Ali Asaria
Hello, the kids are asleep so I thought I’d type out these thoughts
- research
- Asaria, Salomone, and Gandhi, “More Canadian than American: the surprising default values of five leading AI models,” Transformer Lab, 2026.
- Asaria, Salomone, and Gandhi, “The Copy Test: How Much of a Foundation Model's Skill Is Already in the Obvious Answer?,” Transformer Lab, 2026.
- Asaria, Salomone, and Gandhi, “Does a reasoning model have a subconscious? A plain lens says probably not,” Transformer Lab, 2026.
- Asaria, Salomone, and Gandhi, “You can't scale a learned physics simulator out of blowing up,” Transformer Lab, 2026.
- Asaria, Salomone, and Gandhi, “How much can you cut off a 3D shape before it can't grow back?,” Transformer Lab, 2026.
- Asaria, Salomone, and Gandhi, “It's the language, not the lab: the prompt language, not the model's origin, is what shifts its framing on contested topics,” Transformer Lab, 2026.
- Salomone, Gandhi, and Asaria, “The best-sounding open-source music generator is also the only one you can steer,” Transformer Lab, 2026.
- Asaria, Salomone, and Gandhi, “Everyone assumes bigger AI models win. We built a 34,000-parameter earthquake detector that beats a model 8× its size,” Transformer Lab, 2026.
- Asaria, Salomone, and Gandhi, “Reasoning models don't have much trust in their own thoughts,” Transformer Lab, 2026.
- Asaria, Salomone, and Gandhi, “Scaling a protein model is what lets RL beat trial-and-error protein design,” Transformer Lab, 2026.
- Salomone, Asaria, and Gandhi, “What's the best way to run a 30B coding model on a 24 GB Mac?,” Transformer Lab, 2026.
- Asaria, Salomone, and Gandhi, “Does fine-tuning a chatbot make it more brain-like? We checked,” Transformer Lab, 2026.
- Asaria, Salomone, and Gandhi, “When an AI searches for the equation behind your data, one dial steers it,” Transformer Lab, 2026.
- Asaria, Salomone, and Gandhi, “Training an AI to be correct collapses its variety,” Transformer Lab, 2026.
- Salomone, Gandhi, and Asaria, “How modular is a frontier Mixture-of-Experts?,” Transformer Lab, 2026.
- Asaria, Salomone, and Gandhi, “Teaching a language model to say "I'm not sure" using its own doubt,” Transformer Lab, 2026.
- Asaria, “How I built my own Tesla-style self-driving AI,” Transformer Lab, 2026.
- Asaria, Salomone, and Gandhi, “Judging to improve: a 3D judge you can train against, and the limit of cheap 3D specialization,” Transformer Lab, 2026.
- Asaria, Salomone, and Gandhi, “Train, retrieve, or both? What it takes to make a language model cite the law correctly,” Transformer Lab, 2026.
- Asaria, Salomone, and Gandhi, “Judging single-image 3D generation without humans (and why cheap proxies fall short),” Transformer Lab, 2026.
- Asaria, Salomone, and Gandhi, “A Universal Post-Training Improvement for Open Audio Models,” Transformer Lab, 2026.
- Asaria, Salomone, and Gandhi, “Neither parallel nor sequential: how DiffusionGemma actually commits tokens,” Transformer Lab, 2026.
- Asaria, Salomone, and Gandhi, “Making INT8 actually fast: a fused kernel for Ideogram 4 on a 3090,” Transformer Lab, 2026.
- Gandhi, Asaria, and Salomone, “Quantizing Ideogram 4.0 onto a 3090: an INT8 build that matches FP8 and a 4-bit GGUF that beats NF4,” Transformer Lab, 2026.
- selected writings
- 2026-04-21 Between Clay and Light: A Quranic Framework for the Age of Intelligence
- 2025-12-18 Ambitious Canada Part 9: Permission to Build... What?
- 2025-11-24 Ambitious Canada Part 8: The Mega Project Playbook
- 2025-11-06 Ambitious Canada Part 7: Who is a Builder?
- 2025-10-27 Ambitious Canada Part 6: A Million Internships
- 2025-10-24 Ambitious Canada Part 5: I am Old
- 2025-10-19 Ambitious Canada Part 4: What is Special About Canada?
- 2025-10-17 Ambitious Canada Part 3: You Can't Copy the US
- 2025-10-02 Ambitious Canada Part 2: The Current Ideas are Boring
- 2025-10-01 Ambitious Canada Part 1: This is our Moment
- 2025-09-09 Opensource is Crazy
- 2025-09-09 The secret to arguing with CEOs
- 2025-03-27 Tech Leaders: Let's be Honest That We Have Sucked