Micro1's CEO Ali Ansari says the company wants to recruit 10,000 robotics trainers in seven days.

The advertised work: review and label videos of robots performing tasks. His September 5 post lists $50 to $90 an hour, says applications are accepted globally and says previous AI experience is not required. Original announcement

These are advertised rates and a recruitment target, not confirmed placements or guaranteed hours. The post alone does not establish what an applicant in India will qualify for.

Still, the work itself is interesting. A company is recruiting thousands of people to help machines understand everyday actions.

Why does a robot need so much help doing something we barely think about?

I am Alex. Welcome to ShortCu8 by Innov8.

Lets Dive Deep🐰🐰.

When a robot performs a task independently, the human work may have happened much earlier.

People can demonstrate the action, describe the important moments and evaluate attempts. Training then uses those examples to improve the model controlling the robot.

Demonstrate -> capture -> label -> train -> test -> collect better examples.

This is a simplified loop. Different robotics teams combine human demonstrations, simulation and the robot's own attempts in different ways. DeepMind's RoboCat example

Imagine asking a robot to pick up a cup.

It must reach the right place, grip the object and lift it without dropping it. Change the cup's position or put something in the way, and the required movement changes.

A video can show what happened, but turning that footage into useful training material takes additional work.

Micro1 describes services that record demonstrations, mark the stages of an action, track interactions between hands and objects, and check data quality. Its robotics work includes household tasks, repair, assembly and logistics. Micro1's robotics programme

For example, a reviewer might distinguish reaching for a cup from grasping it, then identify whether the lift succeeded. That is an illustration of the annotation work, not a published task from this recruitment drive.

The useful information includes where an action failed. A robot briefly touching the right object is different from completing the job.

Watching someone fold a shirt can teach a system something about the task. It does not automatically supply the exact commands needed for a particular robot's joints and grippers.

Robotics researchers therefore combine different kinds of evidence. DeepMind's RoboCat, for example, learned from sequences of images and robot actions, human demonstrations, simulated tasks and additional attempts generated by robots themselves. How RoboCat learns

Human labelling is one part of this process. Simulation and automated data collection also help teams expand training without asking a person to supervise every attempt.

The challenge is getting enough varied, useful examples for a machine to cope when the next room, object or task differs from its training.

In a separate announcement on September 3, Figure signed a partnership with Nscale for up to 100,000 Nvidia Vera Rubin GPUs. The initial compute commitment is $3.5 billion, with deployment targeted to begin in the second half of 2027. Figure's announcement

Figure says its Index programme is generating 35 minutes of data every second, and that both data and compute constrain progress on its Helix models.

Micro1 and Figure have not announced these as a joint project. Their announcements show different parts of the same broader effort: producing training material and securing the computing capacity to learn from it.

Thousands of reviewers do not replace the need for training compute. More GPUs do not automatically produce better demonstrations or correct labels.

Micro1's announcement is evidence of demand for human input into robotics. It cannot tell us whether these roles will provide stable careers or how long the advertised demand will last.

But it makes part of the industry visible.

Someone has to capture examples, assess their quality and help establish what successful behaviour looks like. Ordinary physical knowledge becomes valuable when a machine needs to learn it.

That does not mean every robot is secretly being controlled by a person. Demonstrating a task during training, labelling a recording and remotely operating a deployed robot are different jobs.

The person teaching the machine may never be present when it finally performs the task on its own.

We read every reply - just reply to this email and let us know how we can improve !

#AD1

AI made PMs faster. Multiplayer mode is still broken.

A PM can summarize research, draft a PRD, and mock up a prototype before lunch. The hard part starts when the team has to decide what actually gets built.

Jira Product Discovery gives product teams one place to capture insights, prioritize ideas with consistent frameworks, and build living roadmaps stakeholders can rally around.

And because it’s connected to Jira, the context behind every decision stays with the work—so developers and their agents know not just what to build, but why.

AI helps PMs move faster. Jira Product Discovery helps the whole team build with confidence.

Get Started for Free.

#AD2

Most SaaS finance teams know their billing process is slow.Most SaaS finance teams know their billing process is slow. Few know what it's costing them.

The Tabs Billing Lag Calculator puts a dollar figure on it in two minutes — benchmarked against top SaaS companies.

See your number.