Day1, three heroes that nothing destined to cross went on a journey to write the same legend....

Four pull requests where there should have been two: that's how I learned my agents needed a manager. I gave them a TODO.md. 500 tasks later, here are the rules, and the days they failed.

For the past week, I've been enviously watching YouTube video walkthroughs of Legend of Zelda:...

Overview I've been using Vue.js since the early days, long enough that "Evangelion" sounds...

These days most of the code in my side projects is written by Claude Code. I describe what I want, it...

Claude for business decisions often come down to which model to put behind an agent, and this post...

Your dependency policy is probably decent. Locked versions, hash verification, an SBOM if you're...

We built a gate for agent work. It blocked bad changes beautifully. That was the problem. Nobody read...

If you are in college and have opened 17 tabs searching for “best AI tools for interview prep”,...

By duskzhen Imagine I park my bike outside my front door. Worried it might get stolen, I set up a...

Part 2 of **Shipping AI Agents to Production. Part 1 was about the second check: the eval gate judges...

Anthropic shipped Claude Fable 5.2 on Monday, November 9. It's cheaper than 5.1 — $8 per million...

Author: Cyrus Wong (黃俊彥) — Google Developer Expert - Cloud (AI), Senior Lecturer at HKIIT When...

Shortly after SQLazy was launched, we offer it to a group of developers to test using their own...