by admin | Sep 2, 2026 | AI Guides
You paste a long report into ChatGPT, ask questions about it for twenty minutes, and then it answers something that flatly contradicts what you told it at the start. Most people assume the AI got confused. Usually it is simpler than that. The chat ran out of room, and that room has a name: the context window.
It takes five minutes to understand, and once you do, the strange behaviour stops looking random and starts looking predictable. That is the whole point, because predictable problems can be worked around.
What is a context window, in plain English
A context window is the total amount of text an AI model can look at while writing its answer. Google’s own developer documentation uses the clearest analogy: think of it as short term memory. Anthropic’s documentation calls it “working memory” and defines it as all the text the model can reference when generating a response, including the response itself.
Here is the part that surprises people. The model does not remember your conversation the way you do. Every time you send a message, the tool quietly re-sends the whole chat so far, and the model reads all of it again from scratch before replying. Nothing is stored in its head between messages. The context window is simply how much of that pile it can hold at once.
The pile is measured in tokens rather than words. A token is a chunk of text, usually a short word or part of a longer one. Our guide on what a token is in AI explains it with examples.
Everything counts, including things you forgot about
People assume only their own typing fills the window. According to Anthropic’s documentation, everything in the request counts: the hidden system instructions the company wrote, every message in the conversation, any files or images you attached, and the answer the model is currently writing.
So a chat where you uploaded three PDFs and got five long answers is far fuller than it looks. The scroll bar shows your side of it. The window is holding both sides plus the attachments.
Why the AI forgets what you told it
When a conversation gets close to the limit, something has to give. Anthropic notes that chat interfaces can manage the window on a rolling first in, first out basis, which is exactly what it sounds like: the oldest part of the conversation drops off the front to make space at the back.
That is your answer. The instruction you gave in message two, the one about writing in British English or never using bullet points, was not ignored. By message forty it was no longer in the window at all. The model was not being careless. It could not see it.
Some tools handle this more gracefully by summarising the earlier part of the chat and carrying the summary forward instead of the full text. Anthropic calls this compaction. It helps, but a summary is still lossy, so small details from early on can quietly disappear.
A bigger context window is not automatically better
This is the part most beginner articles skip, and it matters more than the headline numbers. Anthropic’s documentation says it plainly: more context is not automatically better, because as the token count grows, accuracy and recall degrade. They call the effect context rot, and their engineering team writes that while some models degrade more gently than others, the pattern shows up across all of them.
Google says something similar in its long context documentation. Finding one specific fact buried in a huge amount of text works well, often around 99 percent of the time. Finding several specific facts at once does not work as reliably, and performance varies a lot depending on the material.
So when a company advertises a million token context window, read it as “this much will fit” rather than “this much will be read carefully”. Both companies say so themselves, in their own developer docs.
How big are context windows now?
They have grown quickly. Google’s documentation traces the path: earlier models handled around 8,000 tokens, then 32,000, then 128,000, and Gemini was the first to accept a million. Google makes a million tokens concrete like this.
- All the text messages you have sent in the last five years
- Eight average length English novels
- Transcripts of more than 200 podcast episodes
These numbers change every few months, and free tiers often have a smaller window than paid ones, so treat any figure you read online as a snapshot. The shape of the problem does not change: there is a limit, everything counts toward it, and getting close to it costs you accuracy.
Context window and memory are two different things
These get mixed up constantly. The context window is the limit inside one conversation. Memory is a separate feature that saves facts about you and carries them across different conversations. OpenAI’s Memory FAQ describes it as ChatGPT automatically remembering useful context from your chats, files and connected apps, and says you can turn it off in Settings at any time.
The practical difference: memory is why a brand new chat already knows you are a nurse in Dublin. The context window is why that same chat forgets a formatting rule you gave it half an hour ago. We covered the memory side in how AI memory works, including how to see and delete what has been saved about you.
Five habits that fix most “the AI forgot” problems
1. Put your question at the end. Google’s own documentation recommends this: when the context is long, put your question after all the background material rather than before it. Paste the document first, then ask.
2. Start a new chat instead of arguing. If a long conversation starts going in circles, that is a room problem, not a reasoning problem. Open a fresh chat and paste a short summary of what matters.
3. Give each chat one job. One chat for the cover letter, one for the spreadsheet, one for the holiday plan. Long mixed conversations fill the window with material that is irrelevant to whatever you are asking right now.
4. Repeat the rule that matters. If a formatting or tone instruction is important, restate it in the message where you need it. One line, no guesswork.
5. Upload the ten relevant pages, not the whole book. Trimming the input is the most reliable improvement, because you are spending the model’s attention on the part you care about.
From my own work on client websites, this changed how I use these tools. I used to paste an entire plugin’s documentation into a chat and then wonder why the answers drifted by the fifth question. Now I paste the one page I need and start a new chat when the topic changes. The answers got noticeably better without changing tools or paying for anything.
Important tip: if a long chat starts contradicting itself, do not try to correct it in place. Open a new chat and paste a five line summary of what has been decided so far. You will get a better answer in less time than the argument would have taken.
What this means for the way you write prompts
Once you picture the window, good prompting stops being a list of tricks. You are deciding what deserves the space. Anthropic’s engineering team frames it as finding the smallest set of high value information that gets the outcome you want, and that scales all the way down to writing one email. For the practical version, see our guide to writing better AI prompts. For the wider picture, start with what a large language model is, and keep our AI glossary open if the jargon slows you down.
Common Questions
Does a longer chat make the AI slower?
Usually yes, a little. Google’s documentation notes that longer inputs generally mean higher latency before the first word appears, because there is simply more to read.
Can I check how full my context window is?
In most consumer chat apps, no. Developers can count tokens before sending, but the everyday apps do not show a meter. Watch for the warning signs instead: repeated questions, forgotten instructions, and answers that contradict earlier ones.
What happens if I go over the limit?
In a chat app, the oldest messages usually drop out or get summarised, and the conversation carries on without telling you. If a single input is too big on its own, for example one very large file, you will normally get an error instead.
Does a bigger context window mean fewer wrong answers?
No. A bigger window reduces forgetting, but it does not stop the model inventing things, and both Google and Anthropic report that recall gets less reliable as the window fills. Check important facts either way.
Final takeaway
The context window is the AI’s desk, not its brain. It only holds so much, everything you put on it competes for space, and the pile at the back gets pushed off the edge when new work arrives. Keep the desk tidy and the work gets better.
You do not need a bigger plan or a smarter model to see the difference. Trim what you paste, put your question last, give each chat one job, and start fresh when things drift. Those four habits are free, and they fix most of the moments where AI feels unreliable.
by admin | Aug 31, 2026 | AI Tools
You take a photo you actually like, and then you notice it. A stranger’s elbow in the corner. A bin by the wall. A wire across the sky. Ten years ago that meant learning Photoshop or letting it go. Today your phone removes it in about four seconds, for free.
That is the good news. The part almost nobody explains is what happens to the photo afterwards. AI photo editing does not only change the picture. In several of these tools it also writes a note into the file saying the picture was changed. This guide covers both halves: which tools genuinely help, where the free versions stop, and what your photo quietly remembers.
What AI photo editing actually means
Old photo editing moved pixels that were already there. You cropped, brightened, or cloned one patch over another.
AI photo editing generates pixels that were never in the shot. When you erase a person from a beach photo, the tool has to invent the sand and sea that were hidden behind them. It is making a confident guess.
That distinction matters twice over. Invented detail can be wrong, especially in faces, text, hands and repeating patterns. And invented detail is exactly what the newer transparency systems are built to flag. Our guide to AI image generators for beginners covers the creating side of the same technology.
The free AI photo editing you probably already have
If you use Google Photos, you already own most of what people pay for. In April 2024 Google announced that its AI powered editing tools would be available to anyone using Google Photos, with no subscription required, rather than staying locked to Pixel phones and paid Google One plans.
Google’s current help page for editing on Android documents the tools worth knowing:
- Magic Eraser removes distractions and unwanted people from the background.
- Unblur sharpens a photo that came out soft.
- Best Take blends a burst of group shots so nobody is mid blink.
- Auto frame widens or recomposes a badly framed shot by generating the missing edges.
- AI Enhance combines several fixes and offers you multiple results to choose from.
Two honest caveats. You need a personal Google Account, because these features do not work on Workspace or school accounts. And the branding moves constantly, so if a tutorial names a button you cannot find, look for the same job under a different label.
What an iPhone can do without any account
On Apple’s side, Clean Up in the Photos app does the same distraction removal job. Apple describes it as “a feature powered by Apple Intelligence in the Photos app that can help remove distracting objects in a photo.”
The catch is hardware. Apple lists iPhone 16 models, iPhone 15 Pro and Pro Max, an iPad with A17 Pro or M1 and later, or a Mac with M1 and later, running iOS 18.1, iPadOS 18.1 or macOS Sequoia 15.1 and above. On an older phone, Clean Up will not appear no matter how many times you update.
What makes it useful for privacy minded people is that it runs on the device. No upload, no service account, no credit counter. That is rarer than it sounds, and it is why I reach for it first with anything sensitive.
Where the free versions actually stop
This is where a lot of online listicles are out of date.
Adobe does have a free route. Adobe’s own generative credits FAQ says free users “get a limited number of free daily generations across all generative AI features, available on a curated set of models. These limits refresh each day.” Notice what that does not say. There is no published number, and Adobe states the limits and models can change. Treat it as a daily taste, not a plan.
Canva is the one people get wrong most often. Canva lists its one click Background Remover under Canva Pro, and the only route in from a free account is to start a free Pro trial. Older articles that list Canva’s background remover as simply free are out of date.
Important tip: before you upload a photo of another person to a free web editor, ask whether you would be comfortable telling them you did it. From building client websites, this is the mistake I see repeated. Someone runs a staff photo or a scanned document through a free tool they found on page one of Google, and nobody checks what that site does with the upload. Our guide on using AI safely and protecting your privacy covers the habits worth keeping.
Your edited photo now carries a record
Here is the part that changes how you should think about all of this.
Google Photos now supports Content Credentials. Its help page describes them as “a technology that provides media history and transparency about the use of artificial intelligence.” Open a photo, tap More, then About, and you may see a section called “How this was made” with a line such as “Edited with AI tools.”
The sentence worth reading twice is this one from the same page: if an original image does not have Content Credentials, Google Photos will add them when an AI edit is saved. The record is created by your edit, not inherited from the camera.
Content Credentials come from the C2PA, an open standard for showing “the origin and edits of digital content.” The same page has the clearest beginner description I have seen: they “function like a nutrition label for digital content.” The steering committee includes Adobe, Amazon, BBC, Google, Meta, Microsoft, OpenAI, Sony and TikTok, so this is not one company’s experiment.
Two limits. Google says Content Credentials are not available on Google Photos on the web, only in the apps. And Apple’s support pages do not document any AI edit marker for Clean Up, so do not assume every tool leaves a trail.
Will Instagram or Facebook label your edited photo?
Probably not, and it is worth being precise about why.
Meta’s transparency page on labelling AI content says the “AI Info” label is applied when it detects industry standard AI image indicators or when someone discloses AI content themselves. The same page adds that its methodology “may not capture labels on some content that is edited with AI.”
These systems were built to catch fully AI generated images, not a tidied up holiday photo. Metadata can also be stripped when a file is re-saved or re-uploaded, which is why a label is a helpful signal and never proof. To check an image you were sent rather than one you made, our post on how to check if a photo or video is AI generated walks through what works.
A workflow that keeps you out of trouble
- Start in the app you already have. Free, fast, and no upload to a stranger.
- Keep the original. Both apps let you revert, and you will want that the first time an erase invents something odd.
- Zoom in on the repaired area at full size before sharing. Repeated patterns and fingers are where these tools fail.
- Say so when it matters. For a listing photo, a news image or anything a person will act on, describe what you removed.
Point four sounds fussy until it is your reputation. Removing a bin from a garden photo is housekeeping. Removing damage from something you are selling is a different thing entirely, and the file may now say the picture was edited.
Common Questions
Is AI photo editing free?
Partly. Google Photos gives the core AI edits to any personal Google Account with no subscription, and Apple’s Clean Up is free on supported hardware. Adobe gives free users a limited number of daily generations with no published quota, and Canva lists its background remover as a Canva Pro feature with a free trial.
Does AI photo editing ruin image quality?
The edit itself is usually invisible at normal sizes. The failure mode is not softness but invention. Check text, faces, hands, straight lines and repeated patterns at full zoom, since that is where a generated patch gives itself away.
Can someone tell my photo was edited with AI?
Sometimes. If the tool wrote Content Credentials into the file and nobody stripped them, a viewer in a supporting app can see a note such as “Edited with AI tools.” Screenshots, re-uploads and re-saves often remove that information, so the absence of a label proves nothing either way.
Which AI photo editing tool should a complete beginner start with?
The one already on your phone. Learn Magic Eraser or Clean Up properly before installing anything else, since those two cover most of the fixes people actually need. For more everyday uses of the AI already in your pocket, see our guide to using AI on your phone.
Final takeaway
AI photo editing has quietly become a normal phone feature rather than a specialist skill. Use it. Remove the bin, fix the blur, straighten the group shot.
Just carry two facts with you. The free tier is generous inside Google and Apple and thinner everywhere else, so check before trusting a listicle. And your edit may now be written into the photo itself, which is fine if you were tidying up, and a reason to think twice if you were not.
by admin | Aug 30, 2026 | Future Jobs
Sending out job applications has always been slow, tiring work. Now there is a tool sitting on your desk that can produce a full cover letter in about nine seconds, and the temptation to let it do the whole thing is very real.
The problem is that recruiters are now reading a lot of those nine second letters, and most of them can tell. So the question is not whether you should touch AI at all. The question is where it genuinely helps and where it quietly costs you the interview. This guide walks through how to use AI to apply for jobs in a way that saves you real time without flattening the one thing that actually gets you hired, which is you.
What AI is genuinely good at in a job application
Before we get to the warnings, it is worth being fair to the tools. There are parts of applying for a job that are pure admin, and AI handles them well.
- Reading a long, badly written job advert and pulling out the skills and keywords that actually matter
- Researching the company so you are not walking into an interview blind
- Turning a rambling paragraph about your experience into something tight and readable
- Catching spelling, grammar and clarity problems you stopped seeing an hour ago
- Restructuring a story you already lived into a clear Context, Action, Result shape
- Building a checklist of everything a specific application asks for so nothing gets missed
Notice what is missing from that list. AI does not know you. It has never watched you fix something at eleven at night, or talk a frustrated client down, or teach yourself a tool because nobody else on the team would. That material has to come from your head.
The three step method careers advisers actually recommend
You do not have to invent a process here. University careers services have already worked one out. The University of Manchester Careers Service teaches a simple three step approach: Prepare, Prompt, Proofread. It is boring, and it works.
1. Prepare
Gather your raw material first. The job advert, the responsibilities, your honest reason for wanting the role, and two or three specific things you have done that map onto what they asked for. If you skip this step, AI has nothing real to work with and will hand you a generic letter that reads like every other generic letter in the pile.
2. Prompt
Manchester points to a prompt structure developed at Durham University called RTF: Role, Task, Format. You tell the AI who to be, what to do, and how to give it back to you.
Role: You are an experienced careers adviser who reviews CVs for this sector.
Task: Here is my CV and the job description. Point out where my existing experience matches their requirements and where it does not.
Format: Give me bullet points, with the specific phrasing I should consider and any gaps I should address.
Notice that this prompt asks for feedback, not for a finished document. That single shift keeps the writing yours.
3. Proofread
Read every line back and ask three questions. Does this sound like me? Is every fact in here true about me and about this company? Is it in the format the employer asked for? AI invents things. It will happily give you a year of experience you never had, and that is the kind of error that ends a hiring process badly.
Important tip: never paste a full job description together with your personal details, address or ID information into a random free AI tool without reading its privacy policy first. From my own experience working with websites, online tools and cybersecurity, the free tools are usually free because your data is the product. Strip out anything personal before you paste.
Check the employer rules before you use AI at all
This is the step almost everyone skips. Employers do not agree with each other on this, and some of them tell you exactly where they stand if you bother to look.
The UK Civil Service, for example, publishes an open guide on artificial intelligence and recruitment that spells out what is fine and what is not. Checking spelling and clarity, or asking AI to help you structure an example, sits on the acceptable side. Using AI to supply misleading information, to complete personality or ability tests, or to help you during a live assessment does not.
Other organisations encourage AI use openly. Some ask you to declare it. Some try to detect it. Five minutes on the employer careers page will tell you which one you are dealing with, and that is a cheaper way to find out than being rejected without ever knowing why.
What you should never hand over to AI
- Experience you do not have. If it is not true, it will come apart at interview, and it is a much bigger problem than a weak cover letter.
- Timed or live assessments. Almost every employer treats this as cheating, and many now design the assessment specifically to catch it.
- Your final voice. If the letter does not sound like a person who could walk into the room and say those words out loud, rewrite it.
- Confidential information from a current employer. Do not paste internal documents into a chatbot to help you describe your own work.
Remember that AI is reading your application too
This works in both directions. Research from the Institute of Student Employers, cited by Manchester, found that 28 percent of employers were using AI in recruitment as of November 2023, up from 9 percent the year before. Your carefully written letter may well be sorted by software before a person sees it. We covered what that process actually looks like in our guide to what happens after you click apply.
There is a legal side to this now as well. Under the EU AI Act, systems used for recruitment and for filtering job applications are classified as high risk under Annex III, which brings extra obligations for the employers using them. In practice that means more employers will have to be able to explain how their screening works, which is good news for applicants.
The demand side is moving too. The World Economic Forum Future of Jobs Report 2025 found that 70 percent of employers plan to hire people with new AI related skills. Being able to say honestly that you use AI tools well is starting to count for something, which is why it is worth thinking about how AI skills sit on your CV. If you want the wider picture, our overview of how AI is changing future jobs is a good place to start.
A realistic workflow for one application
- Paste the job advert into your AI tool and ask it to list the top eight requirements in plain language.
- Against each one, write two lines yourself about what you have actually done. Rough notes are fine.
- Ask the AI to tell you which of your points are weak or vague, and why.
- Write the letter yourself using its feedback.
- Ask the AI to check clarity, grammar and length only.
- Read it out loud. If a sentence makes you cringe, cut it.
That whole loop takes about twenty five minutes and produces something a recruiter has not read fifty times already.
Common Questions
Can employers tell if I used AI to write my application?
Sometimes, and not reliably. AI detection tools are not accurate enough to be treated as proof, as we explained in our post on whether AI detectors really work. What experienced recruiters notice is not the software output, it is the flatness. Generic phrasing, no specific detail, and nothing that could only have been written by you.
Should I tell the employer I used AI?
If they ask, yes, always. If they do not ask, using AI for editing and structure is normally treated the same way as using a spellchecker. The line most employers care about is whether the content is honestly yours.
Will using AI get my application rejected automatically?
Not on its own, unless the employer has said clearly that AI is not allowed. What gets applications rejected is a letter that could have been sent to any company for any role.
Which AI tool is best for job applications?
The tool matters far less than the process. A free version of any mainstream assistant, used with the Prepare, Prompt, Proofread method, beats a paid tool used to generate a letter in one click.
Is it worth using AI for the CV as well as the cover letter?
Yes, but for a different job. Use it to check that your CV covers the keywords in the advert and that your formatting is clean and machine readable. Do not let it rewrite your achievements, because that is where invented details creep in.
Final takeaway
The best way to use AI to apply for jobs is as a demanding editor, not as a ghostwriter. Let it read the advert, question your first draft, and clean up your grammar. Keep the thinking, the examples and the voice for yourself. Check the employer rules before you start, never let it invent anything, and read the final version out loud before you send it.
Applications that get interviews are still the ones where a specific person explains why they want a specific job. AI can help you get there faster. It cannot get there for you.
by admin | Aug 29, 2026 | Opportunities
You finished the free course. The certificate is sitting in your email somewhere. And then, honestly, nothing much changed.
That gap between learning about AI and having something to show for it is where most beginners get stuck. AI competitions for beginners are one of the cheapest ways to close it. They cost nothing to enter, they run all year, and when you finish you have a public link to something you actually did rather than another line on a list.
Here is where to start, what each platform is good for, and how to pick a first one you will actually finish.
Why a competition does something a certificate cannot
A certificate says you watched the lessons. A competition entry says you took a messy real problem, made choices about it, and put your answer next to everyone else’s.
From my own experience reviewing people’s work for websites and digital projects, a link to something someone built tells me more in thirty seconds than a list of course titles does in five minutes. The list tells me what they sat through. The link tells me what they can do when nobody is giving them the answer.
There is a second benefit that nobody mentions. A competition has an end. Courses can be paused forever, but a submission deadline makes you finish something, and finishing is the skill most beginners are actually missing.
Start on Kaggle, but stay off the main leaderboard
Kaggle is the biggest home for this kind of thing, and it sorts its challenges into types. Featured competitions are the premier ones with prize money. There are also Research, Community, Simulations and Hackathon categories. None of those are where you begin.
Two categories exist specifically for people in your position. Getting Started is described by Kaggle as approachable machine learning fundamentals, and Playground is described as fun practice problems. Getting Started competitions do not close, so there is no clock running and no pressure to be clever on a schedule.
The ones beginners usually meet first are:
- Titanic: Machine Learning from Disaster, which Kaggle literally labels “Start here”
- Digit Recognizer, for computer vision basics using the classic handwritten digits dataset
- House Prices, for predicting numbers rather than categories
- Natural Language Processing with Disaster Tweets, if text interests you more than tables
- Spaceship Titanic, a friendlier modern remake of the first one
You write your code in Kaggle’s notebooks in the browser, so there is nothing to install and nothing to break on your own machine. If the machine learning ideas behind these are still fuzzy, the free courses in our roundup of free AI courses from Google, Microsoft and Kaggle line up almost exactly with these starter problems.
Devpost, if you would rather build something than tune a model
Not everyone enjoys squeezing accuracy out of a spreadsheet. Devpost runs public hackathons and app contests, including a steady stream of AI ones, and there you submit a working project instead of a predictions file.
That suits beginners better than it sounds. A small tool that solves one annoying problem, explained clearly, often does better than a technically impressive thing nobody understands. Your entries also stay on your Devpost profile afterwards, which quietly turns into a portfolio without you ever having to build a portfolio website.
MLH, if you are a student
Major League Hacking describes itself as the official collegiate hackathon league, and its season calendar lists student hackathons across North America, Europe and Asia Pacific, including events tagged for high school students and events with a diversity focus.
Most of those are in person, which rules them out for a lot of readers. The part that does not is Global Hack Week, which runs online, is open worldwide, and is themed. Recent editions have been built around AI agents, data and generative AI. You join from home, for free, and you are hacking alongside thousands of other people who are also figuring it out as they go.
Free live events, if a competition still feels too big
If entering anything sounds like too much this month, a gentler middle step is a free live session. IBM runs free online SkillsBuild events on AI topics that you can simply attend. We covered what else is on that platform in our guide to IBM SkillsBuild free AI courses.
How to choose your first AI competition
- Pick a problem you can explain to a friend in one sentence. If you cannot, you will lose interest by week two.
- For your first attempt, choose something with no deadline, or one that already closed. You can still enter old competitions to practise.
- Read three public notebooks or past entries before you write anything yourself.
- Give it one week, not a whole summer.
- Enter alone the first time so you learn the whole process, then find a team.
Important tip: your first goal is a submission, not a score. Submit something bad on day one, confirm it went through, and then improve it. Most beginners who quit never got to the point of submitting anything at all.
Read what you are signing up for
This is the part I care about most, because it comes from working on the security side of websites and online tools. Competition and hackathon sign-ups often ask for a lot: your CV, your university, sometimes identity documents if there is prize money involved. That can be perfectly legitimate, but it is worth reading rather than clicking through.
Two habits will keep you out of trouble. First, remember that public notebooks and public repositories are genuinely public, so never upload employer data, client data, or anything containing other people’s personal details. Work only with the data the organisers give you. Second, check the eligibility rules for age and country before you spend three weeks on a prize competition you were never allowed to win. The same “check the terms first” logic applies to free AI student offers, where the catch is usually in the small print rather than the price.
What to do after it ends
Finishing is not the last step. Write up what you tried in plain English, including what did not work, because that is the part experienced people find convincing. Then put the link somewhere it will be seen. Our guide on how to put AI skills on your CV covers how to phrase it without overselling.
If the code itself was the hardest part, that is useful information rather than a failure. It tells you exactly what to study next, and our list of free AI coding courses for beginners is a reasonable place to go from there. Then enter a second one. The second is always easier than the first.
Common Questions
Do I need to be good at coding to enter an AI competition?
You need some, but far less than people assume. Kaggle’s Getting Started competitions come with public notebooks you can read, copy and adapt, and Devpost hackathon teams regularly need people who can design, write, test or present rather than only code.
Are AI competitions for beginners really free to enter?
The ones listed here are. Kaggle competitions, Devpost public hackathons and MLH Global Hack Week do not charge an entry fee, and Kaggle even gives you the notebook environment to run your code in. Travel to an in person hackathon is the one real cost, which is exactly why the online options matter.
Do I need a team?
Not for Kaggle, where solo entries are normal. Hackathons are usually more fun in a team, and most of them run a channel where people without a team find each other in the first hours. Going in alone is a completely normal way to arrive.
Final takeaway
Courses teach you the vocabulary. Competitions are where you find out whether you can use it, and they leave behind something a stranger can look at. Open the Kaggle Getting Started list this week, pick the one that sounds least intimidating, and aim to make one submission by Sunday. It will not be good, and that is completely fine. It will be yours, and it will be more than a certificate.
by admin | Aug 28, 2026 | AI Tools
You are halfway through typing something into a chatbot when you stop. Maybe it is a client contract, a letter about a family health issue, or a draft you are not proud of yet. The question that makes you pause is fair. Where does this text actually go once you press send?
There is a second option most beginners never hear about. You can run AI on your own computer, with no account, no internet connection and no monthly fee. The model sits on your hard drive and answers you locally. This guide covers what that means in plain English, what your laptop needs, the two free apps that do the hard part, and the limits nobody mentions.
What it means to run AI on your own computer
When you use a normal chatbot, your words travel over the internet to a company’s servers. The model lives there, does the thinking there, and sends an answer back. You are renting access to something you never touch.
Local AI flips that around. You download a model file once, and after that the calculation happens on your own processor. No internet needed. Nothing leaves the machine.
The models you can download this way are usually called open-weight models, which simply means the company has published the trained file for anyone to download and use. If the vocabulary is new, our plain-English guides to what a large language model is and the common AI terms explained will make the rest of this post easier.
This is not a fringe hobby any more. In August 2025 OpenAI released two open-weight models, gpt-oss-120b and gpt-oss-20b, under the permissive Apache 2.0 licence, and said the smaller one can run on a device with just 16 GB of memory. Google publishes its Gemma models with a similar goal, describing them as built to run anywhere from cloud servers to laptops and even phones.
Why anyone would bother
Four reasons come up again and again, and only one of them is about saving money.
- Privacy. Text you type into a local model never reaches anyone’s server, so there is no policy to read and no setting to double check.
- It works with no internet. On a plane, on a train with bad signal, or during an outage, a local model still answers.
- No message caps and no subscription. Ollama’s own pricing page states plainly that running models on your own hardware is always unlimited.
- You finally see how it works. Watching a model load into memory, and noticing what a smaller one gets wrong, teaches you more in an evening than a month of reading about AI.
What your computer actually needs
Memory is the deciding factor, not the brand of your laptop. A model has to fit into RAM to run, so the size of the model you can use is set by how much memory you have spare.
LM Studio’s own system requirements put it plainly: 16 GB or more is recommended, and on an 8 GB machine you should stick to smaller models and modest context sizes. Those small models are still useful for summarising, rewriting and simple questions. At 16 GB you get real choice, including OpenAI’s gpt-oss-20b, though that 16 GB is the model’s own requirement, so it will be tight once your operating system takes its share.
You also need disk space, since each model is a file of a few gigabytes upwards. A dedicated graphics card or an Apple Silicon chip makes answers appear noticeably faster, but neither is required. An ordinary laptop will run a small model, just more slowly.
Important tip: start with the smallest model that does your job, not the biggest one your computer can technically load. A fast small model you actually use beats an impressive large one that takes forty seconds per answer and drains your battery.
Two free apps that do the hard part
You do not need to touch Python or compile anything. Two well-known free apps handle the setup for you, and both were launch partners when OpenAI published its open models.
LM Studio is the friendlier starting point if you like buttons and menus. It is a desktop app where you browse and download models inside the app itself, then chat with them in a familiar window. Under the hood it uses the MLX and llama.cpp runtimes, so you get good local performance without configuring any of it. The free tier costs nothing and now covers offline voice transcription as well as chat.
Ollama suits people who are comfortable typing a command, though it now ships desktop apps too. Its free plan costs nothing and runs models on your own hardware, which is the part that matters here, because text a local model reads never leaves your machine in the first place.
One honest caveat that most guides skip. Both companies now also sell access to much bigger models running on their own servers, and they charge for it differently. Ollama’s paid plans start at 20 dollars a month, while LM Studio sells cloud credits you top up as you go. Price is not the thing to watch, though. Ollama’s free plan now lists cloud model access too, so free no longer means local. Those cloud modes are convenient, but they are not local, so the privacy benefit disappears the moment you switch to one. Check which mode you are in before you paste anything sensitive.
A first setup that takes about twenty minutes
- Check your available memory first, so you pick a realistic model instead of guessing.
- Install one app, not both. LM Studio if you want a normal window, Ollama if you do not mind a terminal.
- Download one small model. Resist queueing up five, because you will not compare them fairly anyway.
- Test it on three real tasks from your own week. Summarise an email you already answered, rewrite a paragraph you wrote badly, explain a term you half understand.
- Compare those answers with your usual chatbot, then decide honestly whether the local one is good enough for that kind of task.
That last step matters more than the install. Most people who give up on local AI never tested it against real work, so they had nothing to judge it by.
The limits, honestly
A small model on your laptop is not the equal of the best hosted model, and anyone who says otherwise is selling something. Expect slower answers, weaker reasoning on long problems, and confident mistakes. It also knows nothing about today’s news unless the app’s search tool is switched on, and those built-in search features usually send your query out to a web service, which quietly undoes the offline privacy you installed the thing for. The habit of checking anything that matters still applies, and our guide to using AI safely covers that in more detail.
There is a practical cost too. Running a model works your processor hard, so a laptop gets warm and the battery drops faster than usual.
Where local AI is genuinely the better choice
From my own experience building and maintaining websites for other people, the strongest case is client material. When someone sends you their unreleased copy, their pricing or a database export, running that text through a local model keeps a promise you already made about their data. It removed a decision I used to make far too casually.
The same logic applies at work, where employees often paste company information into whatever free tool they found, a habit now widely called shadow AI. A local model is one of the few answers that keeps people productive without pushing internal documents to an outside service.
It is also the obvious choice for offline study, travel, and anyone on an expensive connection. If you mostly work from a handset, our guide to using AI on your phone is a better place to start.
Common Questions
Is it really free to run AI on your own computer?
The local part is free. The apps cost nothing, the open-weight models are published for free download, and there are no message limits. You pay in disk space, a warmer laptop and some electricity. Any subscription you see advertised is for running models on the company’s cloud servers instead.
Do I need a graphics card?
No. A dedicated GPU or an Apple Silicon chip makes replies come faster, but a standard laptop can run a small model. Speed is the thing you trade away, not the ability to run it at all.
Is a local model as good as ChatGPT or Gemini?
Not in general, no. For everyday jobs like summarising, tidying up your writing, drafting a reply or explaining a concept, a good small model is often close enough. For hard reasoning, long documents and current information, the hosted models are still clearly ahead.
Does it work with no internet at all?
Yes, once the model is downloaded. You need a connection to fetch the file the first time, and after that you can switch off the wifi and keep working. LM Studio documents this directly, saying the app can operate entirely offline once you have the model files.
Final takeaway
You do not have to pick one and abandon the other. Most people end up with both, using a hosted model for heavy thinking and a local one for anything private, offline or repetitive. The useful move this week is small: install one app, download one small model, put three real tasks through it. That evening will tell you more about whether you should run AI on your own computer than any comparison table will.
