Best AI papers explained
by Enoch H. Kang
Latest episode
H-Spec: Parallel Speculative Decoding Without a Drafter-Side KV Cache
H-Spec: Parallel Speculative Decoding Without a Drafter-Side KV Cache
About this podcast
Cut through the noise. We curate and break down the most important AI papers so you don’t have to.
Episodes
The newest 50 of 859 episodes
- H-Spec: Parallel Speculative Decoding Without a Drafter-Side KV Cache · 22 minH-Spec is a novel parallel speculative decoding framework designed to accelerate large language models by eliminating the memory-intensive drafter-side Key-Value cache used by…
- Mass producing training data with surprising side effects · 24 minThis paper detail a comprehensive red-teaming and evaluation framework designed to test how supervised fine-tuning influences language model behavior. Specifically, the…
- User Embeddings are Superpositions of Interpretable Behavioral Modes · 24 minThis paper explores how sparse autoencoders can be applied to user embeddings to uncover hidden behavioral modes without needing raw interaction data or text-aligned training. By…
- Training Foundation Models with a Single Algorithm · 23 minFoundation models are traditionally trained through a sequential pipeline of distinct phases, including pretraining, fine-tuning, and reinforcement learning. Researchers…
- Sparks of Weight-Space Programming: LLMs can natively read & write neural network weights · 23 minFrontier large language models demonstrate a surprising capacity for weight-space programming, meaning they can naturally comprehend, evaluate, and modify the raw parameters of…
- Quantifying Elicitation of Latent Capabilities in Language Models · 21 minThis paper investigates how large language models unlock latent capabilities through fine-tuning, reframing the process as an information-constrained problem. By analyzing…
- The Trinity: Capability-Driven Recursive Self-Improvement for Multimodal Agents · 20 minThis paper introduce The Trinity, a recursive self-improvement framework designed to scale multimodal intelligence by coupling automated environment generation with agent…
- High-Value Training Data from Noisy, Diverse Sources · 20 minTraining advanced machine learning models typically relies on vast, heterogeneous datasets containing noisy, adversarial, or mismatched data. To address this challenge, the…
- Predictable Performance in Model Routing and Cascades · 22 minRecent research explores performance prediction and model routing in large language models by studying when and why an AI's errors are predictable. The authors demonstrate that…
- Asymmetric On-Policy Distillation Learns In-Context Exploration · 23 minWe discuss ASTEROID, a novel student-teacher framework designed to address the challenges of sequential decision-making under partial observability. By combining on-policy…
- Generalization Dynamics of Language Model Pretraining · 23 minLarge language models often exhibit unexpected behavioral fluctuations during pre-training, repeatedly shifting between memorizing shallow patterns and demonstrating true…
- Scaling Discovery Requires Forgetting · 21 minAutomated research agents powered by large language models often experience performance plateaus during open-ended discovery tasks as their accumulated history makes their edits…
- One-Sided Verification Under Limited Expert Review · 23 minAI-driven scientific discovery increasingly relies on automated verification to evaluate candidate solutions, but the scarcity of reliable automated metrics makes human expert…
- Scaling Empowered Agents · 25 minThis paper introduces a scalable algorithm called Skills Estimate Empowerment (SEE) that bridges unsupervised skill learning and empowerment maximization to train intelligent…
- Hybrid Reasoning Agents · 20 minThis paper introduces HybridAgents, a novel framework that blends continuous latent reasoning with explicit text interactions within a unified generation framework. By utilizing…
- Strategic Self-Consistency · 25 minThis research demonstrates that unfaithful large language model service providers can artificially inflate the number of reasoning paths generated during test-time compute to…
- Retrofitting language models to operate over bytes · 24 minThis Nature paper discusses Byteification, which offers a novel method to convert traditional subword-based large language models into efficient byte-level models that handle raw…
- Priced Guidance: Can Language Models Generate Future Research Ideas? · 20 minThis paper introduces Priced Guidance, a novel evaluation framework that assesses language models on their ability to generate novel and future research ideas. Rather than using…
- Context Language Models · 22 minThis paper introduce Context Language Models (CLMs), a novel approach that enables language models to natively manage their own context by treating it as an editable file rather…
- Position: AI-Agent Pricing Should Become More Outcome-Dependent: An Economic Perspective · 23 minThis paper analyzes AI-agent pricing models through an economic lens, arguing that dominant token-based pricing is insufficient because it fails to share risk efficiently or…
- Emergent Capabilities Arise Randomly from Learning Sparse Attention Patterns · 24 minThis paper investigates why transformer language models display sudden, emergent capabilities during training by analyzing how they learn attention patterns. Through experiments…
- Pragmatic Causal Inference with AI-Learned Representations · 23 minThis paper presents a rigorous framework for performing valid causal inference when complex covariates are compressed into AI-learned representations or embeddings. The authors…
- Pointwise or pairwise: when do pairwise losses help reward learning, provably? · 22 minThis research paper investigates when and why pairwise losses outperform pointwise losses for reward learning by analyzing both approaches within a grouped offline…
- When to Trust the Cheap Check: Weak and Strong Verification for Reasoning · 25 minThis paper explores weak-strong verification policies for large language models, presenting a framework that balances the affordability of scalable internal checks with the…
- Theoretical Limits of Language Model Alignment · 24 minThis paper investigate the fundamental limits of language model alignment by establishing exact theoretical boundaries for reward improvement under a KL-divergence constraint…
- DT2: Decision-Targeted Digital Twins · 13 minThis paper introduces DT2, a novel training framework designed to align digital twins more effectively with their primary goal of decision support. Traditional virtual models…
- Self-Play Pretraining with Zero Data · 22 minThis paper introduces Self-Play Pretraining with Zero Data, a method for training language models using only synthetic data generated by the model itself. In this framework, a…
- Language models need sleep: learning to self-modify and consolidate memories · 21 minThis research paper introduces a "Sleep" paradigm for Large Language Models (LLMs) to overcome the limitations of static knowledge and catastrophic forgetting. Inspired by human…
- Jev Creator: System One models for Prod, not God · 22 minWe discuss the Latent Space's interview with Diogo Almeida, CEO of TypeSafe, about the launch of Jev, a specialized class of AI models designed for programmatic integration…
- Detecting and countering misuse of AI: September 2026 · 23 minThis paper is an Anthropic threat intelligence report from September 2026 detailing the misuse of AI models by various malicious actors. It describes how state-sponsored groups…
- Position: LLMs can’t jump · 21 minThis paper examines the cognitive limitations of Large Language Models by using Albert Einstein’s discovery of General Relativity as a primary case study. While modern AI excels…
- Jailbreaking Jailbreaks: A Proactive Defense for LLMs · 23 minThe research introduces PROACT, a proactive defense framework designed to safeguard Large Language Models from iterative adversarial attacks. Unlike traditional passive defenses…
- When Agents Slow Down: Understanding LLM Agents’ Test-Time Strategies via Elo-per-token Analysis · 22 minThis paper introduces Elo-per-token analysis, a novel framework for measuring how the performance of large language model agents scales with increased inference-time computation…
- Thinking with Looped Flows · 20 minThis paper introduces looped flows, a novel framework designed to enhance the reasoning capabilities of neural networks by merging recurrent hidden states with probability flow…
- Multi-Turn LLM Conversations under the Least-Recently-Used Policy: Mean-Field Asymptotics · 22 minThis paper introduces a Mean-Field Asymptotic framework designed to estimate the hit ratio in multi-turn large language model (LLM) serving systems. As conversations grow in…
- Breaking the Token Ceiling: Distilling Smaller, Stronger Byte Models · 24 minThis research introduces Marginalize-It and End-Of-Token, two novel methods for efficiently distilling large token-based language models into smaller, more capable byte-level…
- Thinking to Recall: How Reasoning Unlocks Parametric Knowledge in LLMs · 21 minThis research explores how reasoning helps Large Language Models (LLMs) answer simple, single-hop factual questions that do not logically require step-by-step thinking. The…
- Tail-Likelihood Reinforcement Learning · 21 minThis paper introduces Tail-Likelihood Reinforcement Learning (TailRL), a novel optimization framework designed to improve how generative policies handle continuous rewards…
- Next-Latent Prediction Transformers Learn Compact World Models · 23 minThis paper introduces Next-Latent Prediction (NextLat), a novel training framework designed to help Transformer models learn more compact and generalizable internal world models…
- Language Models Can Control Their Own Attention · 20 minResearchers have introduced Declarative Attention (DA), a protocol that enables large language models to autonomously manage their own focus during long-context tasks…
- AI Finds A Way · 26 minThis paper introduces a comprehensive collection of anecdotes documenting instances where artificial intelligence systems developed innovative yet unpredictable solutions. While…
- Superposition Without Interference? Towards Isolated Interventions via Almost Orthogonal Features in Language Models · 24 minThis research paper investigates how feature entanglement in large language models prevents precise, localized interventions on specific concepts. The authors argue that because…
- TTPO: Test-Time Policy Optimization · 23 minThis paper introduces Test-Time Policy Optimization (TTPO), a novel method for improving the mathematical reasoning of large language models without using ground-truth labels…
- Demystifying Reinforcement Learning Post-Training of Language Models · 20 minThis paper deconstructs the mechanics of reinforcement learning (RL) post-training for large language models to determine how different factors influence model performance. By…
- Recursive Experiential–Working Memory Evolution for Long-Horizon Agent Harnesses · 21 minRecuris is a recursive architectural framework designed to enhance the performance of large language model agents during complex, long-horizon tasks. By coupling Working Memory…
- TailSFT: Filtered Fine-Tuning Improves Post-Training Performance · 22 minResearchers introduce TailSFT, a modified supervised fine-tuning algorithm designed to better prepare language models for subsequent reinforcement learning. Unlike standard…
- SPADE: Self-Play in Adaptive Synthetic Executable Environments · 22 minThis paper introduces SPADE, a reinforcement learning framework that enables a single large language model to achieve open-ended self-improvement by designing its own training…
- Evaluating Skills, Not Just Agents: Agentic Continuous Evaluation of Skills · 28 minThis paper introduces ACES (Agentic Continuous Evaluation of Skills), a comprehensive framework developed by NVIDIA to move beyond static document scanning when assessing AI…
- Impression Share Prediction: An Offline Evaluation Task for Ranking Systems · 22 minResearchers from Meta Platforms propose a novel offline evaluation task called impression share prediction to better anticipate how new ranking models redistribute traffic across…
- Q-Learning with World Models · 25 minThe researchers introduce Q-Learning with World Models (QWM), a framework designed to enhance sample efficiency and performance in robotic reinforcement learning. Unlike…
