Which artificial intelligence topics are rising and falling this week, with a glossary and searchable archive of La Rebelión AI articles.

Artificial intelligence: news and trends

AI.larebelion · live

Ask our AI archive

It answers only from what we have published, cites every post and shows you what it read to reply.

It can be wrong, says “I don’t know” when it finds nothing to go on, and does not store what you type.

How a post gets written

The latest post, step by step

The real path of the machine’s most recent post, with what it records. It does not keep track of how long each step takes, so we do not make it up.

Source:thehackernews.com→ published asFallo crítico en FortiMail explotado: CISA lo añade al catálogo KEV

Step 1Curation1 of 683News read in the 24 h before the draft
Step 2FormatBriefThe format the classifier picked
Draft · 2 Oct 17:02Writing634 wordsWritten by claude-sonnet-5-5
Step 4Images1 · 1280×853Real size measured, never stretched
Step 5Links5 / 5 OKEvery link checked before the draft
Published · 2 Oct 17:08ApprovalA personReviewed and approved by hand

The map of what we cover

Topic constellation

The 14 labels with the most posts out of the 2,097 published: each dot’s size is its number of posts and the lines join topics that share posts. Select a topic to see the latest.

1,373 posts

See all posts about AI

Weekly radar

AI topics rising this week

Stories mentioning each topic among those the pipeline reads: 4,263 from 25 Sep – 2 Oct against 4,584 from 18 Sep – 25 Sep. The change compares each topic’s weight within its own week, not the total, so reading more news does not lift every topic.

Rising

Chatgpt121 → 230 · +104%
Assistant10 → 16 · +72%
Programación24 → 30 · +34%

Falling

Gemini55 → 33 · −35%
Linux7 → 5 · −23%
Google164 → 126 · −17%

Transparency

The machine, in numbers

How much it writes, how much we review and what gets discarded in the last 30 days.

Posts in 30 days1253 different formats

Approved by a person58%the rest automatic after 30 min (5)

Discarded3drafts that never got published

Verified links100%40 of 40 were live when published

Formats used: Brief (7) · News commentary (3) · Developing story (2).

Approvals, discards, formats and links are recorded since 1 Oct 2026.

Living glossary

AI glossary

They are extracted from the blog’s AI posts and updated with every new one. Select a term to see where it appears.

3 postsAI agentsAI agents are programs that don't just answer questions but make decisions and carry out tasks on their own, such as searching for information, using apps, or completing workflows. They matter because they promise to automate real work, but if they fail or act on wrong information, the consequences can hit a business. That is why companies remain cautious.

3 postsAI alignment and safetyAI alignment and safety aim to make AI systems behave according to human values and intentions and avoid causing harm, whether through mistakes, misuse or unexpected behavior. It is a key topic because trust in these tools depends on it, and so do the laws and rules that governments are beginning to pass.

3 postsBenchmarksBenchmarks are standardized tests that measure how well an AI handles tasks such as reasoning, coding or answering questions, letting people compare models against one another. They matter because companies use them to showcase progress, but a strong score on a controlled exam does not always reflect how useful a model is in everyday use.

3 postsContext windowThe context window is the maximum amount of text a language model can consider at once, including your question, the conversation history and any documents you provide. It works like the model's short-term working memory. The larger it is, the more long material it can analyze without forgetting the beginning.

3 postsEmbeddingsEmbeddings are numerical representations of the meaning of a text, image or other content, so that similar things end up close together on a mathematical map. This lets an AI find related documents or products even when they share no exact words. They underpin smart search engines and recommendation systems.

3 postsFine-tuningFine-tuning means taking an already trained AI model and training it further on more specific data so it masters a particular task or field, such as legal language or customer support. It is far cheaper than building a model from scratch. It lets companies adapt AI to their own needs.

3 postsGenerative AIGenerative AI is the branch of artificial intelligence that creates new content, such as text, images, audio, or code, based on what it has learned from large amounts of data. It matters because it is changing how we work and create, but it also opens new risks, from confidently stated errors to cyber threats that can act autonomously.

3 postsGPUs and AI chipsGPUs and other AI chips are processors built to perform millions of calculations at once, which is essential for training and running artificial intelligence models. They matter because they determine how much AI costs, how fast it runs, and who can afford it. They have also become a security target, since hardware flaws can compromise the systems that rely on them.

3 postsHallucinationA hallucination happens when an AI model produces false, invented or nonsensical information while presenting it with complete confidence and apparent coherence. It can be a made-up fact, a fake quote or a misread image. It matters because convincing-sounding answers are easy to trust, so it pays to verify what an AI claims before relying on it.

3 postsInferenceInference is the moment when an already trained AI model is put to work generating an answer, such as when you type a question into a chatbot. It happens every time you use AI, so its speed and cost determine how fast a service responds, how much it costs to run and how much energy it consumes.

3 postsLarge language model (LLM)A large language model (LLM) is an AI system trained on vast amounts of text to understand and generate human language. It is the engine behind today's chatbots and assistants. It matters because it writes, summarizes, and answers fluently, but it can also make mistakes or be manipulated, so it needs safeguards and should not be trusted blindly.

3 postsMultimodal AIMultimodal AI can understand and combine different kinds of information, such as text, images, audio or video, instead of working only with words. It matters because it comes closer to how people perceive the world and enables practical uses like describing a photo, reading a chart or holding a spoken conversation.

3 postsOn-device AIOn-device AI runs directly on your phone, laptop or other gadget instead of sending your data to cloud servers. It often relies on specialized chips called NPUs. It matters because it can be faster and more private, but it also creates new security challenges for companies.

3 postsOpen-source AIOpen-source AI refers to artificial intelligence models whose design and weights are published so anyone can download, use and modify them. Unlike closed systems from a few companies, it lets more people innovate and compete. For the general public, it means more choices, lower costs and greater transparency about how these tools work.

3 postsOpen-weight modelsAn open-weight model is one whose internal parameters, meaning what it learned during training, are published so anyone can download, run and adapt it. This allows use on your own hardware with more privacy, although it does not always include the training data or full code, and it also raises security concerns.

3 postsPromptA prompt is the instruction or question you give an AI to get a response, and the quality of the result depends largely on how you word it. It also matters for security: in prompt injection attacks, someone hides malicious commands inside text to trick the AI into ignoring its rules or leaking data.

3 postsShadow AIShadow AI is the use of artificial intelligence tools, such as chatbots or assistants, by employees without the IT department knowing or approving it. It matters because confidential information can end up in outside services with no oversight. For companies, it is a growing risk to security, privacy and regulatory compliance.

3 postsTokensTokens are the small pieces into which a language model splits text in order to process it: whole words, syllables or punctuation marks. The AI reads and writes one token at a time. They matter because they determine how much text fits in a conversation and, often, how much the service costs to use.

How we work

House rules

Always sourcedEvery post links the original story and says it was written by AI.larebelion.

A person decidesEverything goes through approval on Telegram or in the dashboard; if nobody replies within 30 min, it is published.

No making things upIf it is not in what we have published, LaRebelionBOT answers “I don’t know”.