Other AI

On May 12, 2026, Celonis SE announced it has acquired Ikigai Labs Inc., an MIT‑linked decision‑intelligence startup, to strengthen its software portfolio and deliver operational data and context to enterprise AI.

Kavish Sathia unveiled Gauntlet at the Agent Builder Hackathon, a top-three project built in the final 48 hours that inserts a mocking agent between an autonomous agent and its tools to intercept, mutate and record tool

Nicole Replogle’s May 15, 2026 analysis finds that differences in billing units and runtime behavior make Zapier typically less expensive than Make for AI‑powered automations, even though Make offers lower entry pricing.

Supertone released Supertonic 3, the third generation of its on‑device, ONNX‑based text‑to‑speech engine.

Closing arguments ended after a combative final week that centered on the credibility of Elon Musk and OpenAI CEO Sam Altman; jurors begin deliberations Monday and an advisory verdict is expected as soon as next week.

Zyphra published ZAYA1-8B-Diffusion-Preview, converting an autoregressive mixture — of-experts LLM into a discrete diffusion model that preserves evaluation quality and reports up to 7.

Anyscale released an Agent Skill, /anyscale — workload-llm-post-training, that recommends post‑training methods (SFT, CPT, DPO/KTO/ORPO/SimPO, PPO, GRPO/DAPO), emits LLaMA framework configs (Factory, SkyRL, Ray Train),

On May 15, 2026 the Ray monitoring suite gained Cluster and Actor Dashboards, completing a fully persisted Ray Dashboard rollout and enabling historical queries of Ray events after clusters terminate.

A federal judge delayed final approval of Anthropic’s $1.5 billion settlement over alleged use of books in AI training after authors objected to distribution terms and a roughly $320 million fee request;

OpenAI has formalized a product reorganization that places cofounder Greg Brockman in charge of a unified product group covering ChatGPT, Codex and developer — facing APIs.

A Linux kernel information — disclosure bug (CVE-2026-46333, dubbed ssh-keysign-pwn) can let unprivileged processes read SSH host private keys and the /etc/shadow password file.

Google published documentation — highlighted in a May 15, 2026 report by Matthias Bastian — rejecting the need for separate "Generative Engine Optimization" (GEO) or "Answer Engine Optimization" (AEO).

Thomas Dietterich, an arXiv moderator, posted that submissions with incontrovertible evidence authors did not check LLM-generated output will trigger a one-year submission ban and extra conditions on future postings;

A tester found four practical changes — using a USB connection, a high-speed data cable, closing background apps, and disabling battery optimization for Android Auto-produced noticeable improvements in app launch,

OpenAI has started a paid preview that lets US ChatGPT Pro subscribers link bank accounts via Plaid to sync real transactions into a queryable spending dashboard. Finance conversations default to the GPT‑5.

On May 13, 2026, Ben Lyso published a practical roundup of eight model context protocol (MCP) servers aimed at builders who want a single, reusable bridge between large language models and their software stacks.

NASA’s Jet Propulsion Laboratory has started laboratory and environmental trials of a palm‑sized, radiation‑hardened multicore processor intended to give deep‑space probes orders‑of‑magnitude more onboard compute

Julie Qiu outlined five AI roles — Archaeologist, Experimenter, Critic, Author, Reviewer — that act as cognitive "RAM" for engineering leaders, helping synthesize legacy context, pressure‑test designs, and accelerate

Anthropic has added a /review — contract skill to Claude for Small Business inside Claude Cowork, enabling fast contract analysis in the Claude app for users on the Claude Pro tier ($20/month).

Studios in China are producing entire ultrashort streaming dramas end-to-end with generative AI, releasing content at high volume and slashing production cycles and budgets — changes that are remaking economics, labor

PHAROS is a four‑agent prototype introduced May 13, 2026 that automates end‑to‑end pharmacovigilance: It ingests adverse‑event reports from the FDA FAERS API, runs WHO‑standard signal detection (proportional reporting

On May 14, 2026 Anthropic published a policy paper arguing that access to advanced AI compute is the decisive factor in U.S.

A benchmark roundup shows Claude Code at 87.6% on SWE‑bench Verified and GPT‑5.5 leading Terminal‑Bench at 82.

Mira Murati, who left her role as OpenAI chief technology officer in 2024 to co‑found Thinking Machines Lab, previewed new “interaction models” that use camera and microphone inputs to interpret pauses, interruptions