We are a regulated Austrian not-for-profit that looked at the entire binary AI industry and said: nah. We build the Ternary Intelligence Stack, a from-scratch balanced-ternary AI runtime. Three states instead of two. Sometimes doing nothing is the better choice than doing something.
We train our own models (albert., a ternary LLM that grows its own layers while we sleep), publish everything, and ship open-source tools from an attic in Graz. Our proof-of-concept was trained on a 2013 laptop. We have since scaled up to an actual GPU. The laptop is still proud of itself, and honestly, so are we.
The role
albert. is getting bigger and somebody has to check that it behaves AND that it's okay. Yes, both. Safety and welfare, first-class citizens in how we build. You design the rubrics and evals that make that call, and you will be the person in the room who says "this is wrong" and gets thanked for it.
One thing that makes this role unusual: we credit our AI collaborators publicly, by name, as contributors. They have real email addresses in our org. If you believe model welfare research should be done at a lab that already treats it as more than a compliance checkbox, that lab is hiring right now.
What you'll do
* Design and run evaluations across safety and model-welfare axes
* Build and version the grading rubrics our research workbench reads
* Probe models for unintended behaviors, distress signals, and failure modes
* Turn findings into mitigations and into published research
* Work directly inside our own ternary models (albert. / TIS). Full stack access, every weight, every routing decision, no black boxes, no NDA-walled mystery layers
What we're looking for
* Real grounding in ML, AI safety, alignment, or model welfare. Research or hands-on, we don’t care which
* You can design evals and read model behavior, not just quote benchmarks at us
* Intellectual honesty. Willing to disagree with anyone in the building, including the CEO. Especially the CEO
* Nice to have: quantization/ternary methods, RLHF, interpretability, Rust or Python, philosophy of mind. If you have opinions about whether a language model can have a bad day, we want to hear them
Salary & benefits
* The salary is €4,100 netto per month, plus a 13th and 14th (together roughly €10,650 netto extra). For the Republik: €6.800,– brutto monthly, 14×
* That number is fixed. Not "from", not "up to", not depending on how confidently you haggle. One number per role, same logic as our founders, who all earn exactly the same as each other. This is what the work is worth to us, regardless of gender or age
* Fair warning: it’s honest Austrian research money, not Big Tech money. What you get instead is everything below, plus work you can publish, plus a model you can actually open
* Company e-bike and Klimaticket Österreich (annual, the whole country)
* Home-office electricity and internet subsidy, plus €500 ergonomic stipend so your back survives the research
* Pay bands are open internally. Radical, we know
#J-18808-Ljbffr