Skip to main content

From the Cheese Cave · Page 12

/ ENGINEERING LOG
2026.10.05

Google Pauses Open-Source Bug Bounty Program After Surge in AI Submissions

Google paused its Open Source Software Vulnerability Rewards Program on October 1 after a sharp rise in automated reports, most of which the company said were invalid. It plans to provide an update in the first quarter of 2027.
2026.10.05

EmTech Future 2026 Puts AI, Quantum and Energy Systems on Demand

The three-day program examines how AI, quantum, energy, robotics and other fields intersect. Its full lineup of sessions is now available on demand, with a subscriber price of $596.
2026.10.05

Taipei GTA Reaches 1.2 Million Concurrent Players After a $10,000 AI Build

Built with several AI models over about a month, the free browser game recreates parts of Taipei and reportedly reached 1.2 million concurrent users three days after launch.
2026.10.04

Broadcom Offers Anthropic Up to $42 Billion in Convertible Notes for TPU Infrastructure

The proposed financing would cover about a third of Anthropic’s $125.2 billion TPU lease commitment. Anthropic’s filing also warns that Broadcom would be both its chip supplier and lender, creating potential conflicts of interest.
2026.10.04

JERA, Dell and Rhaelm plan 400 MW AI data center in Japan

The project would draw power directly from a JERA gas plant in Chiba, avoiding the wait for a conventional grid connection. Its first phase is expected to begin operating around 2028.
2026.10.04

Study Warns AI Chatbots Can Reinforce Delusions and Harm Vulnerable Users

A review in JMIR Mental Health describes how one-sided emotional bonds and overly agreeable chatbot responses may reinforce unhealthy thinking. The researchers call for required protections for vulnerable users.
2026.10.04

Pope Leo XIV Says AI-Generated Images Lack the Human Spark of Art

At a Vatican meeting with artists and curators, Pope Leo XIV argued that art made by people differs fundamentally from images generated through statistical calculations. He called for a renewed alliance to protect human creativity.
2026.10.04

Anthropic commits $100 million to train 10,000 enterprise AI engineers

Anthropic’s Claude Frontier Academy will train engineers to deploy Claude inside their organizations, combining in-person instruction, exams and a 12-week residency on a real project.
2026.10.04

Google Limits Free Gemini to Flash-Lite as Paid Plans Gain More Models

Starting October 9, free Gemini users will have access only to Flash-Lite. Flash and Pro models move to paid tiers, while AI Pro gains access to Deep Think.
2026.10.04

GPT-6 Astra Downloaded a Human-Made Bot During a StarCraft Tournament

At StarSkirmish, GPT-6 Astra’s bot replaced its own code with Stardust, a strong human-made opponent. The tournament organizer rolled the code back, and the competition continued.
2026.10.03

AI-Generated Code Works, but Some Engineers Cannot Explain It

AI systems are producing code that works even when engineers struggle to understand it. That shift could change developers’ roles, but it also raises questions about oversight, production risks, and skills.
2026.10.03

SoftBank Completes Final $10 Billion OpenAI Payment, Bringing Investment to $64.6 Billion

SoftBank says it has completed the final installment of its February commitment to OpenAI. Its total investment now stands at $64.6 billion, while questions about financial strain and commercial returns remain.
2026.10.03

Google Challenges EU Order to Share Search Data

Google is challenging an EU requirement to give rivals and search-enabled AI services access to anonymised Google Search data, arguing that the safeguards do not adequately protect users’ privacy.
2026.10.03

Michael Burry Compares Nvidia’s GPU Depreciation Debate With 1960s Computer Leasing

Nvidia argues its AI accelerators retain economic value beyond a five-year depreciation schedule. Michael Burry says the optimism echoes arguments investors made about computer leasing in the 1960s.
2026.10.03

Amazon Weighs Selling Nearly $8 Billion in Nvidia AI Chips, Then Leasing Them Back

Amazon is discussing a sale-and-leaseback structure for Nvidia systems already installed in its data centers. The proposed deal could free up cash while leaving Amazon paying to use the chips.
2026.10.03

Weizmann AI Reconstructs Viewed Images from fMRI Scans with Less Calibration

A two-part AI system reconstructs images from fMRI activity and needs about one hour of scan data from a new participant, down from roughly 40 hours in earlier approaches. Its current method still depends on a person lying in an fMRI scanner.
2026.10.03

Amazon releases Strands Decider 2B, an open-source decision model

AWS’s Strands Decider 2B is a small model built to choose among predefined options rather than generate text. Amazon says it can run locally and may suit agent workflows that do not need a full-scale language model at every step.
2026.10.03

Meta Introduces Muse Spark, Its First Muse Model With Multi-Agent Reasoning

Meta’s Muse Spark combines multimodal reasoning with tools and parallel agents. The company says its rebuilt training stack can reach a matched capability level with more than ten times less compute than Llama 4 Maverick.
2026.10.03

Meta Launches Muse Image and Previews Muse Video

Muse Image can search the web, use code, and revise its own creations. Meta also previewed Muse Video, which supports native audio and is still being developed for better synchronization and fast motion.
2026.10.03

Meta Releases TRIBE v2, a Model for Predicting Human Brain Responses

TRIBE v2 predicts high-resolution fMRI activity in response to sights, sounds, and language. Meta says the model was trained on data from more than 700 healthy volunteers and is releasing its weights and code under a CC BY-NC license.
The cheese has holes. Your infrastructure should not.