MIT-IBM Team Finds Multimodal AI Models Can Ace Benchmarks on Pure Noise
MIT News AI · rss · 2026-11-21
An MIT News Q&A with Jim Glass's MIT-IBM team covers key failures of multimodal foundation models: models like GPT-4o can't localize objects from few-shot examples (fixed via a pseudonym-substitution trick), and audio-LLMs often game multiple-choice benchmarks without listening — they still score well on pure noise. The team's approach emphasizes weakly supervised, data-efficient adaptation (including test-time RL) and public benchmarks, and is now moving to full-duplex spoken dialogue systems that listen while speaking.
More from Multimodal
- Selfie-to-Hologram Pipeline Perfected Using Gemini Neon and Scenario — AIandDesign · 2026-11-21
- Open-Source Editor Velorn 0.4 Adds Free Idle-GPU Sharing, Phone Control and a Built-In Agent — VisualFXMan · 2026-11-21
- AI-generated illuminated manuscript revives medieval book art, 'making the past alive again' — repligate · 2026-11-21
- Building a SpaceX Raptor 3 engine in 3D with Gemini Neon, viewed on Apple Vision Pro — dreamwieber · 2026-11-21
- Human-made or AI-generated? Reddit challenges users to tell the difference — RainbowSovietPagan · 2026-11-21
- Muse Code adds 3D model generation with printable textured output in CLI — charlieholtz · 2026-11-21