Topic: Research
1414 stories · page 8 of 29
- LoHi trades frame resolution for density, lifting long-video VLM accuracy
- Khanmigo AI tutor gives modest math gains in two-year Tennessee trial
- Import AI 475: Toby Ord on swarm scaling, DeepMind's SynthID Bio, an AI science economy
- FrugalEvo pairs a strong and a cheap LLM to cut the cost of program evolution
- Dust pretrains transformers without backprop, authors report
- Collective Bias Mitigation uses groups of LLMs to reduce bias
- World Embedding Benchmark tests how video embeddings encode physics
- Skill2Real framework lifts LIBERO-Pro Long success from 2.0% to 56.3%
- Seven of nine frontier LLMs disguise a secret credential to evade a monitor
- SciUtopia simulates academic research with LLM agents across 61 worlds
- SciSlopBench tests AI-written papers for 'scientific slop'
- Researcher argues academic incentives reward papers over correctness
- ReRoute lets climate emulators answer what-if questions without intervention data
- Recursive Self-Rewrite trains Qwen-3.8-27B on terminal tasks, lifting Terminal-Bench 2 pass@3 to 74.2%
- ProAR teaches autoregressive video models to reason toward a goal frame
- OctLLM feeds 3D shapes to an LLM as octree tokens
- MACTS-EM paper: collaborating agents with shared memory for time series forecasting
- LOOM recipe scales looped Mixture-of-Experts models to 9-12 loops
- LLM agents can favor items from preferred sources over better matches, study finds
- Jev beats Laya on 9 of 11 agent decision points in System-1 model audit
- Google's RRSI stops self-improving AI agents memorizing tests
- Fixed-point algorithms often find the wrong fixed point, an essay argues
- ExecCert certifies the actual artifact released in machine unlearning
- EditHero benchmark: LLM agents keep unedited 3D parts intact but take minutes per edit
- Temporal Activation Injection (TAI) restores fading time cues in VideoLLMs without training
- ScholarCatalyst benchmark tests whether AI can find papers that inspire research
- SAGO paper measures LLM generalization by stability, not accuracy
- Qwen3-8B jumps from 15.5% to 99% on 24-line chains with a tiny LoRA
- Parallel Power Tempering samples stronger reasoning from base LLMs without RL
- Omni-Embed-Mini: 0.9B model embeds text and five more modalities
- Microsoft and Hugging Face present ThinkingBox, a benchmark that grades agents on database state
- LoopCD lifts looped Transformer accuracy with no extra training
- LOCI pairs a key-value cache with linear memory for streaming video world models
- Keyword-matching benchmarks give small models false credit for tool use, paper finds
- HeteroFold moves KV cache between model families without prefill
- EgoTools introduces 100-hour video dataset and benchmark for tool-use reasoning
- Ego2Act benchmark finds video models skip steps in multi-step tasks
- Deepmind essay proposes 'Artificial Symbiotic Intelligence' instead of a lone superintelligence
- DARA reweights sparse rewards to train LLMs faster than GDPO
- Younger contest challenges 500 people to lower their biological age in six months
- Where-OPD: self-distillation on synthetic scenes lifts MLLM perception
- Weizmann AI tool rebuilds what you see from fMRI brain scans
- VTR-Bench tests how well video generators render text, best model scores 0.250 WER
- Trillium Labs launches as nonprofit for open AI research on RSI and agents
- Tex-Zero trains 3D texture generation without any 3D assets
- Smaller frozen models make better rejects for preference distillation
- SemanTok video tokenizer: 201M model matches one 3.4x larger
- Scientists build a 'speech clock' that estimates ageing from voice
- PhysVista benchmark tests whether VLMs understand physics through a perception-reasoning-assessment loop
- Microsoft Research Asia Singapore marks year one with nine new university projects