A 2-hour AI podcast tackles multilingual models, alignment, and evaluation
simi_97k · x · 2026-07-28
A 2-hour AI podcast on multilingual systems, cultural alignment, and evaluation
The episode centers on a long conversation about open problems in AI research. Topics include:
- multilingual and multimodal AI
- evaluation for creative and subjective tasks
- alignment and how it should be encoded in models
- personalization for users with different backgrounds and experiences
- how to choose research problems based on personal goals
The guest also describes a PhD project on machine translation for vision inspired by a real-world need: translated storybooks for children need the images contextualized too, not just the words.
More from Research
- Continual learning on Qwen3.5-397B is said to match Opus 4.8 for about $450k — josh_wills · 2026-07-28
- Climate Change AI sets NeurIPS 2026 workshop dates for Sydney on climate ML — kklmmr · 2026-07-28
- Harvard and MIT find end-to-end RL can hide role drift in compound LLM systems — omarsar0 · 2026-07-28
- Paper says AI control cut wastewater plants’ electricity use 5.4% and emissions 6% — paulnovosad · 2026-07-28
- OpenAI says small teams use AI as a cross-functional generalist tool — OpenAI · 2026-07-28
- A sparse-matrix idea tries to mix tensor-core efficiency with larger hidden capacity — mgostIH · 2026-07-28