Applied AI & Industry Deployment
Key Questions
What does the PwC report indicate about enterprise AI investment?
The report shows that investing at least 1.6% of revenue in AI yields significant EBITDA and TSR gains, marking an investment tipping point.
What is Gemini Embedding 2?
Gemini Embedding 2 is Google DeepMind's native multimodal embedding model that strengthens retrieval and RAG infrastructure.
What does the Stanford AI hiring study find?
The study of 4M applications reveals role-level racial disparities in recruitment driven by AI hiring tools.
What memory law does How LoRA Remembers? identify?
It reveals a parametric memory power law linking loss reduction to parameters and sequence length, with a phase transition at p>0.5 for verbatim recall.
What benchmark evaluates AI on chip co-design?
IBM and Columbia's HSCO-Bench tests AI agents on full chip co-design for heterogeneous SoCs.
PwC enterprise AI benchmarking report shows investment tipping point at 1.6% of revenue yields significant EBITDA/TSR gains. New: SMART embedding trick extracts multi-vector performance from single-vector models via late interaction on frozen hidden states. New: Google DeepMind releases Gemini Embedding 2 — native multimodal embedding model, major signal for retrieval/RAG infrastructure. New: First open benchmark for early Parkinson's speech detection. New: AutoResearch AI survey maps shift from task-level to workflow-level AI for science. New: IBM/Columbia HSCO-Bench benchmarks AI agents on full chip co-design for heterogeneous SoCs. New: AI hiring bias study (Stanford) — 4M applications, role-level disparities. New: How LoRA Remembers? reveals a parametric memory law for LLM fine-tuning, showing a power law linking loss reduction to parameters and sequence length, with a phase transition at p>0.5 for verbatim recall; MemFT's threshold-guided budget redistribution is practical.