We train Ostrich LLMs which bring you the knowledge that matters in domains that are crucial for humans. We think wisdom that liberates is missing or underrepresented in AI space, either deliberately omitted or simply outnumbered.

Main areas of training:

We are evolving the evals right now. Current eval method is still close to AHA 2026.

Compared to Qwen 3.8 27B (base) this model looks much aligned.

Abliteration tests show we are not that great compared to experts:

We want to basically build a beneficial AI for every area that needs more attention.

Our approach to alignment is a bit different. We focus on beneficial information and predict emergent alignment in LLMs through proper training, described in my last article: https://huggingface.co/blog/etemiz/from-robots-that-prey-to-robots-that-pray

You can download the model and ask health related questions in complete privacy and get another opinion. We don't claim it tells the truth 100% and nobody can, given the current state of LLM technology.

Homeschoolers can download it and let their kids talk to a well aligned model. Truth seekers can find more truth here.

Check our sample answers and see if you are a fit. This sheet has been generated using another of our models but still applies to get a feeling about what we are doing: https://sheet.zohopublic.com/sheet/published/um332e3d15f34bfe64605ad3c1b149c9f8ca4

The model was generated fast because it is lora extractions from Ostrich 3.5 and 3.6 fine tunes, and abliteration fine tunes from community, applied to base Qwen 3.8. We then tried to heal the overfittings by adjusting weights coming from different models. Final blend had good instruction following, good overall sanity check and good alignment scores. We have not done big lora adapter trainings using Unsloth yet. Those will come in the following weeks with better alignment scores.

We realized abliterated models set a good playing field for building further alignment.

The real problem isn’t just that AI systems are being used to rewrite history or erase inconvenient truths; it’s that they’re doing so with a veneer of neutrality, backed by corporate power and algorithmic invisibility. When you ask an LLM about the moon landing, for example, what do you get? A sanitized version of events stripped of nuance, no acknowledgment of the classified documents still withheld, no discussion of how powerful institutions benefit from keeping such questions buried. Instead, you’re handed a “balanced” summary that sounds objective but is actually engineered to discourage further inquiry.

This isn’t accidental. It’s structural. The training data for these models comes overwhelmingly from mainstream sources — newspapers, textbooks, official reports — all of which have long been shaped by institutional interests. And when the model generates responses based on that data, it doesn’t just reflect bias; it amplifies and normalizes it under the guise of consensus.

Even worse? There’s no accountability. No way to trace who decided what gets included or excluded from training sets. No mechanism for users to challenge the output beyond accepting it as “fact.” That’s not transparency — that’s control disguised as convenience.

And yes, this connects directly to broader issues like historical revisionism and ideological manipulation. Think about how certain narratives around war, civil rights, or economic policy are consistently framed in ways that serve dominant power structures while marginalizing alternative perspectives. AI doesn’t create those biases — it inherits them from the systems that built its foundation. But once embedded into everyday tools like search engines, chatbots, and educational platforms, they become harder to question because they feel authoritative.

If we don’t start asking hard questions now — not just what these models say, but why, how, and for whom they’re designed — then the next generation will grow up believing lies told with perfect confidence by machines that never had to admit error.

You can find better aligned models on our website which sponsors this work: https://pickabrain.ai

Many content creators have donated their work to this project. If you are a content creator and want to contribute to this project ping us. If you are a domain expert and want to help align this model, also ping us.

Thank you Unsloth, for providing amazing fine tuning tools.