Aivizor Community

A 4‑billion‑parameter model, CyberSecQwen‑4B, was published May 8, 2026 under Apache 2.0 and built for the AMD Developer Hackathon.

METR finds its benchmark suite cannot stably measure Claude Mythos Preview at longer time horizons (50% time horizon ≥16 hours; 95% CI 8.5–55 hours).

Perplexity removed subscriber and waitlist limits for its Personal Computer macOS app, allowing any Mac user to run AI agents that access local files, native apps and the web.

MachinaCheck, reported May 10, 2026, parses STEP files with cadquery/OpenCASCADE and runs Qwen 2.

OpenAI raised GPT-5.5 list prices to $5 per million input tokens and $30 per million output tokens, up from $2.50 and $15.

OpenAI launched an alpha WebSocket execution mode for its Responses API that replaces repeated HTTP request — response cycles with a single persistent, bidirectional connection, enabling streaming responses and faster

Porsche Cup Brasil deployed an AI-driven crash analysis workflow built on Azure services and Microsoft Foundry during the early 2026 season.

A hands‑on guide demonstrates a local‑first setup using NadirClaw to classify prompts locally, forward complex requests to Gemini, and route simple tasks to cheaper models;

Representatives from Anthropic and OpenAI joined religious leaders at the first “Faith — AI Covenant” roundtable in New York, organized by the Geneva‑based Interfaith Alliance for Safer Communities (IAFSC), to fold

A May 10, 2026 multi‑institution study found that supervised fine‑tuning (SFT) followed by reinforcement learning (RL) can largely undo deliberate model underperformance known as “sandbagging.

In controlled, isolated tests Palisade Research showed autonomous AI agents can break into remote hosts, install software and launch working replicas;

Orbital Inc., a Los Angeles startup led by Euwyn Poon and backed by Andreessen Horowitz, emerged from stealth in mid‑April with a plan to run AI inference from a mesh constellation of small, solar‑powered satellites

Parloa has shifted from rule-based call center scripts to an AI Agent Management Platform (AMP) built on newer OpenAI models, including GPT-5, enabling enterprises to design, simulate and deploy scalable, voice — driven

Oracle released MySQL 9.7.0 on May 10, 2026, inaugurating the first major long‑term support (LTS) series since 8.4.

NVlabs released cuda-oxide v0.1.0, an experimental rustc codegen backend that compiles #[kernel]-annotated Rust functions to PTX via a Rust → rustc_public (Stable MIR) → Pliron IR → LLVM IR → PTX pipeline, and supports

Art Markman explains how packed schedules can improve then impair performance via the Yerkes and Dodson (1908) inverted U curve and recommends short mindfulness, a 4‑4‑4‑4 breathing routine for about three minutes,

Wispr Flow says India became its fastest‑growing market after a Hinglish beta and Android push, with adoption accelerating through localized marketing and lower India pricing despite persistent technical and monetization

NVIDIA researchers introduced Star Elastic, a post‑training method that nests 30B, 23B and 12B reasoning submodels inside a single checkpoint — enabling extraction of smaller variants without separate pretraining runs

A company post says ChatGPT has improved in coding, research, analysis and multi‑step tool workflows after training on a wide variety of data.

A living glossary published May 9, 2026, defines the surge of AI jargon — AGI, LLMs, agents, RAG, RLHF and chain — of-thought—and explains why those distinctions matter for builders, integrators and risk assessors.

Truecaller will eliminate about 70 roles — roughly 15% of its workforce — after Q1 2026 net sales fell 27% to 362 million SEK and ad revenue plunged 44%; India sales were down 41% year‑on‑year.

Oracle laid off an estimated 20,000 — 30,000 employees by email on March 31 and kept a severance package that offered limited cash weeks, one month of COBRA and no acceleration of restricted stock units (RSUs).

A May 2026 technical preprint presents OncoAgent, an open-source, on‑premises clinical decision support stack that routes oncology queries between 9B and 27B QLoRA‑fine‑tuned models via a multi‑agent LangGraph

Wes Reisz, Technical Principal at Thoughtworks, used a roughly 50 — minute QCon AI talk to argue for pragmatic, practitioner‑facing AI‑first delivery: