Let me start by abruptly inserting the text that I fed some chatbots with. It goes as follows:

Years ago, various online jobs that didn’t require much expertise bar a perfect knowledge of a language seemed questionable to me, even if apparently legitimate. Beyond the payment that seemed low, I found their basic mechanism stupid: If I was in Germany but not a German, I couldn’t enroll, because I needed a perfect mastery of German. It didn’t matter that I could perform in English and French, and that the job was purely online. It has to be “about the local culture” because such tasks could include, e.g., “evaluating the impact of an advertisement.” OK, so I forgot about them.

Now, thanks to an ad on Facebook, I discovered that there are many such jobs targeting various tasks meant to help train AI. Most likely, not just the major LLMs but also customized agents, depending on the client. From the platforms offering such gig jobs, I selected the following ones to analyze.

Platforms:

§1. Outlier.AI and Remotasks (subsidiaries of Scale AI)

(a.) Outlier: “40,000 experts, 3.4M assignments, $100M paid out.” It doesn’t look that well-paid to me!

“AI training is never boring! Apply for an opportunity in your area of expertise.”

“Outlier helps the world’s most innovative companies improve their AI models by providing human feedback. Are you an experienced writer who would like to lend your expertise to train AI models?”

“How much will I be paid? Competitive rates are set based on the domain of focus and your qualifications.”

OK, but the hourly rates are set by location, not by language, despite being online jobs! The advertisement shown (on Facebook) to people in or from Romania claims “up to $15/hr,” but in other countries it varies:

  • https://outlier.ai/languages/ro-ro – Language: Romanian – “Earn up to $15/hr”
  • https://outlier.ai/languages/fr-fr – Language: French – “Earn up to $25/hr”
  • https://outlier.ai/languages/de-de – Language: German – “Earn up to $29/hr”

For tasks involving texts:

  • “Currently, pay rates for core project work by Romanian writing experts average USD $15 per hour.”
  • “Currently, pay rates for core project work by Italian writing experts average USD $20 per hour.”
  • “Currently, pay rates for core project work by French writing experts average USD $25 per hour.”
  • “Currently, pay rates for core project work by German writing experts average USD $29 per hour.”

For experts:

  • “Currently, pay rates for core project work by Math experts in France/Germany/Italy/UK range from $30 to $50 USD per hour.”
  • “AI Training for Sociology” pays “up to $30, depending on your level of expertise.”
  • “Higher rates might be offered for PhD.”

Other jobs:

  • US, Canada: “Currently, pay rates for core project work for voice recording are up to $40 USD per hour.”

Unfortunately, most jobs (or tasks) are for the US, Canada, the UK, Australia.

From the Reddit threads I examined, I gathered contradictory testimonials:

  • The onboarding process tends to be long, complex, and time-consuming.
  • Some people are happy, but many are not.
  • Complaints about the lack of assignments for up to several weeks.
  • Complaints about the long onboarding followed by nothing.
  • Complaints about being kicked off a job or even having the account closed without any explanation or based on alleged use of AI, which wasn’t the case. But those rejections were automated, and no human examined them.
  • Complaints about abusive downgrading of one’s status, without justification and means to appeal.
  • Complaints about tasks starting e.g. at midnight with expiration in, say, 12 or 15 hours, and a task needing 8 hours. If one cannot end it in the required timeframe, all work is lost, meaning no payment!
  • Complaints about onboarding or preparatory tests requiring you to say that a true statement is false or that a false statement is true, because the tests are wrong. PhDs or experts have had such experiences.
  • PhDs and experts complained that expertise is not properly rewarded, and that “regular people” with little or no expertise are preferred.

In such conditions (wrong assessment tests, etc.), I suspect that AI models will become worse and worse if trained by the people selected by Outlier.AI!

I also question the reliability of such a side job. Many people say one has to work at least 30-40 hours a week, and during the night too, so it’s a tough decision.

(b.) Remotasks: “Meet entrepreneurial taskers like you. Join a community of 240,000+ taskers in 90 countries.”

“Learn how to do tasks with our quick online courses or free hands-on training + Work on tasks from projects you’ve unlocked + Get paid based on your quality & number of tasks completed.”

Tried to sign up with Romania: “Unavailable country.” Wow. 90 countries, but not a EU one!

§2. Pareto.AI

From Reddit, I learned that it used to be largely aimed at providing employment for women in developing countries, mostly in Africa, and specifically Kenya and Nigeria.

Pareto’s new contractor agreement has a Non-Disparagement section: “j. Non-Disparagement: The Contractor agrees not to disparage or make any negative statements about the Company, its employees, or its clients, during the term of this Agreement and for a period of one year following its termination.”

Negative statements can be perfectly true. One can even make them in court, in a lawsuit against Pareto. A company who thinks so much about the possibility of “negative statements” against it most likely knows that its business practices are not correct.

§3. Alignerr

I know nothing about them.

Questions:

  1. How trustworthy are such platforms?
  2. How big is the risk of becoming a disappointed or a wronged gig worker?
  3. How much can one expect to earn, realistically, if not located in a major English-speaking country but targeting English or French?
  4. How feasible is to start without certifications, PhDs, or something “big” as provable qualifications?
  5. Any fiscal issues if located in the EU, say in Germany or in Romania?
  6. Any alternative platforms are welcome.

Now, the answers from various chatbots:

🤖

❶ Grok: https://grok.com/share/bGVnYWN5_97535217-0f1a-44db-8b6e-15acc32df29a

The answer is 100% useless and it doesn’t even address the specified platforms individually! It rather focuses on “online gig platforms, such as Upwork, Fiverr, and similar marketplaces.” 1,000% pathetic!

Grok seems increasingly stupid lately when web searches are necessary.

🤖

❷ ChatGPT Research: https://chatgpt.com/share/6835e3c7-99b4-800c-908a-f710ae4e5b49

“Research completed in 6m · 38 sources · 61 searches.” The analysis seems pretty much OK, albeit “Online reviews and forum posts paint a mixed to poor picture for most of these platforms.” Nonetheless, “Pareto.AI receives the most positive feedback of the four.” And “Alignerr is viewed as the least trustworthy.”

ChatGPT based most of its data on Reddit, just as I did. “Remotasks is easiest to start: one simply registers, completes free training modules and a benchmark test. There’s no formal qualification needed, though poor test performance can bar you from certain projects. Outlier.AI and Pareto.AI have more intensive onboarding.”

Declaring and taxing gig income is a bit complex in Europe, and I don’t believe it’s worth the effort if the revenue is sporadic and questionable. ChatGPT mentioned the basics of it, but that’s boilerplate info.

The suggestions were nice, yet unconvincing:

Given the mixed reputations above, consider more-established crowdsourcing and freelance platforms:

  • Appen/Lionbridge (Remotasks’ original parent companies): …
  • OneForma (TELUS International): …
  • Teemwork.ai, DataForce, Clickworker, CrowdAI: …
  • Amazon Mechanical Turk (MTurk): …
  • Freelance Marketplaces (Upwork, Freelancer, Fiverr, etc.): …
  • Crowdsourcing/Gig Platforms (Clickworker, Spare5/Mighty AI, SearchEvaluation sites): …

A decent answer, but meh.

🤖

❸ Qwen3: https://chat.qwen.ai/s/ab888642-0733-4d0d-b418-4ce956432e37?fev=0.0.104

Supported by 9 links to pages on Brainly, Chegg, Quora, Stack Exchange, and other sites, Qwen3 gave a very modest and useless answer, which at least had the decency to insert “[[file]]” when an assertion had as the unique source the info provided by me.

🤖

❹ Perplexity: https://www.perplexity.ai/search/the-questions-are-in-the-attac-N9Qtfq49RSy6nG449wSvog

Based on 51 sources, the answer, which can be exported as PDF, Markdown, or DOCX, only includes a link to this article published on April 2, 2025: I worked remotely at Outlier AI for 30 days—here’s everything they don’t tell you. Perplexity’s answer was 100% useless, bar for the insight from that article. Some excerpts from it:

Here are the various tasks that Outlier AI provides across multiple fields:

  • Writing original and challenging prompts for an AI model and crafting original responses.
  • Creating complex Math & Reasoning prompts.
  • Teaching LLMs to deliver highly efficient code.
  • Ranking AI model responses based on accuracy and correctness.
  • Editing AI model responses to ensure accuracy, factuality, and relevance.
  • Providing reasoning & logic behind different responses and evaluations.
  • Open rewrite assignments where experts provide a starting point or specific context for an AI model to expand upon.
  • Multi-modal assignments help train models to generate and classify information from multiple modalities, including images, audio, or video.
  • Rating and ranking assignments that ask contributors to identify the best model responses out of two or more options.

He applied for the job of “AI Content Reviewer” (it’s actually a project or a task). “Outlier AI compensates individuals for successfully completing the onboarding process. … I got $10 for mine.” This guy compared Outlier AI to Appen and DataAnnotation Tech. Then:

Before I got to work, I thought my time with Outlier AI would be pretty laid-back and mono-dimensional. I thought, “Okay, I’ll be teaching robots to be less robotic,” but it was nothing like that. Instead, I was thrown into a multifaceted workspace filled with a variety of tasks that challenged my skills and provided valuable insights into the niche of AI training.

What follows is “a detailed account of the tasks I worked on.”

The author of the article, an SEO expert based in Lagos, Nigeria, seemed pretty happy with Outlier AI, but he couldn’t persuade me. Oh, and he never mentioned the hourly rates offered to him!

🤖

❺ Mistral: After having attached the text file: “I’m sorry, but I currently don’t have the capability to access or read files.” I therefore pasted its contents at the prompt. Result: https://chat.mistral.ai/chat/5c394507-1435-4cdd-8bdf-fd89d62c1520

Backed by 16 sources, Mistral’s answer was rigidly structured and of zero practical usefulness! As a side note, here’s the list of suggested alternatives: Appen, Lionbridge, Clickworker, DataAnnotation.tech, Gigwalk, RapidWorkers, iSoftStone, Amazon Mechanical Turk, Upwork, OneForma.

🤖

❻ Copilot: https://copilot.microsoft.com/shares/7GP5Drmo6fhPzGcYJJzvt

Yet another useless answer that doesn’t include much that I didn’t already know! Except, of course, for the suggested alternatives: Appen, Clickworker, Toloka by Yandex, Amazon Mechanical Turk (MTurk), Freelance Marketplaces (Upwork, Freelancer).

🤖

❼ Claude: https://claude.ai/share/1f9b2793-a1d7-4741-a8d1-b041afe4dc19

As it cannot perform web searches, the answer is based on its embedded knowledge, plus mine.

UPDATE: Now it can search the web, but I won’t repeat the task.

So it mostly judges the info provided by me! “The risk of disappointment appears high based on your research.” Its final recommendations:

6. Alternative Platforms

Consider these potentially more reliable options:

Direct freelancing platforms:

  • Upwork, Fiverr (for writing/editing)
  • Rev.com (transcription/captioning)
  • Gengo, One Hour Translation (translation work)

Academic/research focused:

  • Prolific.co (research studies, more transparent)
  • User interviews and testing platforms
  • Language tutoring (iTalki, Preply)

Content creation:

  • Technical writing for documentation
  • Course creation platforms
  • Copywriting agencies

My Recommendations

  1. Avoid platforms with the red flags you’ve identified
  2. Diversify income sources rather than relying on any single gig platform
  3. Consider building direct client relationships instead
  4. Leverage your multilingual skills in more stable markets like translation or language tutoring
  5. Factor in the opportunity cost – time spent on unreliable gig work could be invested in building more sustainable income streams

Your instinct about these platforms being “questionable” appears well-founded. The AI training space is lucrative for the companies but seems to exploit workers through inconsistent work, poor communication, and unfair practices.

Haha, he literally tells me to avoid the platforms brought up by me! As a life coach, Claude’s advice seems nonetheless sound 🙂

🤖

❽ Gemini: https://g.co/gemini/share/11fb1f5a5d48

A completely useless piece of crap! It cites me, FFS! Not that I wasn’t expecting it to be this way.

🤖

❾ Deepseek‘s chats are not shareable, so I’ll paste the answer here:

1. How trustworthy are such platforms?
Trustworthiness varies but is generally low. Platforms like Outlier.AI and Remotasks (Scale AI subsidiaries) have mixed reviews, with widespread complaints about inconsistent work, abrupt account closures, unfair payment practices, and flawed onboarding processes. Remotasks excludes certain EU countries (e.g., Romania), contradicting claims of global availability. Pareto.AI’s non-disparagement clause raises ethical concerns, suggesting efforts to suppress negative feedback. Alignerr lacks transparency. While some workers report satisfaction, systemic issues like automated rejections and location-based pay disparities undermine reliability.

2. How big is the risk of becoming a disappointed or a wronged gig worker?
The risk is significant. Common issues include:

  • Unstable income: Tasks may dry up for weeks, or work may go unpaid due to strict deadlines.
  • Account instability: Automated bans without recourse or explanation.
  • Exploitative practices: Tasks requiring night shifts, downgraded statuses, and non-disparagement clauses.
  • Poor quality control: Flawed assessment tests may prioritize speed over expertise.
    Reddit testimonials highlight frustration among PhDs/experts, who feel undervalued compared to casual workers. Proceed with caution.

3. How much can one earn realistically (outside major English-speaking countries)?

  • English/French: Rates depend on location, not language. For example:
  • Romanian speakers: ~$15/hr (but limited tasks).
  • French speakers: ~$25/hr (higher demand in EU).
  • German speakers: ~$29/hr (better rates but competitive).
    Realistically, earnings may fall short due to task scarcity, unpaid onboarding, and sudden project cancellations. Consistency is rare.

4. How feasible is starting without certifications or PhDs?
Feasible but unrewarding. Platforms often prioritize volume over expertise. While basic roles (e.g., text evaluation) may not require credentials, advanced tasks (e.g., AI training for sociology/math) claim to value qualifications. However, PhDs report being underpaid or ignored in favor of cheaper labor. Beginners can start but should expect low pay and limited upward mobility.

5. Fiscal issues in the EU (Germany/Romania)?

  • Taxes: Income must be reported as self-employment. Germany requires registration (€0–400/month may be tax-free; higher earnings incur income tax + social contributions). Romania taxes freelance income at 10% (micro-enterprise) or progressive rates.
  • Legal compliance: Platforms may not handle tax withholding, shifting responsibility to workers. EU gig workers should consult local regulations to avoid penalties.

6. Alternative platforms:

  • Upwork/Fiverr: Freelance hubs for writing, translation, and data tasks (global access, but high competition).
  • Appen/TELUS International: AI training roles with better reputations (check country eligibility).
  • Prolific: Academic surveys (reliable pay, limited to specific demographics).
  • Clickworker: Microtasks (low pay but consistent availability in the EU).
  • Amazon Mechanical Turk: Low-skilled tasks (notorious for poor pay but widely accessible).

Note: Research platforms for regional availability and worker testimonials to avoid pitfalls.

Meh. Boilerplate stuff.

🤖

❿ Manus: https://manus.im/share/pNVYW5hgLDLis76Z37mHeG?replay=1

Also, the website report: Analysis of Online Gig Jobs for AI Training. Subtitle: “A comprehensive guide to AI training platforms, their trustworthiness, risks, earnings potential, and alternatives.” Comprehensive, my ass!

It cost me 544 credits, and I rated it 1/5 because it’s just pathetic! Most of the ideas were simply retrieved from my document, with very little relevant input from its web search and “analysis.” Its most original contribution comes at the end:

Alternative AI Training Platforms

  1. Appen (formerly Figure Eight)
  2. TELUS International (formerly Lionbridge)
  3. Clickworker
  4. Welocalize
  5. Remotasks (but this one was in my list!)

General Freelancing Platforms with AI Projects

  1. Upwork
  2. Fiverr

Specialized Platforms for Language Experts

  1. Unbabel

Obviously, it just forgot to analyze Remotask, and it added it as a suggestion! WTF.

100% useless.

🤖

Bottom line: 🍆 🍑