LivingScreen Exposes How 13 Top AI Models Fail at Scrolling TikTok
LivingScreen exposes a blind spot in every frontier AI agent: none can decide how long to watch a video, and all fall far short of human performance on TikTok-style feeds.
20 stories
LivingScreen exposes a blind spot in every frontier AI agent: none can decide how long to watch a video, and all fall far short of human performance on TikTok-style feeds.
LifeSide exposes a critical gap: frontier LLMs that ace memory benchmarks still fail at sustained emotional understanding across sessions
CYGNET puts a pre-execution gate between LLM agents and Neo4j, catching 100% of structural Cypher errors before they hit production — at just 5.6 ms overhead.
A new benchmark reveals that video-understanding models give better summaries to some videos just because of where they appear in the input queue
TAP-PER replaces per-user LoRA adapters and RAG prompts with compact learned prefix embeddings, cutting per-user storage 130x while beating every baseline on the LaMP benchmark
A new retrieval framework tackles the messy reality of long group chats by finding coherent multi-message, multi-image fragments instead of isolated utterances
AutoTool teaches multimodal models to decide when NOT to use tools, cutting inference time by up to 44.9% while boosting accuracy
A new fine-tuning method cuts reasoning model token usage by up to 66% using just 1,000 training samples, with no RL required
RAMPART treats LLM context assembly like a compiler pass, using five primitives to control what goes in, where, and when — with measurable gains across five models.
A head-to-head trial pitting three frontier LLMs against real headache specialists reveals where AI falls short — and where it eerily blurs the line
GOTabPFN uses graph-based feature ordering and neural compression to squeeze high-dimensional biomedical data into TabPFN's token budget without retraining the model.
Researchers from the University of Bucharest propose a cycle-consistent machine translation pipeline that bootstraps coreference resolution for languages with little to no training data.
Researchers propose using an auto-discovered schema as a shared contract for both knowledge graph ingestion and multi-source QA retrieval, beating RAG and agentic baselines across four benchmarks.
A new study finds LLMs used as mutation operators in genetic programming reliably converge to the same structural templates, undermining the diversity that evolutionary search depends on.
A new Harvard-led study maps 403 U.S. hyperscale data centers and finds their carbon intensity runs 48% above the national grid average, driven by where they are built.
A new paper surgically removes safety refusals from code LLMs to generate labeled vulnerable code for security research, revealing that willingness and capability are entirely separate properties.
A new residual coding framework beats SZ and fixes the high-fidelity blind spot of learned scientific data compressors by up to 60%
TimeClaw gives LLM agents a time-series-native runtime, evolving toolbox, and episodic memory — beating models up to 235B parameters while using fewer tokens
A new open-source benchmark tests whether AI agents can wait patiently for the right moment, rather than burning compute hammering refresh buttons
Researchers analyzed 837 GitHub fault threads across 473 MCP servers to map the 73 distinct ways your AI tool integrations can silently break at runtime