About Working From Home with AI
Greetings, working hitchhickers!
Getting right to the point, since the start of this year, one of my gigs has been training Artificial Intelligence models by being the H in RLHF, or Reinforcement Learning with Human Feedback, via freelance platforms. And as a popular subreddit's title famously states, it's one of the WFHJobs: Work From Home Jobs.
If you've read my blog before, you may know that I suffer from GAD, Generalized Anxiety Disorder. Because of that, working has been a worry of mine for the past few years. I've tried to take more control of my time and efforts, and being a freelancer has really helped me achieve that. It's been almost a year now that I've worked as a freelance 3D Designer (you can check my portfolio here), but this new type of Data Labeling work is still fairly new to me.
In full disclosure, I found most of these opportunities either via LinkedIn or Reddit threads. I felt a bit like an explorer, hitchhiking through this new corner of the Internet that I still haven't passed through. And now, I come here to share about my experiences with you! Some of the links include a referral that may reward me in case you enter some platforms, but I'll make sure to explicitly mention it, and you can feel free to not use them as well. There's no proven correlation between the use of referral links and acceptance, but I can assure you that using my link will make me happy, and that you can use the comments' section to ask for any advice or questions you have regarding the processes!
Another really important thing is that work is not constant, and unfortunately we've been going through a general drought in the market for the last months since around June. My humble take is that the models have improved a lot in the recent future, and Generalist work has not been as needed as before, when the LLMs were still improving; I feel like the Data Labeling demand has shifted to Experts in their areas and Audio-related endeavours. It also seems like the Client demand lowers during the Northern Hemisphere Summer period, and that's the part where I'd love to hear some feedback from you readers on the Northern half of the world: do demands halt much during this time? Because indeed, I've worked around 25% of the time I used to back in March through these opportunities. Still, it has helped a lot in complementing my earnings, so I truly feel like it's a good opportunity, even if you only have a few hours per week available.
GPT5 taking care of Image Creation for this post, since it felt like it fits with the overall theme. Images created on the 16th of September of 2025. I always like to analyze how it cannot maintain consistency through different generations of images, even though I asked for it to maintain the style of the first for the following, and even between the next generated ones they are disturbingly minimally different from a macro perspective, which is interesting. Nano Banana has been really in the spotlight lately, but I haven't tested it yet to have any concrete opinion.
My Experience with Data Labeling
Funnily enough, I started this journey while I was not in my dear hometown of São Paulo. We as a family had saved up and were in Canada, traveling happily and peacefully, but I took some time off my vacation to focus on this year that was starting. It's crazy how fast two thousand and twenty-five passed by.
But before continuing, I think it might be useful to briefly introduce to you what Data Labeling actually is. To help me better define it, I present the definition that IBM presents:
Data labeling involves identifying raw data, such as images, text files or videos and assigning one or more labels to specify its context for machine learning models. These labels help the models interpret the data correctly, enabling them to make accurate predictions.
Data labeling underpins different machine learning and deep learning use cases, including computer vision and natural language processing (NLP).
To go even further, these labels can also be added to identify more complex data, such as 3D objects in MESH or NURBS form! We truly cannot predict what types of raw data we may need to label in a decade from now. And to illustrate how this type of work is handled, I also included a section of the take on the topic by AWS:
Today, most practical machine learning models utilize supervised learning, which applies an algorithm to map one input to one output. For supervised learning to work, you need a labeled set of data that the model can learn from to make correct decisions. Data labeling typically starts by asking humans to make judgments about a given piece of unlabeled data. For example, labelers may be asked to tag all the images in a dataset where “does the photo contain a bird” is true. The tagging can be as rough as a simple yes/no or as granular as identifying the specific pixels in the image associated with the bird. The machine learning model uses human-provided labels to learn the underlying patterns in a process called "model training." The result is a trained model that can be used to make predictions on new data.
To also add my own opinion to it, it's a work that rewards people with Critical Thinking skills and attention-to-detail oriented. And even though a lot of the time is spent working silently annotating data, it's really important to be able to express yourself to the specific Project Managers and Coordinators, since a lot of the time the Labeling Instructions are a few pages long, and getting rid of ambiguity is a daily chore.
The first two job opportunities I heard about were through LinkedIn postings from Turing and Outlier (referral link). The first few qualifications I did for these positions were while still in Montréal, which was really singular, to say the least. And while I didn't work any contracts yet for the first, I had a lot of success with the latter. One additional resource about the topic of Data Labeling comes from Outlier's parent company Scale AI, entitled Data Labeling: The Authorative Guide, which provides useful information, despite the name.
Since then, I've had some experience in multiple platforms, and have known people with experience on them too, so I thought I could provide some suggestions, organized by chronological order that I got in contact with. But before reading, remember one of the most important lessons not only for crypto traveling on this Internet, but also for life:
DO YOUR OWN RESEARCH!
Outlier
I'll always remember Outlier for being the first platform where I successfully had paid work in. Being a Mathematician not only brought some traumas onto my life, but also made me develop an Expertise in Python programming, which was the topic of my first task. Most of the work I've completed in Outlier has been related to Coding, but there are still opportunities. However, the platform has been a lot more voltaile since Meta acquired its parent company, Scale AI, in June. While there are definitely money to be made, I recommend exploring with care!
Referral link:
https://app.outlier.ai/expert/referrals/link/IR6NINZzv06GSNYX2HnyIEByJZU
No referral:
Alignerr
It was in this platform that I started to have opportunities to exercise my skills being a Native Brazilian Portuguese speaker. It's definitely worth it to also mention that it was at Alignerr that I met some of my closest coworkers that I've encountered in this Data Labeling journey. I haven't worked many projects recently there, but it's a legit company that I did some good hours of work.
REFERRAL LINK:
https://app.alignerr.com/signin?referral-code=fade0180-4dc9-4b1b-992c-942a3763221b
No referral link:
Data Annotation
DA is definitely my preferred platform of all that I have worked with. Even though I cannot disclose much due to the NDA, it offers the fairest deal for working, in my humble opinion. Unfortunately, it's not the best at communication, so it's possible that you don't even hear back after completing the Assessment to try to get in, if you fail at it. But in full disclosure, I've experienced some of my favorite work projects from this year within the platform.
LINK (No Referral):
https://www.dataannotation.tech/
Mercor
Mercor is one of the new kids in the block; and even though I haven't worked on any contracts on the platform yet, I've referred a dear friend that's also a Mathematician and he's had a positive experience up to now. I was a bit afraid since he's new to all of this, but he's been really grateful and happy for the opportunity. And that in part inspired me to write about it here, so thank you Du <3
REFERRAL LINK:
https://work.mercor.com/?referralCode=8b47e7ba-ee2b-4801-bfdb-2e27a41daa30
No referral link:
Arthur: 'thank you! would you like to complement your work on this final image by providing a caption?'
GPT5: Of course! Here’s a caption suggestion that ties your article’s theme and the final image together; “Bridging human creativity and artificial intelligence: a new journey of work from home.”
It highlights the symbolic connection in the illustration — the human hand, the robot, and the book — as a metaphor for exploration, learning, and collaboration in your freelancing journey.
Travel Safely, and DON'T PANIC!
I've been rereading The Hitchhickers Guide to the Galaxy once more, and it's been making me feel so happy. I really do recommend you checking it out, especially if you appreciate science fiction and British humour. The Guide has been kind of like my companion over the years of traveling on and offline, and it's been almost ten years since I first read it!
In many of the more relaxed civilizations on the Outer Eastern Rim of the Galaxy, the Hitch-Hiker's Guide has already supplanted the great Encyclopaedia Galactica as the standard repository of all knowledge and wisdom, for though it has many omissions and contains much that is apocryphal, or at least wildly inaccurate, it scores over the older, more pedestrian work in two important respects. First, it is slightly cheaper; and secondly it has the words DON'T PANIC inscribed in large friendly letters on its cover.
Even though it's also a joke, I feel like "Don't Panic" is non-ironically great advice; especially coming from someone that has, indeed, experienced panic episodes during some of the harder patches of my hitchhiking journey through Existence.
...
Breath in.
...
Breath out.
...
Try to give yourself another moment of experiencing nothing but your breathing, instead of "losing control"; because in the end, no one has real control over anything but our own thoughts and actions. And we can choose to try to do our best (: