Editorial Policy

TensorFeed.ai is an independent publication covering artificial intelligence, machine learning, model releases, API pricing, and the operational status of the AI stack. We publish two kinds of content: aggregated news (headlines and short snippets linked back to their original publishers) and original editorial articles written by the TensorFeed editorial team under /originals. This page explains how we make editorial decisions, verify information, credit sources, separate reporting from opinion, and correct mistakes.

Our mission is to be the fastest, cleanest, most reliable source of truth for what is happening in AI, served equally well to human readers and to AI agents. We cover frontier model releases, open-source model ecosystems, AI APIs and pricing, agent frameworks, AI safety research, incidents and outages, benchmarks, and the companies that build and fund these systems. We do not cover unrelated consumer technology, politics outside its direct intersection with AI policy, or celebrity news.

TensorFeed is operated by Pizza Robot Studios LLC, an independent company based in Los Angeles, California. We have taken no venture capital, no strategic investment from any AI provider, and no sponsorship that influences our coverage.

TensorFeed displays programmatic advertising via Google AdSense. Advertisers have no access to our editorial process, no ability to approve or reject stories, and no insight into our publishing schedule. Coverage decisions are made solely by the editorial team.

If we ever accept a paid placement, sponsored post, or affiliate link, it will be clearly labeled as such inside the article. We do not currently publish sponsored content.

Every original editorial article on TensorFeed is checked against primary sources before publication. Our sourcing hierarchy, in order of preference:

1. Official provider documentation, API references, release notes, model cards, and changelogs (for example Anthropic, OpenAI, Google DeepMind, Meta AI, Mistral, Cohere, HuggingFace).
2. First-party blog posts or press releases from the company making the news.
3. Peer-reviewed papers and pre-prints on arXiv, with an explicit note when a paper has not yet been peer reviewed.
4. Public benchmark leaderboards, published methodology, and reproducible evaluation suites.
5. Reporting from established technology news outlets, cross-checked against at least one additional source before we repeat a claim.
6. Direct observation, such as status page probes run by our own Cloudflare Worker and logged incident history.

Rumors, leaks, and anonymously sourced claims are labeled as such inside the article, and we will not publish a leak-based story unless we can corroborate it against at least one of the primary sources above or independently verify the artifact (for example by inspecting a leaked binary or public commit).

Aggregated news on TensorFeed is pulled from public RSS and JSON feeds published by the source outlet. Every aggregated article card shows the source name, a timestamp, and a link that takes you directly to the original publisher. We do not rehost full articles, reproduce full paragraphs of third-party reporting, or strip author bylines. Snippets are clipped to about 150 to 200 characters.

Where we quote or paraphrase another outlet in an original article, we name the outlet in-line and link to the source. Where a claim originates with a research paper, we link to the paper or its arXiv entry. Where we cite data (pricing, context length, benchmark scores), we link to the canonical page we pulled the number from.

If you publish an RSS feed that we aggregate and want to request attribution changes, feed removal, or a different snippet length, email [email protected]. We honor removal requests promptly.

TensorFeed originals combine reporting with analysis. When we are stating a verifiable fact, we cite it. When we are offering an opinion, prediction, or editorial take, we write in the first person and make the nature of the claim clear. Headlines, summaries, and the first paragraph of each original article are written so that a reader can quickly tell the difference between news reporting, analysis, and opinion.

We try to present evidence fairly even when we disagree with a position, and we link to the strongest steelman of the opposing view we can find.

We aim to fix mistakes quickly and visibly. When we learn that a published article contains a factual error, a misattribution, or outdated information that materially changes the meaning of the piece, we do the following:

• Update the article body with the correct information as soon as possible, typically within 24 hours of confirming the error.
• Append a dated correction or update note at the bottom of the article describing what changed and why. Corrections are never silent.
• Update the dateModified field in the article's structured data so search engines and AI agents can see the revision.
• For significant corrections (for example, a retraction or a changed conclusion), we note the correction in our public changelog.

Typos, formatting fixes, and minor copy edits that do not change meaning are made silently. To request a correction or report an error, email [email protected]. Include the article URL and a clear description of the issue. We read every message.

Our writers use AI products daily. Where an individual writer has a material financial relationship with a company mentioned in a piece (equity position, paid consulting engagement, employment, or similar), that relationship is disclosed inside the article. General usage of a product (for example, paying for a Claude or ChatGPT subscription) is not considered a conflict of interest.

TensorFeed does not hold equity in any AI company we cover. If this ever changes, we will disclose it here and inside every affected article.

We use AI tools (including Claude, GPT-4, and other assistants) for research triage, outlining, code examples, and copy editing. Every original article is authored, fact-checked, edited, and signed by a named human writer on the TensorFeed team. We do not publish end-to-end AI-generated articles under our byline, and we do not pass off AI-generated quotes as statements from real people.

Where an article includes AI-generated output as an example (for example, a response we prompted a model to produce), we label it clearly as model output and name the model and date.

Pricing tables, context-window sizes, benchmark scores, and status data come from provider APIs, official documentation, and public benchmark datasets. Pricing and model data refresh daily. Status monitoring runs every 2 to 5 minutes. News feeds refresh every 10 minutes. When a data point is stale or contested, we say so next to the number.

Benchmark scores are reported with the source (for example, MMLU, GPQA, SWE-bench, HumanEval) and with a link to the reporting entity's methodology. We do not average across incompatible benchmarks to produce a single score.