The LGTM (Page 37)

The LGTM

Sign in Subscribe

Gemini’s Workspace Outage Was a Reminder: AI Tools Need SRE Like Everything Else

Gemini’s Workspace Outage Was a Reminder: AI Tools Need SRE Like Everything Else

Gemini’s Workspace disruption was not the loudest Google AI story this week, but it may be the most useful one for anyone building agentic software. The interesting detail is not that a chatbot had errors. The interesting detail is Google’s root cause: a backend database performance issue affected

Gemini in Chrome Is Expanding Globally. The Browser Is Becoming the Agent Runtime

Gemini in Chrome Is Expanding Globally. The Browser Is Becoming the Agent Runtime

Google framed Gemini in Chrome’s latest move as a regional expansion: desktop and iOS availability across Latin America, Africa, the Middle East, and more supported regions. That is the distribution headline. The engineering headline is sharper: Chrome is becoming a high-permission AI runtime, and the web is about to

Gemini Wants to Run the Front Desk for Small Businesses

Gemini’s new small-business features are not flashy in the usual AI-demo sense. Nobody is asking a model to solve graduate math or write a startup pitch in the voice of a pirate. Google is doing something more commercially important: putting Gemini next to the public listing, customer reviews, questions,

Google’s AI Scam Lawsuit Is the Security Story Hiding Inside Gemini’s Growth

Google’s AI Scam Lawsuit Is the Security Story Hiding Inside Gemini’s Growth

Google’s lawsuit against the group it calls “Outsider Enterprise” is easy to file under consumer scam enforcement and move on. That would miss the useful part. The story is not that criminals discovered phishing. The story is that the same AI productivity loop companies are adopting — generate variants, ship

On-Policy Distillation Looks Dense on Paper, but the Parameter Updates Are Sparse Where It Counts

On-Policy Distillation Looks Dense on Paper, but the Parameter Updates Are Sparse Where It Counts

On-policy distillation sounds like it should rewrite models broadly. The student generates its own trajectories, the teacher provides dense token-level feedback, and the optimizer gets a lot more signal than sparse reward RL. The surprise in this paper is that the final parameter updates still look selective. “Dense Supervision, Sparse

EvoArena Shows Why Agent Memory Needs Git-Style History, Not Just a Bigger Scratchpad

EvoArena Shows Why Agent Memory Needs Git-Style History, Not Just a Bigger Scratchpad

Agent memory keeps getting described as a bigger notebook. EvoArena makes the better argument: it should look more like git history. The benchmark suite evaluates agents in environments that change over time: terminal workflows, software repositories, and user preferences. Its paired method, EvoMem, wraps ordinary memory systems with append-only patch

RA-RFT Says Retrieval for Reasoning Should Find Analogies, Not Keyword Neighbors

RA-RFT Says Retrieval for Reasoning Should Find Analogies, Not Keyword Neighbors

Retrieval-augmented generation has a bad habit: it assumes the useful thing is the thing with the most similar words. RA-RFT is a reminder that reasoning does not work that way. The paper, from researchers affiliated with Meta Superintelligence Labs and Rice University, proposes Retrieval-Augmented Reinforcement Fine-Tuning: a post-training method that

OpenClaw’s Default-Model Doctor PR Is the Gemini Migration Story in Miniature

The ugliest AI-tooling failure is not “the model is unavailable.” It is “the tool says this dead model is your default, then tells you the model does not exist, then fails your agent run anyway.” That is not an error message. That is a trust withdrawal. OpenClaw PR #92292 addresses