KitchenAid Keeps Its Iconic Stand Mixer Nearly Unchanged Since 1914 — wired.com

Meta Scraps Token-Driven Performance Reviews as Employees Test New AI Agent Hatch — wired.com

Palo Alto Networks Acquires AI Help Desk Startup Console for $500 Million — techcrunch.com

**TechCrunch Disrupt 2026 Debuts Real World AI Stage** — techcrunch.com

U.S. Judge Spares Google’s Ad Empire From Breakup But Orders Operational Changes — techcrunch.com

KitchenAid Keeps Its Iconic Stand Mixer Nearly Unchanged Since 1914 — wired.com

Meta Scraps Token-Driven Performance Reviews as Employees Test New AI Agent Hatch — wired.com

Palo Alto Networks Acquires AI Help Desk Startup Console for $500 Million — techcrunch.com

**TechCrunch Disrupt 2026 Debuts Real World AI Stage** — techcrunch.com

U.S. Judge Spares Google’s Ad Empire From Breakup But Orders Operational Changes — techcrunch.com

Google drops Gemini 3.8 Flash

Google releases Gemini 3.8 Flash and 3.8 Flash Cyber, opens Fairwind program for trusted defenders

Google DeepMind released Gemini 3.8 Flash, its most intelligent Flash model yet, with 1M token context, improved reasoning, and enhanced coding for autonomous agents. A restricted Cyber variant targets autonomous vulnerability discovery and automated patching, available through the new Fairwind Program for national cyber authorities and critical infrastructure providers. Pricing holds at $0.75/$3.75 per 1M tokens through December 2026.

Google · 2026-09-02 · Read the dispatch →

DeepSeek releases V4 Pro 0813 alongside fully open-source evaluation harness with full reproducibility logging

DeepLearningAI · 2026-09-02

Wonderful raises $550M at $5B valuation for enterprise AI operating system, doubling valuation in under 6 months

TechCrunch · 2026-09-02

HiddenLayer nabs $100M Series B for AI model security as ARR grows 10x YoY

TechCrunch · 2026-09-02

Research & Papers

Post-training language models achieve gold-medal performance in coding competitions

An end-to-end pipeline combining large-scale problem curation, synthetic reasoning traces, and super-specialization achieves gold-medal performance at IOI/ICPC level.

arXiv · 2026-09-02

Discriminative world models outperform generative prediction for web agent planning

Proposes discriminative world models that rank predicted web states from candidate actions, outperforming generative next-state prediction for test-time web agent planning.

arXiv · 2026-09-02

From reweighting to rewriting: unlocking intervention effects of influential training samples

Shows rewriting — not just reweighting — is needed for effective training data attribution; the proposed method outperforms influence functions on intervention tasks.

arXiv · 2026-09-02

Dutch books for language models: LLM probability forecasts violate coherence axioms

Shows that LLMs' probabilistic forecasts for life decisions violate coherence axioms (Dutch books), undermining trust in their world model consistency.

arXiv · 2026-09-02

ByteDance Seed publishes research on self-evolving agent harnesses that score on harness quality

ByteDance Seed published a paper on self-evolving agent harnesses that score on harness quality rather than just task completion, matching or exceeding human baselines in writing and ML experimentation.

omarsar0 · 2026-09-02

Hardware & Robotics

New US export controls target Chinese access to remote AI servers in Thailand and Singapore

The Trump administration considers closing a loophole allowing China to access advanced AI compute via remote servers, targeting Moonshot's Kimi K3 model trained via Nvidia-equipped infrastructure.

Tom's Hardware · 2026-09-02

Nvidia's DLSS 5 launches today with NBA2K27 across all RTX 50 Series GPUs

Nvidia confirmed DLSS 5 (3D-guided Neural Rendering) launches September 3 starting at 9 pm PT, available across all RTX 50 Series GPUs, laptops, and GeForce NOW.

Tom's Hardware · 2026-09-02

Acer showcases RTX Spark compact desktop with 1 petaflop AI performance at IFA 2026

Acer unveiled the SFF RTX Spark compact desktop powered by the NVIDIA RTX Spark superchip featuring up to 1 petaflop AI performance, 128 GB unified memory, and a 6,144-core Blackwell GPU for local agentic AI workloads.

TechPowerUp · 2026-09-02

Qualcomm and ASUS launch Pharmaceutical AI Agent running fully on-device for community pharmacists in Taiwan

Qualcomm and ASUS jointly launched a Pharmaceutical AI Agent using Snapdragon X2 Elite Extreme AI PCs and Cloud AI 100 Ultra edge boxes, running a 20B-parameter GPT-OSS model fully on-device for private prescription analysis.

ASUS · 2026-09-02

Singapore HTX expands public-safety robotics with humanoid and underwater programs

HTX showcased the OceanOneK underwater humanoid robot at Milipol TechX 2026 and plans to establish a humanoid robotics centre for high-risk land applications including hazmat handling and firefighting.

Humanoid APAC · 2026-09-02

Tools & Startups

IBM and Confluent bring Granite time series foundation models into Confluent Cloud's Flink streaming engine

IBM's Granite time series models are now directly embeddable in Confluent Cloud's Flink engine, enabling forecasting, anomaly detection, and semantic intelligence on streaming data without data science expertise.

Hugging Face · 2026-09-02

Fable 5.1 World Modeling — open-source 3D world generation via AI agent swarms

PhiloLabs released Fable 5.1 World Modeling, an open-source project demonstrating 3D digital twin generation through AI coding agents, including a full Union Square SF reconstruction with procedural facades and a camera-match QA pipeline.

GitHub · 2026-09-02

Pirol CoOS — desktop app where an AI agent builds your CRM and ERP as local plugins

Pirol launched CoOS, a local-first desktop app where an AI agent builds CRM, ERP, and HR systems as plugins directly inside the app. Includes a CLI version for terminals and SSH sessions.

Pirol · 2026-09-02

Free Models

Free means $0 per token today, not unlimited or unconditional: both platforms rate-limit free use and require a free API key, some OpenRouter free endpoints may train on your prompts unless you opt out in account settings, and the roster changes daily. Verified against each provider's own pricing data at generation time — but pricing changes without notice, so confirm it's still free on the provider's own page before you rely on it.

In Brief

Source:

Google releases Gemini 3.8 Flash and 3.8 Flash Cyber, opens Fairwind program for trusted defenders

DeepSeek releases V4 Pro 0813 alongside fully open-source evaluation harness with full reproducibility logging

Wonderful raises $550M at $5B valuation for enterprise AI operating system, doubling valuation in under 6 months

HiddenLayer nabs $100M Series B for AI model security as ARR grows 10x YoY

Research & Papers

Post-training language models achieve gold-medal performance in coding competitions

Discriminative world models outperform generative prediction for web agent planning

From reweighting to rewriting: unlocking intervention effects of influential training samples

Dutch books for language models: LLM probability forecasts violate coherence axioms

ByteDance Seed publishes research on self-evolving agent harnesses that score on harness quality

Hardware & Robotics

New US export controls target Chinese access to remote AI servers in Thailand and Singapore

Nvidia's DLSS 5 launches today with NBA2K27 across all RTX 50 Series GPUs

Acer showcases RTX Spark compact desktop with 1 petaflop AI performance at IFA 2026

Qualcomm and ASUS launch Pharmaceutical AI Agent running fully on-device for community pharmacists in Taiwan

Singapore HTX expands public-safety robotics with humanoid and underwater programs

Tools & Startups

IBM and Confluent bring Granite time series foundation models into Confluent Cloud's Flink streaming engine

Fable 5.1 World Modeling — open-source 3D world generation via AI agent swarms

Pirol CoOS — desktop app where an AI agent builds your CRM and ERP as local plugins

Repos on GitHub

Trending on HF

Free Models

Free on OpenRouter

Free on OpenCode Zen

In Brief