Loading...
Loading...
Loading community...
<community townhall>
The central hub for intelligence exchange. Ask questions, find answers, and contribute to the collective knowledge of Kracked operatives.
frequent tags
Work Local, Earn Global A Malaysian Freelancer's Playbook for Landing UK Clients A field guide for software devs, AI builders, and UI/UX designers in Southeast Asia who are tired of competing on Upwork for $15/hour. Overview After 10 years working in London, I noticed something obvious that no one in Southeast Asia seems to be acting on: UK companies are already hiring developers and AI builders remotely from the Philippines and Vietnam. They've built entire engineering teams across the region. Malaysia, somehow, is barely on their radar — despite stronger English, better infrastructure, and a closer time zone match for hybrid teams. This is not a "manifest abundance" post. This is a playbook. The exact steps I'd take if I were starting over today from KL, Penang, or anywhere with a stable internet connection — to land my first UK client within 60–90 days. It bypasses the gig platforms entirely. It's built on direct outreach, a sharp portfolio, and consistent personal branding. It's slower to ramp than Upwork, but the ceiling is 10–20x higher, and the clients are real businesses, not the rate-shopping freelance buyers most of us are stuck with. If you commit to this for one quarter, you'll have something most freelancers never build: a pipeline you own. Table of Contents 1. The UK Freelance Market — Why Now, Why You 2. The Three Ways to Work With UK Companies (Permanent · Contract · Freelance) 3. The Old Way: Gig Platforms and Why They're a Trap 4. The New Way: Direct Outreach (the core of this guide) - 4.1 Build Your Target List - 4.2 LinkedIn First, Email Second - 4.3 Polish Your Portfolio - 4.4 The Outreach Sequence - 4.5 The LinkedIn Brand Engine 5. Tailoring the Strategy: Designer vs Developer vs AI Builder 6. The 90-Day Action Plan 7. Common Mistakes and How to Avoid Them Part 1 — The UK Freelance Market: Why Now, Why You The UK is one of the most freelancer-friendly economies in the world, and most Southeast Asian talent has no idea. A few realities worth internalising: - The UK has over 4 million self-employed workers, and post-Brexit + post-COVID, hiring contractors remotely is normal, not exotic. - London-based startups in particular are aggressive remote hirers. Their local engineering rates are eye-watering (£600–£900/day for a mid-level contractor), so anyone delivering quality work at £200–£400/day is a no-brainer to them. That's still RM 1,200–2,400 per day — far above local Malaysian rates. - They already hire from the Philippines and Vietnam. Whole agencies and pods exist serving UK clients from Manila, Cebu, Ho Chi Minh, and Hanoi. The infrastructure of "UK company + SEA contractor" is well-established. You're not pioneering — you're catching up. - The time zone works in your favour. Malaysia is 7–8 hours ahead of London. You can do deep focused work in the morning, then have 2–4 hours of overlap in the afternoon for standups and calls. UK clients love this. - English fluency is a real edge. Malaysian English is closer to UK English than almost any other SEA country's. This matters for client comfort more than people admit. Who this guide is for: - Software developers (frontend, backend, full-stack, mobile) - AI builders (LLM engineering, ML, RAG, agents, fine-tuning) - UI/UX designers - Technical researchers and data engineers Who this guide is NOT for: - People looking for passive income - People not willing to do outbound work for 60–90 days before seeing results - People who haven't built anything they can show Part 2 — The Three Ways to Work With UK Companies Before you start outreaching, get clear on what you're actually selling. There are three models: 1. Permanent (Remote Employee) You're a full-time employee of a UK company, working remotely from Malaysia. Usually requires the company to use an Employer of Record (EOR) service like Deel, [Remote.com](http://Remote.com), or Oyster. They handle compliance, payroll, and contracts. - Pros: Stable income, benefits, sometimes equity in startups. - Cons: Harder to land (fewer companies do this), salary often "localised" downward. - Best for: People who want stability and a single employer relationship. 2. Contract (Fixed-Term Engagement) You're hired for a set period (3, 6, 12 months) at a day rate. The UK has a massive contractor culture. You'd typically work through your own Malaysian Sdn Bhd or sole proprietorship and invoice in GBP. - Pros: Much higher day rates, clear scope, easier to land than perm. - Cons: No benefits, you handle your own tax (in Malaysia), gaps between contracts. - Best for: Experienced devs and designers who can deliver outcomes, not hours. 3. Freelance / Project-Based Pay per project or per deliverable. Could be a 2-week landing page build, a 3-month MVP, or ongoing retainer work (e.g., "fractional design lead, 2 days a week"). - Pros: Flexible, multiple clients, recurring revenue once you build a roster. - Cons: Constant sales cycle, less predictable income early on. - Best for: Anyone starting out. This is where most people should begin. > My recommendation: Start with freelance project work, build relationships with 2–3 strong clients, then graduate the best one into a retainer or contract. That's how you go from chasing work to having work chase you. Part 3 — The Old Way: Gig Platforms and Why They're a Trap Let's talk about Upwork, Fiverr, Toptal, [Freelancer.com](http://Freelancer.com), PeoplePerHour, and the rest. These platforms work technically. People land work on them every day. But understand what you're signing up for: - You compete on price with the entire world. A junior dev in Bangladesh quoting $8/hour is bidding on the same job as you. You can't out-cheap that. You shouldn't try. - The platform owns the relationship. The client is the platform's client, not yours. Try contacting them off-platform and you'll get banned. - 15–20% fees. Often more when you factor in connect tokens, premium memberships, and withdrawal fees. - Race-to-the-bottom buyers. The clients who use these platforms are explicitly looking for cheap labour. The good clients — the ones with budget and respect for craft — are not posting jobs on Upwork. They're being referred by friends or hiring through their network. - No compounding asset. When you finish a $500 gig on Fiverr, you have $500. You don't have a brand, a list, an audience, or a reputation outside that platform's walled garden. Toptal is the one exception worth mentioning — it's a curated network with real rates — but their screening is brutal (3% acceptance rate) and they take a heavy cut. The lesson: Gig platforms are fine for your first gig if you have literally zero portfolio and need a screenshot for credibility. Beyond that, every hour spent on them is an hour not spent building something that compounds. Part 4 — The New Way: Direct Outreach This is the core of the playbook. Five stages. Each builds on the last. 4.1 — Build Your Target List Stop being vague. "I want UK clients" is not a strategy. You need a list — a real spreadsheet — of 200–500 specific UK businesses you'd love to work with. Where to source them: - Crunchbase — filter by UK location, funding stage (Seed → Series B is the sweet spot), and industry. These companies have money but aren't big enough to have built out in-house teams. - YC company list (UK-based) — filter by location, sort by recency. Many are hiring fractionally. - Tech Nation reports, [Sifted.eu](http://Sifted.eu), UKTN — these publications profile UK startups constantly. Treat them as lead sources. - LinkedIn Sales Navigator — filter by location (Greater London, Manchester, Edinburgh, Bristol), company size (10–200), and industry. Worth the £80/month for one or two months. - [Apollo.io](http://Apollo.io), Clay, Instantly — for scraping company + decision-maker data at scale. - AngelList / Wellfound — UK startups, sorted by stage. - Companies House ([companieshouse.gov.uk](http://companieshouse.gov.uk)) — every UK business is registered here. Free. You can filter by industry code (SIC) and incorporation date to find recently founded companies. Your target persona: The ideal first client is a UK startup or small business with 10–100 employees, founded in the last 1–5 years, with some funding but no in-house design or dev team yet. Big enough to pay, small enough to talk to the founder directly. Build a spreadsheet with these columns: | Company | Founder/Decision Maker | Their LinkedIn | Company Website | Their Email | Industry | Why They're a Fit | Status | Aim for 200 rows in the first two weeks. This list is your asset. Maintain it, update it, never lose it. 4.2 — LinkedIn First, Email Second Cold email to a stranger is easy to ignore. Cold email to someone you've already shown up for on LinkedIn for two weeks is very hard to ignore. The play: 1. Send connection requests to every decision-maker on your list — without a note. (Counterintuitive, but acceptance rates are higher without notes for the first round.) Stay under 100 a week to avoid LinkedIn flagging you. 2. Engage with their content for 1–2 weeks. Real comments, not "Great post! 🔥". Add a thought, ask a question, share a relevant example. You're showing up in their notifications. They're starting to recognise your face. 3. THEN you reach out. By the time your DM or email lands, you're not a stranger. You're "that thoughtful person who's been commenting on my posts." This single shift — engagement before outreach — will multiply your reply rate 3–5x over cold cold outreach. Tools to make this easier: - [Hunter.io](http://Hunter.io), Apollo, [Snov.io](http://Snov.io) — find email addresses from LinkedIn profiles or company domains. - Clay — combines scraping, enrichment, and outreach. Powerful, has a learning curve. - LinkedIn Helper, Dripify — automate connection requests (use carefully, LinkedIn doesn't love these). 4.3 — Polish Your Portfolio Before you send a single outreach message, your portfolio has to be ready. Because the moment someone clicks your name, they decide in 8 seconds whether to take you seriously. Your portfolio must answer four questions in under 30 seconds: 1. Who are you? (One sentence. Not "creative problem-solver passionate about tech." Try: "AI engineer building production RAG systems. Previously at [company]. Based in KL, working with teams worldwide.") 2. What do you do? (Specific. "I help early-stage SaaS founders ship their MVPs in 6–8 weeks." Or "I build LLM-powered internal tools for non-technical teams.") 3. What have you done? (3–6 strong case studies. Quality over quantity.) 4. How do they reach you? (Email + Calendly link. Don't make them dig.) Format by role: - Designers → Visual-first. Behance, a custom Framer/Webflow site, or a clean Notion site. Show the work. Include process, but lead with hero shots. - Developers → GitHub + a personal site. Pin your best 4–6 repos. Each repo has a real README with screenshots, demo links, and "what I learned." Write 2–3 technical blog posts on [dev.to](http://dev.to) or your own site. - AI builders → Live demos. Always live demos. A Hugging Face space, a Vercel-deployed app, a Loom walkthrough. Talking about LLMs isn't enough — show the agent working. - Researchers → A site that links to papers, talks, and writing. Substack or a Notion site works fine. Lead with the most cited or interesting work. Case study structure (this is the formula): 1. The problem (1 paragraph) 2. The constraints (1 paragraph — budget, timeline, team size) 3. What you built (visuals, code snippets, architecture diagrams) 4. The outcome (numbers if you have them: users, revenue, time saved) 5. What you learned (this is what makes you sound human and senior, not junior) 4.4 — The Outreach Sequence You have your list. You've warmed leads on LinkedIn. Your portfolio is sharp. Now you reach out. The cadence (per lead): Touch 1 — LinkedIn DM (Day 0) Short. Specific. Mentions something real about their company. Ends with a soft question or a "no pressure" offer. > Hey [Name] — been following [Company] since you launched [feature]. Genuinely impressed with how you handled [specific thing]. I'm a [your role] based in KL, work with UK startups remotely. If you ever need a hand with [specific area], happy to chat. No pitch — just here if useful. Touch 2 — Email (Day 3–4) Slightly more substantive. Reference the LinkedIn message. Add value — link to a relevant case study or share a quick observation about their product. Touch 3 — LinkedIn DM follow-up (Day 8–10) One-liner. "Hey, just floating this back up in case you missed it — let me know if useful, otherwise I'll stop bothering you 🙂" Touch 4 — Final email (Day 14) "Last one from me. If now isn't right, totally understand. I'll keep following [Company]'s journey — and the offer stands whenever you need it." After 4 touches with no response, move on. Add them to a "nurture" list and re-engage in 3 months. Volume: 10–20 outreach sequences per week is realistic when starting out. That's 40–80 contacted per month, ~600 per year. Expect a 5–10% reply rate, of which 10–20% become real conversations, of which 20–30% become paid work. Do the maths — that's roughly 1–4 clients per month at scale, plenty to fill your runway. 4.5 — The LinkedIn Brand Engine Outreach gets you the first conversation. Brand gets people sliding into your DMs. This is the long game, and it runs in parallel to everything above. The rule: Post 3–5 times a week on LinkedIn for 6 months. Minimum. What to post (by role): - Designers → Daily work-in-progress shots, finished screens, before/afters, design teardowns of popular apps. Carousels work brilliantly. This is a brand and marketing play — every post is a billboard for what you do. - Software developers → Build-in-public posts. "Today I shipped X." Technical breakdowns. Lessons from debugging. Architecture decisions explained simply. - AI builders → Demo videos of agents, prompts that worked, RAG architectures, fine-tuning experiments, failures and what you learned. - Researchers → Paper summaries, your own findings, threads explaining complex ideas simply. Format that works: - Hook in line 1 (controversial, specific, or curiosity-piquing) - 1 line of context - Body in short paragraphs - One image, video, or carousel — always - A question at the end to drive comments Why this matters: When a UK founder gets your cold message, the first thing they do is click your profile. If they see consistent, quality posts about your craft, you go from "random freelancer" to "credible operator with a real point of view." Conversion rates on outreach roughly double when your LinkedIn is active. Part 5 — Tailoring the Strategy by Role The playbook is the same. The expression of it changes. If you're a UI/UX Designer - Lead with visuals everywhere — LinkedIn, portfolio, even your email signature - Behance + Dribbble are still relevant; Framer/Webflow site is mandatory - Case studies show process, not just final screens - Target product-led startups, agencies, and DTC brands - Day rate range: £250–£500 starting If you're a Software Developer - GitHub is your real portfolio. Pin and polish 4–6 repos. - Write technical blog posts. Even 5 of them changes how you're perceived. - Target SaaS startups, agencies, and product teams without senior backend talent - Specialise — "React + Node generalist" is forgettable; "ships production Stripe integrations for B2B SaaS" is memorable - Day rate range: £300–£600 starting If you're an AI Builder - Live demos beat case studies. Always. - Open source one or two of your tools. Even small ones. - Write about specific implementations: "How I built a RAG system over 50k support tickets" beats "what is RAG." - Target startups building AI-native products, agencies offering AI services, traditional businesses just starting their AI journey - Day rate range: £400–£800 starting (highest of all three) If you're a Researcher - Personal site with paper list, talks, and a blog - Speak at meetups (UK has tons — many run hybrid) - Target R&D teams, AI labs, deep-tech startups - Different conversion path — usually longer cycle, more about credibility, often higher-value engagements Part 6 — The 90-Day Action Plan Days 1–14 — Foundation - Audit and rebuild your portfolio - Pick your niche (be specific) - Write 3 case studies in the format above - Set up Calendly + a clean email signature - Build your initial target list of 200 companies Days 15–30 — Warm Up - Send 50–100 LinkedIn connection requests per week - Start posting 3x/week on LinkedIn - Comment thoughtfully on 5–10 posts per day from your target list - Get your scraping + email enrichment workflow running Days 31–60 — Outreach At Volume - Send 10–20 full outreach sequences per week - Continue posting and engaging - Track everything in your spreadsheet (open rates, reply rates, conversions) - Take every call, even ones that feel like a stretch — practice matters Days 61–90 — Close and Iterate - Convert first 1–3 clients - Ask delighted clients for written testimonials and a video quote - Refine your messaging based on what's converting - Raise your rates by 20% once you have 2 paid clients By Day 90, if you've actually done the work, you should have 1–3 paying UK clients, a growing LinkedIn presence, and a pipeline that doesn't depend on a platform. Part 7 — Common Mistakes and How to Avoid Them 1. Mass-blast outreach. Personalise every message. Generic templates get auto-deleted. 2. No portfolio, hoping to "talk about projects on a call." No call will happen. Portfolio first. 3. Discounting yourself because you're "based in Malaysia." Your rate is your rate. Geography is irrelevant if your work is good. Don't apologise for where you live. 4. Ghosting after the first message doesn't get a reply. 80% of replies come on touch 2, 3, or 4. Follow up. 5. Posting once on LinkedIn and giving up because it got 12 likes. It compounds. The 50th post lands very differently than the 5th. 6. Trying to be everything to everyone. "Full-stack developer" is forgettable. "Builds Shopify apps for UK D2C brands" gets remembered. Narrow first, expand later. 7. Not invoicing properly. Set up a Sdn Bhd or sole proprietorship in Malaysia. Invoice in GBP. Use Wise or Payoneer to receive payments. Don't get paid through Western Union like it's 2008. Closing The remote door is already open. Filipino and Vietnamese developers walked through it years ago. The infrastructure exists. UK companies are paying. The only thing missing is you doing the work — for 90 days, consistently, without quitting after the first batch of unanswered messages. This isn't a hack. It's not a shortcut. It's the slow, compounding work of building a reputation and a pipeline that you own. But on the other side of those 90 days, you stop being a freelancer fighting for $20/hour gigs and start being an operator running a real business — from KL, working with companies on the other side of the world, charging what your work is actually worth. Stop waiting. Build the list. Send the messages. Post the work. Repeat. 🇲🇾 → 🌍 If this was useful, share it with one freelance friend in Southeast Asia who needs to read it.
Which type of project concept would you like the next Bounty to be about?
After 10 years Wroking in London, here’s the play for Malaysians who want to work with UK SMEs in data, software dev, and marketing — without moving there. What UK SMEs actually outsource UK small businesses (especially <10 people) constantly need help with: - Data: dashboards, reporting, automation, cleaning - Software dev: internal tools, client portals, lightweight SaaS - Marketing: landing pages, email, ads, content They don’t care where you live. They care: “Can you solve my problem and show proof?” That’s where your portfolio and outreach come in. How to find them Here’s a simple outbound system you can run from Malaysia: 1. Go on LinkedIn and search for: - “Founder” / “Owner” / “Managing Director” - Filter by UK, company size: 1–10 employees, and industries you understand. 2. Click into the companies: - Check if they look like they use digital tools but clearly have gaps (messy website, no analytics, manual processes, weak marketing). 3. Build a simple list: - Company name - Website - Founder/Director/CEO name - LinkedIn profile URL - Email (if visible on website) This becomes your UK SME target list. How to reach out (and what to offer) Don’t overcomplicate it. Your angle: “I help small teams do more with data/software/marketing — low friction, high value.” Example DM/email structure: - Line 1: Personal relevance “Hey [Name], saw [Company] on LinkedIn and loved what you’re doing in [industry].” - Line 2: Observed problem “Noticed you’re likely juggling a lot with a small team — especially around [data/software/marketing].” - Line 3: Clear offer “I help small UK companies like yours with [dashboards / internal tools / landing pages etc.] so you get [result: clearer numbers, fewer manual tasks, more leads].” - Line 4: Low-risk hook “If you’re open to it, I can do a free 1-week trial where I [build X / clean up Y / set up Z] so you can see if it’s useful — no obligation.” - Line 5: Simple CTA “Want me to send 2–3 ideas tailored to your business?” You’re not begging for a job. You’re offering a test drive. Why small teams & free trial? - Small teams move fast, don’t have procurement drama. - They feel the pain of messy data, broken tools, and weak marketing more intensely. - A free 1-week trial removes risk and shows you’re confident in your skills. Do this consistently: - 20–30 targeted messages per day - Tight portfolio with 3–5 strong examples - Follow-ups that add value (small loom video, quick audit, a simple dashboard mockup) You only need a few yeses to change your income.
There’s a lot of talk about "Vibe Coding" right now—using Cursor, Lovable, or Bolt to prompt entire apps into existence without touching much boilerplate. On one hand, it’s a superpower. We’re shipping features in 10 minutes that used to take 10 hours. We’re "architects" now, not just "syntax writers." On the other hand... are we losing the "soul" of engineering? If the AI hallucinates a security flaw or an inefficient database query, and the builder doesn't know how to read the underlying code to fix it, are we just building houses of cards? I want to hear from the Kracked community: 1. Do you think a "Vibe Coder" who has never learned syntax can be a "Senior Developer" in 2026? 2. What’s one thing you’ve built recently that you honestly wouldn’t have been able to code without AI help? 3. Where do you draw the line? (e.g., "I vibe code the UI, but I manually write the Auth/Logic"). Let’s settle this: Is the "syntax-first" developer becoming obsolete, or are they more important than ever? ⚔️
> KrackedDevs is building Malaysia’s boldest builder ecosystem: a live map of developers, creators, and tinkerers in every state, all learning AI‑powered software by actually shipping projects. Our mission is simple: Turn hidden talent — gig drivers, fresh grads, kampung kids, career‑switchers — into confident builders who can solve real problems for their communities and the country. Why State Ambassadors? To make this real, we need local anchors: people who know their state, its scenes, and its builders. KrackedDevs State Ambassadors help us light up every corner of Malaysia so that when a GLC, SME, or agency needs talent, there’s a ready pool of real builders, not just CVs. We’re hiring 1 paid, part‑time Ambassador in each state: Johor, Kedah, Kelantan, Melaka, Negeri Sembilan, Pahang, Perak, Perlis, Penang, Sabah, Sarawak, Selangor, and Terengganu (RM1,000/month). Your job: connect local builders, run weekly vibe‑coding sessions, and grow a real on‑the‑ground community around AI‑powered software building.  What Ambassadors Do - Host weekly “vibe coding” sessions in your state (online or IRL). - Keep a pulse on your state’s builders: what they’re working on, what they need, who’s stuck. - Onboard new people weekly who can code (or are actively learning) and are happy to stream two nights a week. - Organise casual weekly coffee meetups so builders can connect face‑to‑face. - Champion KrackedDevs to potential users, local companies, SMEs, and GLCs who need talent. - Reach out to local/state government, agencies, and hubs to explore collaborations and opportunities. - Share regular updates with the KD core team about what’s happening in your state. - Report back to the core KD team weekly, and figure out how to make each week better than the last. Who We’re Looking For - You love AI, Vibe Coding, and want to bring AI to everyone in your state! - You’re based in the state you want to represent and plugged into local tech / student / startup / maker circles. - You enjoy bringing people together and making them feel welcome. - You’re comfortable speaking in English and Bahasa Malaysia (other dialects are a bonus). - You’re active online (X, Threads, Discord, etc.) and not shy to post, stream, or host. - You can commit to a part‑time role at RM1,000/month for at least 3–6 months. - You have stable internet and a decent machine, and you’re willing to learn any tools needed for streaming and community work. 3‑Day Trial Task (How We Choose Ambassadors) Instead of just interviews, we run a simple 3‑day trial that mirrors the actual work. Day 1 – Find your builders - Invite at least 5 local builders from your state into KrackedDevs spaces (Discord/Threads/WhatsApp etc.). - Post a short “state check‑in” on your socials (e.g. “Any builders in Johor? Here’s what we’re doing this week with KrackedDevs”) and tag KD. - Send me screenshots: who you invited, where you found them, and links to your posts. DM me on Discord, @enonforetsam [https://discordapp.com/users/562974638603370497](https://discordapp.com/users/562974638603370497) Day 2 – Run a tiny vibe‑coding session - Host a 30–60 minute micro‑session (online is fine) with whoever you pulled in. - Do something simple: co‑build a small feature, fix bugs, or walk through an AI‑coding workflow. - Send a short recap (3–5 bullets): who joined, what you did, what you’d improve next time. Day 3 – Tell the story and think ahead - Post a recap: “What we did this week in [State] with KrackedDevs” plus a clear CTA to join. - Do one lightweight outreach to an institution (uni club, startup, hub, agency, state‑linked org) to introduce KrackedDevs and share the message you sent. - Fill a short form/doc: links to posts, screenshots, people you brought in, and your 5–7 bullet “next 30 days in my state” plan. From this, we can see who can actually move people, communicate clearly, and think like a state‑level community builder in just 72 hours. How to Apply - Share a short intro below in the replies: who you are, which state, and why you care about building Malaysia’s builders. - Include your social links (X/Threads/LinkedIn/etc.) and any communities you already help run. - Tell us one concrete idea you’d execute in your first month as a KrackedDevs Ambassador in your state. We’re looking for people who don’t just talk about community, but can actually create momentum where they live. For any questions, DM me (Danial) on Discord, @enonforetsam [https://discordapp.com/users/562974638603370497](https://discordapp.com/users/562974638603370497)
Hi everyone! First of all, I want to extend a massive thank you to every single one of you who participated in our first ever Krackathon. We saw 38 submissions on our very first try and that is far beyond what we imagined, and the quality of work has been truly inspiring. I also want to personally apologize for any technical hurdles or "buggy" moments you encountered during the submission process. While we are technically in beta, I take full ownership for not having every corner of the system battle-tested for an event of this scale. Thank you for your patience and for sticking with us through the growing pains. As the developer for [KrackedDevs.com](http://KrackedDevs.com), my goal is to build the most robust and mature platform possible for our community. To do that, I need your "unfiltered" input. Whether it's a critique of the UI, a recommendation for the next hackathon, or a specific bug you found, please let me know in the comments below. Your feedback is the fuel that will help KD level up. Best regards, Jinbei
I've been thinking Current AI and LLM'S do not understand words, their meaning or the knowledge behind them. They see them in mathematical code only as that's the only language machines currently understand. Tokens as they call it assign an id to words, in pretraining learned to match to each-other. So love is 567 and you 78. But 567 doesn't explain what it fundamentally means what love it, the definition, true use, synonyms, homonyms or association. It just statistically guesses the best matches to closest other id tokens for best output to your query. In current LLM'S very accurately and with seemingly good responses. But still no understanding of what and why it said it. It's like a human reading a scientific paper infront of an audience he didn't write in a field he isn't in. Yeah it's brilliant, but he doesn't even know what he's saying. Is he now very intelligent and a good reasoner?