TWIML

Yejin Choi discusses her research on making small language models reason more effectively, challenging the assumption that scale is the only path to intelligence.

Nikita Rudin discusses the gap between current robotic capabilities and what is required to deploy fully autonomous robots in the real world.

Aakanksha Chowdhery from Reflection explores the fundamental shifts required to build true agentic AI that can plan, act, and learn autonomously.

Munawar Hayat from Qualcomm AI Research discusses key papers from NeurIPS 2025, including advances in multimodal AI and solutions for object hallucination in vision-language models.

Zain Asgar from Gimlet Labs discusses running AI inference across heterogeneous hardware — mixing GPUs, CPUs, and custom accelerators for optimal performance.

Devi Parikh, co-founder of Yutori, discusses browser use models and how autonomous AI agents can navigate and act within web environments.

Robin Braun from HPE and Luke Norris from Kamiwaza discuss enterprise AI infrastructure challenges and how organizations can scale AI deployment effectively.

A deep dive into reasoning architectures: how chain-of-thought, tree-of-thought, and other inference-time compute strategies work and when to use each.

Advances in robotic manipulation: how AI systems learn to grasp, move, and manipulate objects in unstructured environments.

Research on LLM security: prompt injection attacks, jailbreaking techniques, data extraction, and emerging defense strategies.

Insights from training large foundation models: data curation, compute allocation, and the increasingly important role of data quality over data quantity.

Moving from responsible AI principles (fairness, transparency, accountability) to practical implementation — the organizational structures, evaluation pipelines, and cultural changes required to build AI responsibly.

How diffusion models are being applied beyond image generation: video, 3D modeling, protein design, and audio synthesis.

How organizations must rethink data governance for AI: new challenges around training data provenance, model outputs, and the blurred line between data and code.

Runway CEO Cristobal Valenzuela discusses whether video generation represents a path toward artificial general intelligence — the argument that world models trained on video data must develop physical reasoning.

The state of reinforcement learning: from game-playing to robotics to RLHF. How RL has evolved from a research curiosity to a core component of modern AI systems.

Bridging the gap between AI ethics theory and practice. How organizations can move from principled statements to concrete implementation of ethical AI.

The emerging field of edge AI deployment — running machine learning models on devices (phones, cameras, sensors) rather than in the cloud. The challenges of model compression, latency, and power efficiency.

Alexandre Pesant from Lovable discusses the evolution and practice of vibe coding — building software through natural language descriptions rather than manual code.

Atlassian co-CEO Mike Cannon-Brookes shares internal data on AI tool adoption across 300,000+ enterprise customers — what AI features are actually used, which are ignored, and what this reveals about enterprise AI readiness.

A clinical informatics researcher separates the hype from reality in healthcare AI: which applications are actually deployed in clinical settings, which remain research demonstrations, and why the gap exists.

A critical examination of AI safety evaluation methodologies: red-teaming, benchmarks, and stress testing — what they actually measure, what they miss, and why the field lacks agreed-upon standards.

Arvind Jain, CEO of Glean, discusses the evolution from enterprise search to agentic AI tools for the workplace — how Glean went from finding documents to acting on them.

The convergence of vision, language, and audio in multimodal generative AI systems — how models that can see, speak, and write simultaneously are creating new capabilities and new challenges.

GitHub CEO Thomas Dohmke discusses how AI is transforming software engineering through Copilot and beyond — the shift from AI as autocomplete to AI as a development partner.

How AI is transforming scientific discovery beyond the headline achievements of AlphaFold — applications in materials science, drug discovery, climate modeling, and mathematics.

A deep dive into synthetic data: how AI-generated training data is replacing real data in many applications, the quality challenges, and the risk of model collapse when models are trained on outputs of other models.

Christopher Manning from Stanford discusses the intersection of linguistics and large language models — whether LLMs truly understand language or process it as statistical patterns, and what this distinction means for the field.

A nuanced discussion of the risks and benefits of releasing open-weight AI models, challenging both the AI safety maximalist view (all models should be closed) and the open-source maximalist view (all models should be open).

The AI2 team discusses OLMo, their fully open-source large language model that releases not just weights but training data, training code, and evaluation frameworks — everything needed to reproduce and study the model.

An exploration of how training data composition affects chain-of-thought reasoning capabilities in LLMs, and whether prompting techniques actually elicit genuine reasoning or sophisticated pattern matching.

JPMorgan AI Research discusses DocLLM, their approach to reasoning over complex business documents that combine text, tables, and figures — a fundamental challenge for enterprise AI deployment.

A critical examination of whether emergent abilities in large language models are genuine phase transitions or measurement artifacts caused by nonlinear evaluation metrics.

Kamyar Azizzadenesheli from NVIDIA discusses how the reinforcement learning community is adapting to the dominance of large language models — where RL techniques enhance LLMs and where LLMs change RL.

Naila Murray discusses 2024 trends in computer vision: the integration of vision with language models, the challenge of 3D understanding, and how video generation is reshaping the field.

Thomas Dietterich discusses 2024 AI trends in machine learning and deep learning, including the tension between monolithic LLMs and modular architectures, hallucination challenges, and the role of uncertainty quantification.