标签：#TRAI

"Everyone wants to take that aircraft up now..."

There is a comforting story about how AI models handle the enormous quantities of text and images they are trained on: they do not store any of it, they merely learn general patterns, and once training is done the original data is gone in any meaningful sense. It is a reassuring account, and it is not quite true. Large models memorise fragments of their training data — verbatim, recoverable fragments — and a decade of research has produced reliable ways to detect and extract them. The answer-first version: if your data was in a model’s training set, the model may have memorised identifiable pieces of it, and those pieces can leak. Two families of attack make this concrete. Membership inference works out whether a specific record was in the training data at all. Data extraction pulls memorised content back out word-for-word. Neither is exotic; both are well documented against production systems. This is the mechanism underneath both the newspaper lawsuits alleging near-verbatim reproduction of their articles and the quieter privacy research showing that models leak the people in their training sets. Understanding it is the difference between trusting the comforting story and knowing its limits. Memorisation is a feature of the maths, not a bug Start with why models memorise at all. A large neural network has an enormous number of parameters — enough capacity to do more than compress general patterns. During training it is rewarded for predicting its training data accurately, and one very effective way to predict a specific example accurately is to memorise it. For data that appears once in an unusual form, or many times in an identical form, memorisation is often the path of least resistance for the optimiser. This is measurable. Researchers can show that a model assigns systematically higher confidence, and lower prediction error, to examples it was trained on than to otherwise-similar examples it has never seen. The size of that gap grows with the size of the model

Bankrupt Spirit accused of selling out workers in massive data sale to Google.

"I will tear apart this forest and bring back what it took from me."

Amazon’s team uses a T. rex preparing to devour a book as its logo.

AI firms quietly bulk buying rare books face resistance from booksellers.

A multi-agent system without an orchestrator is just a collection of agents. Each one is capable, but none of them coordinated. They might all be excellent at their individual jobs - searching the web, writing code, calling APIs - but without something deciding what gets done, in what order, by whom, and what to do when a result comes back wrong, the system does not behave like a system. It behaves like a group project with no project manager. The orchestrator is the project manager. Its job is not to do the work. Its job is to make sure the work gets done - and that is a harder, more subtle problem than it sounds. What an orchestrator is responsible for An orchestrator does four things, and only these four things: 1. Decompose the goal. Turn a high-level objective into a concrete set of subtasks. This is a planning problem, not an execution problem. The orchestrator decides what needs to happen, not how to do it. 2. Route tasks to the right workers. Match each subtask to an agent capable of doing it. This requires knowing what tools and capabilities each worker has - not in detail, but well enough to delegate correctly. 3. Manage state across the workflow. As workers return results, the orchestrator decides what those results mean for the remaining plan. Sometimes a result changes the plan entirely. Sometimes it confirms the next step. The orchestrator holds the full picture. 4. Synthesise the final output. Worker outputs are partial. The orchestrator assembles them into a coherent response and decides when the goal has been met. Notice what is absent: the orchestrator does not call APIs, does not run code, does not search the web. It reasons about work and routes it. The moment an orchestrator starts executing, it loses the focus that makes it good at coordination. Building one from scratch Here is a minimal orchestrator in Python. It plans upfront, delegates to type workers, and synthesizes results: import json def orchestrator ( goal : str , workers : dict ) ->

AWS Builder Center now offers free, time-limited sandbox environments for workshops, so developers no longer need to use their own AWS account and credit card or worry about unexpected charges. This has been a long-standing request from the community and removes one of the biggest friction points for practitioners learning new AWS technologies. By Renato Losio

"I'm beginning to suspect we are not in Jumanji."

How agents remember - and why deciding what to forget is the real skill An agent that starts every step with a blank mind cannot really pursue a goal. It would reintroduce itself to you on every message, forget what it just tried, and repeat the same mistake forever. Memory is what turns a stateless model into something that accumulates - that knows who you are, what it has already done, and what it learned last Tuesday. This post is about how that works and, more importantly, about the discipline of deciding what an agent should remember at all. The context window is not memory. The first thing to unlearn: a model’s context window is not its memory. The context window is working memory - RAM, not a hard drive. It is finite, it is reset on every request, and every token in it costs money and dilutes the model’s attention. Stuffing an entire conversation history and knowledge base into the prompt does not scale, and past a point it actively hurts - the model loses the important signal in a sea of stale detail. Real memory lives outside the window and is selectively loaded into it when needed. Four kinds of memory Borrowing loosely from cognitive science, agent memory is usually split into four types, and good systems use all of them: Short-term/working memory - the current conversation and the agent’s recent thoughts and observations. Lives in the context window. Long-term episodic memory - a record of what happened : past conversations, decisions, and the outcomes of previous tasks. Long-term semantic memory - facts and knowledge: who the user is, domain information, documents. This is what retrieval-augmented generation pulls from. Procedural memory - how to do things : learned skills, tool-use patterns, and reusable strategies. Short-term memory: the rolling buffer The simplest memory is just keeping recent turns in the prompt. The problem is that conversations outgrow the window, so the standard move is to keep the last few turns verbatim and summarise the older

"The people I trusted betrayed me. The justice system failed me."

Anthropic blocks authors from opting out of $1.5B settlement at last minute.

"Something's coming. Something we may not be able to deter."

After months of teasing us with reminders about how large Avengers: Doomsday's cast is going to be, Marvel has finally released a proper trailer for the film. Though Robert. Downey Jr.'s Doctor Doom definitely appears to be causing a lot of the chaos in Doomsday's first trailer, the movie seems like it's going to feature […]

"If we can't control the force of nature, at least we can control the narrative."

"You promised me that you would never take power in your name."

The caller ID company says users are increasingly ignoring and blocking calls from India's dedicated business number series.

Holes in socks have become a curious sight at this year’s World Cup. The reasons why are a weird mix of biomechanics, perception, and player habits.

Introduction: The Need for Modern Web/AppSec Training The cybersecurity landscape is evolving at a breakneck pace, but the tools we use to train the next generation of defenders are stuck in the past. Most web/appsec learning platforms still focus on basic, textbook vulnerabilities —XSS popups, simple SQL injection, or trivial IDORs. These labs are like teaching someone to swim in a kiddie pool; they might grasp the concept, but they’re ill-prepared for the open ocean of modern web applications . The gap isn’t just in depth—it’s in relevance . Real-world apps today are complex, API-driven, and riddled with subtle, pattern-based vulnerabilities that don’t fit into neat, isolated lessons. Consider this: a developer misconfigures a GraphQL endpoint, exposing an entire database. Or an API leaks sensitive data because of a flawed rate-limiting mechanism. These aren’t edge cases—they’re common mistakes in modern apps. Yet, most training platforms ignore them, leaving learners to either stumble upon these issues in the wild or remain oblivious. The result? A workforce of security professionals who can theoretically exploit a vulnerability but struggle to identify or fix it in a real-world context . The problem isn’t just outdated content—it’s the lack of ethical, hands-on practice environments . Many aspiring security professionals resort to illegal or gray-area practices to gain experience, risking legal consequences and ethical dilemmas. What’s needed is a platform that simulates real-world scenarios without crossing ethical boundaries, one that teaches not just how to exploit but also why vulnerabilities occur and how to fix them . Here’s the core issue: modern apps are systems, not isolated components . A vulnerability in one part—say, a file upload feature—can cascade into a full account takeover if combined with a session management flaw. Most labs fail to teach this interconnectedness , leaving learners with a fragmented understanding. A practical platform must brid

FTC urged to reject Elon Musk’s bid to end X monitoring amid AI concerns.