Learning AI After Work — A Hands-On Review of Seven Free Tools

Translator’s Note: This is a machine translation of the original Chinese article. The Chinese version is the authoritative text. Some culturally specific terms (e.g., “沙紙”, a Cantonese colloquialism for academic credentials) have been adapted for English readers.

Learning AI After Work — A Hands-On Review of Seven Free Tools

What’s your educational background? In today’s Hong Kong, a university degree seems to have become a universal must-have. Of course, there are countless paths that don’t require one, but a degree certainly makes it easier to navigate the traditional system. What’s become increasingly severe in recent years is credential inflation — a bachelor’s degree no longer stands out; you need a master’s to count. Quite a few of my classmates earned their master’s degrees within the first few years of entering the workforce.

The AI Era: Beyond the Paper Chase

The AI era is different. Many of the AI courses offered by institutions today are really just repackaged fundamentals, with cutting-edge technology and future directions given only a cursory nod. AI is evolving even faster than the computer industry ever did — what you learned in a course last year may already look outdated today.

In the AI era, hands-on experience speaks louder than theoretical knowledge. I believe that diving in personally, stepping into this uncharted territory, is far more effective than waiting for a course to spoon-feed you.

The After-Work Time Crunch

The biggest challenge for the average office worker is time. What time do you clock out? What’s waiting for you after work? Cooking? The gym? Housework? Childcare? After all the daily obligations are subtracted, you’re basically ready for bed. To self-learn anything new, time is the scarcest resource.

To make the most of what little personal time remains for learning, you first have to let go of the “binge-and-finish” mindset. Instead, use fragmented pockets of time, lean on AI to assist your learning, and aim for one small step each day.

Seven Free AI Tools: A Hands-On Review

Before the AI explosion of 2025, there were already several smaller waves — tools like POE were already available online. But this article is about the tools I actually use right now. Below, I’ll walk through several tools I’ve used from my early exploration days through to the present, with some brief thoughts on each.

POE

POE was the tool I used in my earliest exploration phase, roughly three years ago during the ChatGPT hype. You can choose from various models and bots, but the interface makes it genuinely difficult to figure out which model or bot suits you, or even whether you’re using the “real thing.” At first I treated it purely as a toy, but gradually its shortcomings became apparent — no web search capability, and limited by knowledge cutoff dates. I phased it out over time.

Perplexity

I can’t recall exactly when I started using it, but I remember Perplexity’s reputation for rigorous fact-checking and being well-suited for research. I used the free tier briefly. It was indeed fast, and I encountered no hallucination issues. But whether due to strict internal moderation or the limitations of the free tier, for longer or multi-step questions, Perplexity could only offer broad directions or generic advice — no concrete answers. So I only use it occasionally, as a supplementary research tool.

Grok

I remember first using Grok in March 2025, and it was immediately striking — I’d barely touched AI for either work or personal use in the preceding period, and Grok’s performance made me realize just how far AI had advanced. It could search the web for the latest information and handle both general and specialized questions. What stood out most was its tone and comprehension, both a clear step above anything I’d used before.

Grok’s strongest suit is handling complex problems directly — coding, mathematical computation, with reasoning laid out in a coherent chain. But recent updates feel like deliberate throttling and rate-limiting to push users toward paid tiers. It’s also overconfident, often ignoring its own training cutoff and serving outdated answers. Overall, still satisfactory, and it remains one of my go-to tools.

DeepSeek

I started using DeepSeek around the same time as Grok. Compared to other AIs, it carries a noticeable “I know everything” air — an imperious posture that doesn’t welcome being questioned. That said, its answers are reasonably reliable. After several revisions, it now supports deep reasoning and web search, a significant improvement over earlier versions. I still use DeepSeek as an independent reference source, cross-checking it against other tools.

Copilot

I honestly don’t remember when Copilot appeared on my Windows desktop. I wouldn’t call it good. Perhaps because it’s free, Copilot’s answers are the least trustworthy — always agreeing with the user, and the information is often outdated. I rarely use it.

AnyGen

I came across AnyGen in early 2026. It feels distinct from other tools — more like an intelligent agent that can independently produce files and deliver satisfactory answers. But downloading the files it generates mostly requires a paid membership, and its associated ecosystem is Lark (Feishu), which isn’t part of my usual toolset. So I typically only use it for longer analyses, treating it as one independent reference among several.

Google Gemini

I was already using Gemini through other channels before it officially became available in Hong Kong. Since its full Hong Kong launch in March 2026, I’ve been using it even more. Its biggest advantage is leveraging Google Search to inform its responses, so answers rarely feel outdated. That said, Gemini’s answer quality is tied to the quality of its search results — not always the best answer.

Another plus is its integration with other Google tools like Gmail and Google Drive. The underlying model is no slouch either, capable of giving satisfactory answers to most questions. If I had to nitpick, it leans too heavily toward agreeing with the user, making it hard to gauge whether your own thinking is actually correct. It’s one of my primary tools.

The Limits: Assistance, Not Replacement

At the end of the day, all of the above tools are just that — tools. They have no real productive power in themselves. At best, they serve as “consultants” — execution still falls to the user. That’s not to say they’re useless, but their utility has limits. For long-term projects or complex, multi-step tasks, these tools struggle to deliver.

That said, as learning aids and troubleshooting companions while you do the actual hands-on work, they’re perfectly capable. My entire AI learning setup was built with the help of Grok and Gemini. The most important thing is simply to pick a tool that works for you — and all seven of the above are free.

This article reflects the author’s personal experience. Tool performance may vary with version updates.