Cutting Through the Noise

Every day brings a new “game-changing” model, a new course promising mastery, and a new thread of secret prompts. This page gives you five questions and a set of red flags for judging any of it — and an honest list of what you can safely ignore.

Open any social feed and count. A video says a new model “changes everything” — the third such change this month. An advert offers a weekend course that will make you “an AI expert” with a certificate to prove it. A thread promises “50 prompts the top 1% use”. A headline says AI now “beats doctors”. A friend forwards an app that will write, design and invest for you. Each of these asks for your attention, often your money, and sometimes your data. You cannot check them all, and you do not need to. What you need is a quick, repeatable way to decide which deserve a second look — and the confidence to scroll past the rest.

Why there is so much noise

This guide’s framing Most AI content is not made to inform you. It is made to be clicked, shared or bought, and those goals reward different things from accuracy. Understanding who gains from a piece of content is the fastest filter you have.

New technologies often follow a pattern sometimes drawn as a hype cycle: inflated expectations, disappointment, then slower, real progress. AI has been through it before, twice ending in an “AI winter”. Noise is loudest at the peaks. None of this means all AI content is worthless. It means the reliable material is mixed in with a great deal that is optimised for something else — and the burden of sorting falls on you.

Five questions before you trust an AI claim

This guide’s framing These work for a course advert, a post, an app or a headline. You can run them in under a minute. They build on the SIFT method from Checking AI’s Work, adapted for content about AI rather than content from AI.

  1. Who is saying it, and how do they make money?A researcher, a journalist, a vendor, an affiliate marketer and a course seller can all say the same sentence for very different reasons. Search the name in a new tab — what do independent sources say about them?
  2. What exactly is being claimed?Rewrite it in one plain sentence. “Revolutionary new model” usually shrinks to “scored a few points higher on one test”. Vague claims cannot be checked, which is often the point.
  3. What is the evidence, and can I see it?A paper, a public demo you can try, a product you can use for free, a named study? Or a screenshot, a testimonial and a countdown timer?
  4. What does someone independent say?Read laterally — the fact-checkers’ habit of lateral reading. Look for coverage from people with nothing to sell: established news outlets, university researchers, official bodies, practitioners writing about their own experience. If nobody independent has noticed, treat it as unconfirmed.
  5. What would it cost me to be wrong?Ignoring a genuine new tool for a month costs almost nothing. Paying RM3,000 for a worthless course, or handing your files to an unknown app, costs a lot. Match the effort of checking to the stakes.

Alongside the five questions, a handful of warning signs turn up again and again. Tap each to see why it matters.

Judging courses and certificates

There is a genuine skills gap, and good courses help. But the market is crowded, and a course’s marketing tells you little about its quality. Before you pay, look for answers to five practical questions.

The question

Your employer will pay for one AI course. Free university course plus your own project, or a paid certificate programme?

Choose one to hear its case.

Judging influencers, threads and videos

Some of the most useful AI explanations online come from independent creators: researchers explaining their own work, teachers, practitioners showing honest workflows. The problem is not the format but a few recurring patterns.

A good creator explains why something works, shows failures as well as successes, links to sources, updates or corrects old content, and does not make you feel stupid for not buying something.

Judging tools and apps

New AI apps launch daily, and many of them are thin layers on top of the same few underlying models. Before you install one or upload anything to it, ask:

For choosing between the main assistants and specialist tools, see Choosing the Right Tool. The short version: a mainstream assistant from an established company As of 2026 covers most everyday needs, and you add specialist tools only when you hit a real limit.

Judging news and research claims

AI news moves fast and is often written from press releases. A few habits help you read it well.

Practise on these realistic (invented) headlines and adverts.

Judge the claim

“New AI model beats human doctors at diagnosis!” The article links a company blog post about a multiple-choice medical exam. What is the fairest reading?

Exam questions are tidy and complete; real patients are not. The result may be genuine and still say little about clinical practice. Look for independent studies in real settings.

An advert: “Become a certified AI expert in 48 hours. Guaranteed RM8,000 side income. Only 5 seats left!” How many of the red flags does it show?

Any one of these is a warning; three together is a pattern. Skip it, and try a free course from a university or established provider first.

A thread shares a screenshot: “This one prompt made the AI reveal its secret settings.” What should you conclude?

Models often produce plausible-sounding but invented answers about themselves. Without a reproducible method, a screenshot is an anecdote.

A new app: “Upload all your bank statements and our AI will find savings.” It launched last month and has no privacy policy. What do you do?

Who is behind it and what happens to your data come first. A spreadsheet and your own reading of the statements carry no such risk.

“Researchers show new method cuts AI errors by 40%.” The link goes to an arXiv preprint posted yesterday. What is the right stance?

Preprints are how most AI research first appears. They deserve attention, not certainty. “40% fewer errors” on one narrow test may not carry over to anything you use.

What you can safely skip — and what to keep

This guide’s framing This is our opinion, offered so you can stop feeling guilty about everything you have not read.

Other views
  • Keep up daily: people building AI products, investing or reporting on the field argue that daily tracking does matter for them — release details can change their plans within days. That is a reasonable choice for a professional, not a requirement for everyone else.
  • Detectors have a place: some educators use detector scores as one signal to start a conversation with a student, never as a verdict. Watermarking schemes built into some generators are also more reliable than detectors that guess from style alone.

A simple information diet

Instead of trying to keep up with everything, build a small, deliberate habit.

  1. Fundamentals first. Understanding how models work and fail makes every later headline easier to judge. Work through the Novice track and then the Practitioner track; not sure where to start? Use Choose Your Track.
  2. Three to five trusted sources. Choose a mix: one reputable newsletter or news outlet, one official source from a company whose tools you use, one researcher or educator who explains rather than sells, and one source close to home such as Malaysia’s national AI office. Drop any source that repeatedly fails the five questions.
  3. Weekly, not daily. Set aside thirty minutes a week. Save interesting items during the week and read them together; half will have been corrected or forgotten by then.
  4. Try before you trust. When something sounds important, test it yourself on a real task of yours. Your own experience is the best filter there is.
  5. Check this site’s What’s New for changes that affect what we teach.

// Pause and reflect

  1. Think of the last AI post, course advert or headline that caught your attention. Run it through the five questions now — does it still deserve your time?
  2. Which of your current AI sources earn money from your attention or your purchases? Does that change how you read them?
  3. Which three to five sources will make up your weekly information diet, and what will you unfollow?

Your notes are saved privately in this browser.

Try this

For one week, every time you see an AI claim you would normally share or act on, write it down with a one-word verdict from the five questions: trust, check or skip. At the end of the week, look at the list. Most people find that more than half were skips — and that skipping cost them nothing.

Key terms on this page

Sources & further reading