Latest stories
Every story we have published, newest first. Each one links to the filings, papers and official documents behind it.
- Business & Work
Anthropic’s IPO Filing Explained: Who Owns It, and Who Really Controls the Board?
Anthropic’s S-1/A gives public investors economic rights, the founders 50.1% of shareholder votes and its Long-Term Benefit Trust a board majority. We break down the share classes.
- AI Models
Gemini 4 Carbon: Who Gets Google’s Cyber Model First, and What Its Benchmarks Really Show
Google DeepMind is giving Gemini 4 Carbon to cyber defenders first, with no date for a wider release. We explain the access tiers and where each benchmark score comes from.
- Business & Work
GitHub Octoverse 2026: AI Agents Now Start 40% of Pull Requests. Here’s What That Really Means
GitHub says more than 40% of September pull requests began with an agent’s first draft, up from 17% in March. What the number measures, what it misses, and what teams should track.
- India AI
India’s AI Rules Explained: What Is Already Law, and What Is Still Only Expected
India’s IT Rules amendment on synthetic media has been in force since 20 February 2026. A broader AI governance consultation paper has been promised but not yet published. Here’s the difference.
- Guides & Comparisons
ChatGPT vs Claude vs Gemini vs Copilot: AI Assistants Compared (2026)
Compare documented workflows in ChatGPT, Claude, Gemini and Microsoft Copilot. A source-led guide, not a performance ranking.
- Guides & Comparisons
AI Image Generators Compared: What Each Tool Documents in 2026
Compare documented generation and editing workflows in ChatGPT Images, Adobe Firefly, Midjourney and Gemini Apps.
- Guides & Comparisons
Hosted API vs Open-Weight Models: How to Choose in 2026
Compare provider-hosted Gemini access with documented Gemma and Mistral open-weight deployment options.
- Guides & Comparisons
AI Content Provenance Tools Compared: C2PA, SynthID and Watermarks
Compare provenance credentials, watermarks and image classifiers by the evidence each can establish.
- Guides & Comparisons
AI Coding Agent Benchmarks Compared: How to Read SWE-bench and Friends
A source-led comparison of two coding-agent benchmarks, their task designs, graders and limits.
- Guides & Comparisons
IndiaAI Compute vs Commercial Cloud GPUs: Access, Pricing and Trade-offs
Compare IndiaAI Compute Portal access with commercial cloud on eligibility, allocation, billing and capacity evidence.
- India AI
IndiaAI Compute in 2026: What India’s GPU Numbers Really Measure
Official releases show a larger shared GPU pool and millions of reported GPU-hours. They still do not put commitments, live availability and measured use on one public ledger.
- Chips & Infrastructure
AI’s Electricity Demand Is Becoming a Product-Design Problem
The IEA projects data-centre electricity use to rise from 415 TWh in 2024 to about 945 TWh by 2030. Those system-wide figures do not measure energy per AI request.
- AI Models
AI Agents Are Getting Better at OSWorld, but Reliability Is Still the Real Test
Stanford’s 2026 AI Index puts computer-task performance at 66.3%—a sharp gain, but still below its reported human baseline.
- Research
Do Word Labels Steer How LLMs Play the Prisoner’s Dilemma? What a New Preprint Finds
A revised arXiv preprint reports a label-sensitive circuit in tested models; its findings remain limited to specific prompts and game protocols.
- Research
ProteoGPT: How Generative AI Found Peptide Candidates Against Drug-Resistant Bacteria
One protein-language-model pipeline reports in-vitro and mouse-model results; the findings remain preclinical.
- Policy & Regulation
C2PA Content Credentials Explained: Why Provenance Doesn’t Prove Content Is True
EU disclosure obligations, a voluntary code and a contested provenance standard do different jobs.
- AI Models
Mistral Small 4 Explained: Why “6.5B Active Parameters” Isn’t Its Real Size
Mistral lists 119 billion total parameters and 6.5 billion active parameters; its model card does not establish hardware fit or runtime performance.