SpaceX · 2025
The official recap of Starship’s seventh flight test, which flew on January 16, 2025. Historical aerospace context; this is not footage of a 2026 flight.
A DIFFERENT WAY IN
Watch the research.
Visual explanations and company demonstrations. Original dates and credits stay attached.
17 videos · Watch here, free
Tesla · 2023
Tesla’s December 2023 demonstration of its second-generation Optimus robot. A company showcase of that prototype, not evidence of present-day production or independent task performance.
3Blue1Brown · 2017
A visual, math-grounded introduction to neural networks and gradient-based learning.
3Blue1Brown · 2024
A visual explanation of attention and transformer architecture for non-specialist readers.
Google DeepMind · 2025
An official documentary view of scientific research culture and the development of AlphaFold.
NVIDIA / Skild AI · 2026
Skild AI demonstrates its S1 robot model learning from a video. This is a company demonstration, not an independent capability evaluation.
Anthropic · 2025
Anthropic researchers trace computational pathways behind selected Claude responses. Their examples explore planning and intermediate representations; they do not reveal every computation in a language model.
Google DeepMind · 2025
Google DeepMind demonstrates Genie 3 generating interactive environments from text. This is generated world-model footage with the lab’s own selected examples, not a recording of a physical place.
OpenAI · 2024
OpenAI’s May 2024 presentation shows GPT-4o working across voice, images, and text. Keep the original launch date in mind: this documents that release rather than the capabilities or availability of current models.
Stanford / Zipeng Fu · 2024
The Mobile ALOHA team demonstrates a low-cost system for collecting whole-body robot demonstrations. The robot in this housekeeping video is teleoperated by a person; the accompanying paper separately evaluates policies learned from demonstrations.
Andrej Karpathy · 2023
Andrej Karpathy builds and trains a small transformer in Python, showing how attention, token prediction, and the training loop fit together. A detailed coding lesson rather than a frontier-model benchmark.
NVIDIA · 2025
NVIDIA introduces Cosmos and its use of world models and simulated data for physical AI. The animation and selected demos illustrate the company’s proposed workflow, not a guarantee of real-world robot performance.
MIT / Alexander Amini · 2025
Alexander Amini introduces neural networks and the basic ideas behind deep learning in MIT’s 2025 course. The original course page also provides slides and practical labs.
MIT 6.S191 · 2025
MIT’s 2025 lecture on generative modeling explains how learned distributions can produce new examples. Part of the public 6.S191 course, with slides available from the course site.
MIT 6.S191 · 2025
A lecture from MIT’s 2025 deep-learning course on learning through actions, feedback, and rewards. Use the companion course materials to work through the technical details.
MIT / Ava Amini · 2025
Ava Amini discusses generative design and interactive discovery for biology in MIT’s 2025 course. The lecture connects AI methods with physical experiments and protein design.
Meta FAIR · 2025
Meta’s V-JEPA 2 research video shows prediction and planning examples from a model trained on video and then robot interaction data. These are the research team’s demonstrations, with the tested setup described in the linked publication.
Company demonstrations show what the company chose to present. Historical footage is not evidence of current performance.