Research
Recurrent Depth in Transformers: Balancing Compute and Memory Efficiency
An analysis of recurrent depth and Sparse MoE as a way to trade memory efficiency for gradient stability in transformer architectures.
Research
Research
Papers that actually matter
42 articles in this section.
Research
Multi-Pass Prompt Verification: Addressing Qualitative Loss in Quantized LLMs
A new study explores using multi-pass verification to recover accuracy lost in 2-bit and 3-bit quantized models, though critics argue it’s a workaround.
Research
OpenAI Model Disproves Discrete Geometry Conjecture via Counterexample Search
An OpenAI model has disproven a geometry conjecture, highlighting the shift from human intuition to high-speed automated counterexample searching in mathematics.
Research
AI Science Assistants and the Reality of Drug Retargeting
Two AI assistants are accelerating drug retargeting by filtering medical literature, though physical lab validation remains the primary bottleneck in drug discovery.
Research
Agentic Discovery and the Future of Long-Tail Fact Retrieval
Exploring how agentic reasoning models move beyond static RAG to automate the discovery of niche, long-tail facts with high precision.
Research
The Citation Death Spiral: How AI Slop is Breaking Academic Research
AI-generated research papers and fraudulent paper mills are creating a synthetic feedback loop that threatens the validity of academic prestige metrics.
Research
Bridging the Intent Gap: Why LLMs Struggle with Pragmatic Meaning
The IntentGrasp benchmark reveals a critical void in LLM architecture: the inability to distinguish between literal instructions and actual human intent.
Research
Anthropic’s Constitutional AI: Moving Beyond Human-Labeled Data
An analysis of Anthropic’s ‘Teaching Claude Why’ research and the trade-offs between constitutional reasoning and raw model performance.
Research
The Security Risks of Partial Evidence in Enterprise RAG Systems
Exploring how LLM agents can infer restricted data from partial evidence, creating a new vector for privilege escalation in enterprise RAG pipelines.
NCN
Research
Google DeepMind’s AlphaEvolve: Automating Algorithmic Optimization
Google DeepMind’s AlphaEvolve uses Gemini to autonomously evolve code, raising questions about the trade-off between performance and maintainability.