If you actually want to sell data, here are a few things that will give you an edge as opposed to every other Tom, Dick & Harry who has decided to become a data broker:
1. Source data that is actually useful & that labs/researchers want:
> Data sets that are unique and hard to find elsewhere, e.g a wet lab that has managed to increase the number of tines on trophy-fellow deer over 15 years
> Large volumes of data from private companies, typically with a minimum of 5+ years of operating history with 10+ employees, with exclusive licensing to that
> Work with experts (PhD's, industry experts) to curate datasets where the frontier models perform poorly
> Data you have generated that outperforms what frontier models can do today
2. Find buyers for data at the following places:
> Frontier lab researchers
> Neo labs that sell to researchers at frontier labs
> A handful of AI researchers at enterprise companies
> Other brokers who are already selling to all of the above
3. Learn how to build RL environments:
> Python - functions, collections, types, modules, exceptions, pathlib, subprocess and debugging
> Linux & bash - files, permissions, environments, variables, exit code, stdout & stderr, process clean up
> Git - Branches, diffs, commits, review, rollback & Ci checks
> Python packaging - virtual envs, pyproject.toml, pip, dependency locks & reproducible installs
> Formats - JSON, JSONL, CSV, Markdown, JSON schema - maybe Pydantic
> Data querying - SQL, deduplication, aggregation
> Testing - parameterisation, pytest, unit and integration tests, functional and property testing
> Containers - Dockerfiles, CPU & memory limits,image layers
> Math, task design & experiment design - Sampling, scope, tooling, matrices, vectors, gradients, stability, failure taxonomy and controlled comparisons
I'm sure I've missed much more on the above, but with all of the above, you've got a really good shot at selling data for a living!
The robot AI stack is splitting in two. CoRL 2026 showed it clearly.
(1) A planner decides what to do, and (2) a fast model, the actor, does the hands-on work.
@OpenAI's GPT-6 Astra planned while @nvidia's GR00T 2 handled the manipulation, and the pair succeeded in 97 to 98 of 100 toy assembly trials. @GoogleDeepMind used the same split on Mercedes-Benz lines.
5 key takeaways from CoRL 2026:
1. The planner layer is getting cheaper. Luma AI's open 27bn-parameter model matched Astra at $0.29 per task, against $4.87.
2. Training data is getting cheaper. GR00T 2 trained on about 92% human video and about 5% robot teleoperation.
3. How the motion is recorded matters more than raw hours. In LiberAI's test, 10,000 hours of glove recordings beat 2mn hours of web video.
4. Deployment data compounds. Russ Tedrake called mid-training data "commoditized" and said deployment data is something "you have to earn."
5. Touch is the next frontier. @robbyant_brain open-sourced LingBot-Touch, a tactile model pretrained on nearly 5,000 hours of real tactile data.
Full #CoRL2026 recap on @CoreMatter_: https://t.co/ND2uLaHH4y
We expect robot models to continue to outperform LLMs until early 2027 if trends hold. After that, LLMs will outperform robot models.
Cost and latency might be a different story though. More in the thread.
if you want to save the world, pick a career in robotics.
we are currently fucked.
our energy shortage, ageing population and housing crisis is beyond repair.
this cannot be solved with only art, politics or finance.
humanity's silver bullet is more technology.
so the only way out is through: more datacentres, more R&D, more robots.
2037: robots automate labour and bring in an age of abundance where everyone has food, water and shelter
2027: make it 10x easier for engineers to build and deploy robots
2026: make connecting an agent to a robot a 5 minute task
we need more people who want to save the world.
we do not need more leeches getting rich off a finite supply of iron ore (australians)
Catching up with labs and data vendors lately, the same thing keeps coming up: labs now pay most for data on what their models get wrong.
Volume got cheap. Models generate their own code, sim scenes and tasks now. The deals go to data aimed at a weak spot, or data nobody else has.
1. Meta, Nvidia and AWS are buying sim data from Physicl, a Paris startup that just raised a $10M seed. It's built only around scenes their robot models get wrong, like odd lighting or objects they've never seen.
2. 4 of the top 5 US closed labs pay Halluminate, a 9-person team, for finance RL environments like PE due diligence. On its test the best of 7 frontier models averaged 51%.
3. Google paid $10M for a bankrupt airline's 100M emails and 500M Teams messages, and outbid Mercor to get it. Heard exclusive deals like that run 4-5x non-exclusive ones.
The data vendors I'd watch find failures before the lab does, or own something nobody else can sell.
If you sell data to labs: are you getting paid for volume or for failures?
Fable 5.2 just designed:
▸ a battery-powered handheld pet
▸ a home voice-assistant hub
▸ a USB speakerphone
It engineered the electronics, enclosures, and firmware for each one.
With just @diodeinc + some 3D printers in the loop. Wild.
@Mascobot Is part of the future here that the current de facto architecture of production deployments relying on multiple stages of planning->VLA/similar for semantic execution will go away?
Robotics will soon get bitter lesson'd by Claude, Astra, Gemini Argon, etc.
Just as custom coding models, embedding models, text-to-SQL models, computer-use models, etc., have been bitter-lesson'd over the past ~4 years, the same will happen to custom-trained robotics models, VLAs, world models, etc.
Today we officially release Perceptron Ego 1.0, a public endpoint for egocentric annotations.
With the scale the industry is producing egocentric video, scalable high quality annotations have become a bottleneck. Ego 1.0 is built to achieve frontier performance often matching Astra, while being significantly cheaper.
We built this API to scale our own action models and now we’re ready to share it. Give it a try or reach out if you’re interested!
there remains unbelievable alpha in paying attention to data broker procurement postings as well as frontier lab postings.
literally a public roadmap for where the frontier is at a given moment
crazy review for housecleaning robots:
tl;dr:
1. teleoperation works (mostly)
2. it took 4 humans and 2 robots to clean 80% of one floor for $60
3. we are still a long way away from autonomy