As frontier AI models become more capable, ensuring these systems are aligned with human values and fail safely is arguably the most important technical challenge of our generation. But transitioning into AI safety research can be daunting. It usually requires quitting your job, relocating abroad, and navigating a dense, rapidly evolving field on your own.
TARA exists to change that.
TARA is a 14-week, fully funded, part-time technical program designed to build AI safety capacity across the Asia-Pacific region. The program has local cohorts running in major APAC hubs—including Melbourne, Sydney, Brisbane, Auckland, Singapore, Manila, Tokyo, and newly launching in Hong Kong.
The next cohort runs from September to December 2026. Whether you are a software engineer looking to upskill as a participant or an experienced practitioner looking for a flexible teaching role, here is everything you need to know.
The Technical Alignment Research Accelerator is designed to help technically minded people build the skills, portfolio, and network needed to pursue AI safety fellowships and roles.
TARA operates on a hybrid model built for busy professionals:
Saturdays (In-person): Full-day weekend sessions in your local city. These sessions include lectures, pair programming with peers, and guidance from experienced Teaching Assistants (TAs). Start times vary by timezone (e.g., 9:30am in Australia/India, 10:00am in Hong Kong/Singapore/Tokyo/Manila, 10:30am in Auckland).
Weekdays (Remote): Flexible, self-paced independent learning (roughly 2–7 hours a week) supported by Slack access to TAs and your cohort.
Best of all? The program is completely free. TARA covers the cost of expert TAs, Saturday lunches, compute and API credits, and dedicated local study spaces.
TARA’s curriculum is built on the widely respected ARENA (Alignment Research Engineer Accelerator) materials. It is an intense, practical journey through the core pillars of modern AI safety.
Topics generally cover:
Neural Network Foundations: Building CNNs, ResNets, and optimizers from scratch in PyTorch.
Transformers: Coding a transformer architecture from the ground up and loading pre-trained weights.
Mechanistic Interpretability: Using libraries like TransformerLens to reverse-engineer neural networks, extract activations, and understand how models think.
Reinforcement Learning (RL): Building Proximal Policy Optimization (PPO) agents and applying RL from Human Feedback (RLHF).
Model Evaluations: Designing threat models, generating datasets, and running evals using tools like the UK AI Safety Institute’s Inspect library.
The final three weeks are dedicated to an independent, TA-supported project, culminating in a live presentation.
TARA is currently recruiting for two roles across its APAC locations: Participants and Teachers.
If you want to transition into technical AI safety but can’t take time off work, this is for you. You will join an in-person cohort in your nearest participating city.
You are a strong candidate if you:
Have a strong Python programming background (prior experience with PyTorch is a plus)
Have a basic grasp of deep learning, linear algebra, probability, and statistics
Are highly motivated to reduce catastrophic risks from advanced AI
Can commit to attending at least 11 out of 14 Saturday sessions in-person
Can commit around 2 hours a week to independent study
Are highly proficient in English
👉 Apply to be a Participant by July 26
If you already have strong experience with AI safety concepts and want to help mentor the next wave of talent, TARA is hiring remote, part-time teachers for the September cohort. You can be based anywhere globally, as long as you can support the Saturday timezone hours of your assigned APAC cities.
The Role:
Compensation: AUD$80/hour
Commitment: ~13 hours per week (7.5 hours on Saturdays, plus flexible weekday Slack support)
Dates: 12 September – 12 December 2026
Ideal candidates will have:
Completed most or all of the ARENA curriculum
Strong Python and PyTorch skills
A solid grasp of RL, transformers, mechanistic interpretability, and model evals
A patient, encouraging, and proactive teaching style
Applications for teachers close on 8 August 2026, but they are reviewed on a rolling basis, so applying early is highly encouraged.
Want to learn more before applying? TARA hosts a 45-minute Info Session that includes a brief overview of the program followed by an open Q&A. It is perfect for prospective participants, potential TAs, or anyone just curious about the initiative.
Building a local ecosystem of technically skilled alignment researchers is vital. If you know someone in the APAC region who would be a great fit for this—whether as a student or a mentor—please share this post with them.
No posts
Comments
Nothing yet. Say the first thing.
Sign in to join the conversation.