InAltsoph’s blogbyAleksey Tikhonov·Sep 25WTF is Voice Steering?We trained our new TTS model to follow free-form voice instructions, then needed a way to measure how well it followed them. I decided to…
InAltsoph’s blogbyAleksey Tikhonov·Sep 5Auto-cartoons for 2026 Ig Nobel Prize winnersTo test the experimental e2e scientifical cartoon generation pipeline we developed with Alexey RBTZ Ivanov, I used it to generate…
InAltsoph’s blogbyAleksey Tikhonov·Aug 6Artificial tongue twisters for TTS evaluationHow I hacked artificial tongue twisters evolution for fun and profit^W^W^W^W to test our TTS models.
InAltsoph’s blogbyAleksey Tikhonov·May 15What’s Wrong with TTS EvaluationBesides other things, I am the Head of Evaluations at Inworld AI, where we also build TTS models. Our previous TTS model is still top 1 on…
InAltsoph’s blogbyAleksey Tikhonov·Jul 25, 2025Quantum Persona and Test-time Mode CollapseThis time I want to explore a phenomenon I’ve been investigating: what happens when a model trained on contradictory information needs to…
InAltsoph’s blogbyAleksey Tikhonov·Jul 18, 2025What’s Wrong with Chat-Templates Format for LLMI’d like to discuss the current situation with LLM prompting standards; how we got into this mess; and how to live with it now.
InAltsoph’s blogbyAleksey Tikhonov·Jun 30, 2025BIG BANG OF AGENT RULESTLDR: Last weekend, I spent several hours digging through publicly available cursor rules files and analyzing existing usage patterns…
InAltsoph’s blogbyAleksey Tikhonov·Apr 1, 2025COMIC CARTOON GENERATIONSince April Fool’s Day is today, let me share some results on automated comic cartoon generation.
InAltsoph’s blogbyAleksey Tikhonov·Dec 8, 2024HUMOR ARENATLDR: We developed a novel approach to humor generation that gives human-level results on blind tests. To facilitate further progress in…
InAltsoph’s blogbyAleksey Tikhonov·Nov 30, 2024GALLERY OF UNSEENTLDR: It started as a bunch of strange experiments and ideas on unsupervised generation and converged into another NaNoGenMo project. This…