Links for 2024-02-16
AI:
Mind-blowing: OpenAI just released a next-genration AI model that can create realistic and imaginative scenes from text instructions. Sora can generate videos up to a minute long while maintaining visual quality and adherence to the user’s prompt. https://openai.com/sora
Google releases Gemini 1.5 Pro. It features sophisticated multimodal understanding and reasoning capabilities with long context. When given a 44-minute silent film, the model can analyze various plot points and events, and even makes sense of small details you might have missed. Gemini 1.5 was designed using a new Mixture–of-Experts (MoE) architecture, making it much more efficient to train and serve. Gemini 1.5 Pro can consistently run up to 1 million tokens in production, equivalent to: 🔠 Over 700,000 words🛠 Over 30,000 lines of code 🔊 11 hours of audio 📹 1 hour of video https://blog.google/technology/ai/google-gemini-next-generation-model-february-2024/
Meta announces V-JEPA: A method for teaching machines to understand and model the physical world by watching videos. This work is another important step towards AI models that use a learned understanding of the world to plan, reason and accomplish complex tasks. https://ai.meta.com/blog/v-jepa-yann-lecun-ai-model-video-joint-embedding-predictive-architecture/
“If you think OpenAI Sora is a creative toy like DALLE, ... think again. Sora is a data-driven physics engine. It is a simulation of many worlds, real or fantastical. The simulator learns intricate rendering, "intuitive" physics, long-horizon reasoning, and semantic grounding, all by some denoising and gradient maths.” https://x.com/DrJimFan/status/1758210245799920123 (see also: Video generation models as world simulators https://openai.com/research/video-generation-models-as-world-simulators)
More mind-blowing facts about Gemini 1.5: “v1.5 learns to translate from English to Kalamang purely in context, following a full linguistic manual at inference time. Kalamang is a language spoken by fewer than 200 speakers in western New Guinea. Gemini has never seen this language during training and is only provided with 500 pages of linguistic documentation, a dictionary, and ~400 parallel sentences in context. It basically acquires a sophisticated new skill in the neural activations, instead of gradient finetuning.” https://x.com/DrJimFan/status/1758275649373151671
Connor Leahy on Gemini 1.5: “This is the kind of stuff that makes me think that there will be no period of sorta stupid, human-level AGI. Humans can't perceive 3 hours of video at the same time. The first AGI will instantly be vastly superhuman at many, many relevant things.” https://x.com/NPCollapse/status/1758168495164944816
New AI tool discovers realistic 'metamaterials' with unusual properties https://www.tudelft.nl/en/2024/me/news/new-ai-tool-discovers-realistic-metamaterials-with-unusual-properties
Using AI to discover stiff and tough microstructures https://news.mit.edu/2024/using-ai-discover-stiff-tough-microstructures-0214
Google Deepmind presents Transformers Can Achieve Length Generalization But Not Robustly https://arxiv.org/abs/2402.09371
Evan Hubinger (Anthropic)—Deception, Sleeper Agents, Responsible Scaling https://www.youtube.com/watch?v=S7o2Rb37dV8
Miscellaneous:
Homeostatic Feelings and the Emergence of Consciousness — “We propose that a mind can be considered conscious when three processes are in place…” https://direct.mit.edu/jocn/article-abstract/doi/10.1162/jocn_a_02119/119429/Homeostatic-Feelings-and-the-Emergence-of
Researchers design a processor from DNA — microfluidic chip completes math calculations and also stores data in DNA https://www.tomshardware.com/pc-components/storage/researchers-design-cpu-from-microfluidic-dna-processor-completes-math-calculations-and-also-stores-data-in-dna
International research team develops new hardware for neuromorphic computing https://www.lboro.ac.uk/schools/science/news/2024/international-research-team-develops-new-hardware/
Politics:
Why is the U.S. Navy Running Out of Tomahawk Cruise Missiles? https://nationalinterest.org/blog/buzz/why-us-navy-running-out-tomahawk-cruise-missiles-209317
Drones to replace helicopter program that Army developed at $2 billion cost https://www.stripes.com/branches/army/2024-02-09/army-helicopter-drones-12949044.html
We are Legion: One person 'swarm commander' can now control 100 drones https://interestingengineering.com/innovation/one-person-swarm-commander
“Tucker Carlson went to Russia and was wowed by what he saw. But in reality, Russia is a dysfunctional, violent, irreligious middle-income country with widespread poverty and relatively weak family ties.” https://www.noahpinion.blog/p/russia-is-not-actually-a-very-nice
“The Tucker Carlson grocery price video (Russia is so cheap you'll be radicalized, folks!) is tragic & funny. My guy, the grocery bill you're rhapsodizing about is ~SEVENTY PERCENT of a median Russian weekly salary (13.4k RUB)” https://x.com/jsrailton/status/1758254151589409192
Ukraine:
Avdiivka has fallen. Russia has been fighting to achieve this for almost a decade. In the last 4 months alone, they have lost over 650 vehicles here. https://x.com/ThomasVLinge/status/1758228790998675639
A harrowing account from a soldier with Ukraine's 110th Mechanized Brigade who was in the Zenit pocket south of Avdiivka. https://www.instagram.com/p/C3YlMYUtZON/
"A group of allies is going to join forces and transfer 1,000,000 drones to Ukraine, while 20 NATO allies have also agreed to create a mine clearance coalition for Ukraine," NATO chief Stoltenberg said. https://x.com/NOELreports/status/1758142168684753366
Destruction of a whole russian column using FPV drones in Lyman direction by the 60th and 63rd Brigades. https://x.com/wartranslated/status/1758232644733603921
“Information appeared that the ballistic missile that fell in Kyiv region is reportedly a North Korea KN-23 one. All the trees in the 40 meter radius burned down from the impact” https://x.com/Gerashchenko_en/status/1758202296993783821