Posts

Paper of the Day

Weekly paper notes and takeaways. Informal notes written quickly based on what I was exploring at the time of writing. Some personal notes may assume context from ongoing research projects.

2026 36 posts 2025 52 posts 2024 93 posts 2023 16 posts

Log

Experiment logs, notes, and fragments.

Jul 30 Tool call과 Memory

Jul 29 Kanana-2 SLM 개발기 톺아보기

Jul 06 ICML2026 관심 워크샵

Jul 06 ICML2026 관심 논문

Jul 02 초개인화를 어떻게 연구해야할까?

kanana-o 45 posts interject 6 posts REMODE 6 posts

Misc

Uncategorized notes and references.

Apr 29 site publish state — 5 case + listing 축 정리private

Apr 29 감정 — Qwen flash redoprivate

Apr 28 Research Proposal — Modality Gapprivate

Apr 28 감정 — log-17 분리private

Apr 28 감정 — 진행 상태private

Tags

Posts grouped by topic — across logs, paper notes, and misc.

kanana language-modeling rag benchmark reasoning multimodal memory dialogue-system emotion agent self-improvement evaluation

+ more

kmmlu reinforcement-learning prompting long-context modality-gap note respond personalization ops interpretability icl hallucination petl optimization llm-as-a-judge interject function-calling domain-adaptation alignment-learning tool-calling representation-learning peft odqa multi-modality long-horizon knowledge-conflicts factuality ensemble code transformers safety proactive observer-agent multi-linguality multi-agent industry classify agent-benchmark user-simulation rl retrieval post-training multi-turn minicpm laaj knowledge-editing infra hcx exp_d dpo design ai-detection agent-memory activation KoED weight-merging user-model tool-use test-time-scaling terminal-agent setup sae qwen prompt-compression policy-compliance planning on-device memory-systems lrm knowledge icml2026 empathy distillation diffusion attention adaptor MMLU LaaJ workshop workflow user-representation user-preference unlearning tts trust-calibration translate transfer-learning training-free tool-discovery token-level tmux time-series time-sensitive test-time-memorization telegram tau-bench task-oriented-dialogue talk tableqa synthetic-data synthetic-benchmark sycophancy status stateful-workflow slm slides site sft self-reinforcing-error self-learning self-generated-context self-consistency scaling-laws retention research-plan research remode reliability redesign recommender-system publish-state pruning prosody projector proj-memory proj-dialogue procedural-memory probing preference ppo post-retrieval pomdp plan persona pbrl partial-observability partial-credit partial parallel-sampling papers paper outcome-based-evaluation omni-modal omni observer multi-party multi-modal moe modality-preference mllm mid mia memory-management memory-deletion mcp lvlm lost-in-conversation long-term-memory llm-judge llm-agent llm literature-review listing linear-rnn late-interaction korean-bias korean knowledge-graph kakaotalk intervention incident implicit-conflict hypernetwork human-reference human-in-the-loop hci gpu git gan fusion fullset federated-learning false-finish failure-taxonomy experiential-knowledge experience-following exp_c executable-evaluation error-propagation ensemble-scoring embedding dst distributional-analysis disambiguation disagreement dense-reward decoding debug debate data-selection data-management conversational-memory conversational-agent conversation-history contrastive-learning confidence-estimation conference comparison cognitive-science coding clip cli claude-code classification chatbot causal-intervention bfcl benchmark-evaluation belief-state batch baseline audio attentional-bias appropriate-reliance analysis activation-steering Qwen MiniCPM HCX

View posts list →

Enter your search term...