About this role
From Silver.dev
Estamos ayudando a los de G2i a conseguir trainers y task creators para laboratorios de AI, que incluyen a OpenAI y SpaceX.
Las tareas son unidades de trabajo para entrenar a los modelos y se paga por tarea completada efectivamente. Las tareas toman entre 1 y 2 días en completarse, y los creadores de tareas más efectivos llegan a completar más de 30 tareas mensualmente.
El trabajo es async y lo podés manejar a tu tiempo, pero sí requiere varias horas de trabajo y onboarding. Presten atención a las instrucciones de G2i.
About the role
G2i is hiring senior and staff-level software engineers to build the coding tasks a frontier AI lab uses to train its models. Each completed task pays $1,000 USD.
Most AI training work is annotation: engineers pick up a task someone else wrote and evaluate it. This role is one step earlier. You design and build the tasks themselves, which takes a different skill set from typical product engineering.
What you'll do
-
Design and build realistic, challenging software engineering tasks that frontier models will be trained and evaluated on
-
Write tasks across codebases of very different sizes, from a few files to large multi-service projects
-
Work mainly in TypeScript, Python and Go
-
Revise your tasks based on feedback from G2i reviewers and team leads until they meet the lab's quality bar
-
Work mostly on your own, with support from reviewers and a shared engineer channel
-
Who we're looking for
-
We're hiring two kinds of engineers. Both are welcome to apply.
-
Engineers who have built AI training tasks before (top priority). You've designed and built tasks or environments for training models at a frontier lab or a data lab. Contractor experience counts. What matters is that you built tasks, not only annotated, reviewed or evaluated tasks someone else wrote.
-
Very strong senior, staff or principal engineers who haven't done this yet. You have deep, hands-on skill in TypeScript, Python and/or Go, and you learn new kinds of work fast. You may need a few extra days in training to learn the format, and that's expected.
For both profiles:
-
Senior to principal-level engineering experience
-
Strong code reading and writing skills, without relying on AI tools
-
Clear written English, since tasks and reviewer feedback are written
-
Pay
-
You're paid per completed task, not by the hour.
-
Training project: if you pass, you're paid $600 USD and join the core team.
-
Each completed task after training: $1,000 USD
-
Based on early internal data, top performers finish 1 to 2 tasks a day, mid-level performers about 1 every other day, and others 1 to 2 a week. A task usually takes 4 to 10 hours.
-
If you don't pass the training, it doesn't count against you for other G2i AI projects. Those may open to you separately.
-
Hiring process
There are no live technical interviews. The steps are:
-
Apply through the link below.
-
Take-home assessment. It's designed to take about 90 minutes. Taking 4 to 6 hours is fine and won't lower your score, because quality matters more than speed. It doesn't expire, but the sooner you finish it, the sooner you can start.
-
Background check. This usually takes 24 to 48 hours in the US and Canada. In some countries it can take more.
-
Offer. It's usually sent within a day after your background check clears.
-
Training project. You build one task with guidance from reviewers and team leads. You have up to 7 days. Engineers who have built tasks before often finish in 1 to 2. If you pass, you're paid $600 and join the core team. If you don't, you're out of this project, and it's unpaid.
-
Hours and contract
-
Contract: independent contractor, fully remote. Either side can end it at any time.
-
Hours: 40 a week is ideal. Part-time is welcome: 20 a week is preferred, and 10 is the minimum.
-
Schedule: flexible. You can work any day of the week, and some overlap with North American time zones is preferred.
-
Keep your current job: you don't need to leave a full-time role to join part-time.
-
Length: this is a project-based engagement, not a permanent role. When it ends, G2i often has other frontier AI projects available.