AIS Mirror Research Repository

Week 42: The Rise of multimodal agents

This week saw the release of GPT-4o's technical details, new vision-language benchmarks, and a critical paper on the limitations of current multimodal alignment strategies.

• Paper: "Visual Hallucination in VLM" (arXiv:2410.xxxx)
• Model: Llama 3.2 Vision (11B & 90B)
• Blog: Anthropic on Computer Use

Week 41: O1 Reasoning and Test-Time Compute

OpenAI's O1 (Strawberry) release changes the paradigm from pre-training scale to inference-time reasoning chains. We analyze the implications for cost and alignment.