Links for 2025-01-15
AI:
Google presents the successor to the Transformer architecture: Titans marks a significant step in neural network architecture by integrating a bio-inspired long-term memory mechanism that complements the short-term context modeling of traditional attention mechanisms. A key innovation is that the memory module is trained to learn how to memorize and forget during test time. This allows the model to adapt to new, unseen data distributions, which is crucial for real-world applications. The way Titans decides what to memorize is inspired by how the human brain prioritizes surprising or unexpected events. The authors introduce the concept of "momentary surprise" (how much a new input deviates from the model's current understanding) and "past surprise" (a decaying record of past surprises) to guide the memory module's updates. This mirrors the human tendency to remember events that stand out from the norm. https://arxiv.org/abs/2501.00663
Transformer^2: Self-adaptive LLMs — dynamically adapts to new tasks in real-time, using smart "expert" vectors to fine-tune performance. https://sakana.ai/transformer-squared/
Multiagent Finetuning: Self Improvement with Diverse Reasoning Chains https://llm-multiagent-ft.github.io/
Imagine while Reasoning in Space: Multimodal Visualization-of-Thought — MVoT moves beyond Chain-of-Thought (CoT) to enable AI to imagine what it thinks with generated visual images. By blending verbal and visual reasoning, MVoT makes tackling complex problems more intuitive, interpretable, and powerful. https://arxiv.org/abs/2501.07542
VideoRAG: A framework that enhances RAG by leveraging video content as an external knowledge source. https://arxiv.org/abs/2501.05874
O1 Replication Journey -- Part 3: Inference-time Scaling for Medical Reasoning https://arxiv.org/abs/2501.06458
The Lessons of Developing Process Reward Models in Mathematical Reasoning https://arxiv.org/abs/2501.07301
The Future of AI: Exploring the Potential of Large Concept Models https://arxiv.org/abs/2501.05487
UC Berkeley releases a $450 open-source reasoning model that matches o1-preview https://novasky-ai.github.io/posts/sky-t1/
Inference-Time-Compute: More Faithful? A Research Note https://www.lesswrong.com/posts/C8HAa2mf5kcBrpjkX/inference-time-compute-more-faithful-a-research-note
MatchAnything: Universal Cross-Modality Image Matching with Large-Scale Pre-Training https://zju3dv.github.io/MatchAnything/
Chick-fil-A’s Lemon-Squeezing Robots Are Saving 10,000 Hours of Work https://www.bloomberg.com/features/2024-chick-fil-a-lemonade/ [no paywall: https://archive.is/huu7f]
OpenAI is starting to build its own robotics team, hiring for its first hardware roles. https://venturebeat.com/ai/openai-has-begun-building-out-its-robotics-team/
Google’s Gemini AI has quietly upended the AI landscape, achieving a milestone few thought possible: The simultaneous processing of multiple visual streams in real time. https://venturebeat.com/ai/google-gemini-ai-just-shattered-the-rules-of-visual-processing-heres-what-that-means-for-you/
MiniMax-01 is Now Open-Source: Scaling Lightning Attention for the AI Agent Era https://www.minimaxi.com/en/news/minimax-01-series-2
“For the first time, ChatGPT can manage tasks asynchronously on your behalf—whether it's a one-time request or an ongoing routine.” https://x.com/karinanguyen_/status/1879270529066262733
Sam Altman says he now thinks a fast AI takeoff is more likely than he did a couple of years ago, happening within a small number of years rather than a decade https://x.com/tsarnick/status/1879100390840697191
“OpenAI’s AI reasoning model ‘thinks’ in Chinese sometimes and no one really knows why” https://techcrunch.com/2025/01/14/openais-ai-reasoning-model-thinks-in-chinese-sometimes-and-no-one-really-knows-why/
AI politics:
President Biden signed an executive order opening federal land for the development of gigawatt-scale datacenters. The DoD and DoE will both lease land, and sufficient clean energy to match capacity must be built on site. "Clean energy" includes nuclear fission and nuclear fusion(!) https://www.whitehouse.gov/briefing-room/presidential-actions/2025/01/14/executive-order-on-advancing-united-states-leadership-in-artificial-intelligence-infrastructure/
“…even though standard measures of AI quality scale poorly as a function of resources, the financial returns might still scale very well as a function of resources. Indeed, if they scale better than linearly, that would create a paradigm of increasing marginal returns…” https://www.tobyord.com/writing/the-scaling-paradox
Applying traditional economic thinking to AGI: a trilemma https://www.lesswrong.com/posts/TkWCKzWjcbfGzdNK5/applying-traditional-economic-thinking-to-agi-a-trilemma
UK Prime Minister sets out blueprint to turbocharge AI https://www.gov.uk/government/news/prime-minister-sets-out-blueprint-to-turbocharge-ai
A Spymaster Sheikh Controls a $1.5 Trillion Fortune. He Wants to Use It to Dominate AI https://www.wired.com/story/uae-intelligence-chief-ai-money/ [no paywall: https://archive.is/RFAO7]
Bio(tech):
Nanocarrier imaging at single-cell resolution across entire mouse bodies with deep learning https://www.nature.com/articles/s41587-024-02528-1
New computational chemistry techniques accelerate the prediction of molecules and materials https://news.mit.edu/2025/new-computational-chemistry-techniques-accelerate-prediction-molecules-materials-0114
ChemAgent: Self-updating Library in Large Language Models Improves Chemical Reasoning https://arxiv.org/abs/2501.06590
About 5% of cyanobacteria fished from the ocean are connected via nanotubes. These nanotubes are made from lipid membranes. Also, E. coli will sometimes grab onto microbes of *other* species, using these nanotubes, and share nutrients. https://www.quantamagazine.org/the-ocean-teems-with-networks-of-interconnected-bacteria-20250106/
The use of genetically engineered bacteria to recover or recycle chemicals and turn them into useful products is progressing fast https://www.bbc.com/news/articles/cz6pje1z5dqo
Heritability: what is it, what do we know about it, and how we should think about it? https://www.lesswrong.com/posts/xXtDCeYLBR88QWebJ/heritability-five-battles
Synchron to Advance Implantable Brain-Computer Interface Technology with NVIDIA Holoscan https://www.businesswire.com/news/home/20250113376337/en/Synchron-to-Advance-Implantable-Brain-Computer-Interface-Technology-with-NVIDIA-Holoscan
A Mathematical Perspective on Neurophenomenology https://arxiv.org/abs/2409.20318
Politics:
Young people are increasingly avoiding relationships, and it's happening all over the world. Japan's already low marriage rate dropped another 12% since 2019, while a third of 18-34 year-olds globally say they're just not interested in dating or relationships. https://www.ft.com/content/43e2b4f6-5ab7-4c47-b9fd-d611c36dad74 [no paywall: https://archive.is/kyk2L]
In 1965 the US government tried replacing Mexican farmworkers with American high school athletes, and it failed spectacularly. The Department of Labor recruited 18,100 teenagers for the "A-TEAM" program but only 3,300 actually worked the fields, with many quitting within weeks due to brutal conditions like 110-degree heat and minimum wage pay. https://www.npr.org/sections/thesalt/2018/07/31/634442195/when-the-u-s-government-tried-to-replace-migrant-farmworkers-with-high-schoolers
New AI Climate Simulator that you can play with. Visualize how geoengineering can slow global warming. Lets you explore how geoengineering via Stratospheric Aerosol Injection (SAI) gives us new paths to keep warming to 1.5 degrees. Reflecting 1% of sunlight away from earth would lead to an extra ~1 degree of cooling. https://www.planetparasol.ai/
The White House scrambled to get a message to President Vladimir V. Putin of Russia last year after U.S. intelligence agencies said a Russian military unit was preparing to send explosive packages on cargo planes. https://www.nytimes.com/2025/01/13/us/politics/russia-putin-airplane-shadow-war.html [no paywall: https://archive.is/dc7LP]
In Sweden, 63% of those convicted of rape between 2000 and 2020 were immigrants. [PDF] https://journals.sagepub.com/doi/pdf/10.1177/08862605241311611
Ukraine:
I guess most people are rather oblivious to the fact that there are now rather large attacks on Russian infrastructure every few days. Putin's 3-day special military operation truly brought war to Russian soil. https://x.com/NOELreports/status/1879139422031196490
An overview of a Ukrainian shotgun-drone, taking down multiple Russian drones. https://x.com/NOELreports/status/1879462367870034046

teenagers would be the worst age group to choose to do this, they have no staying power at all, it can take them an hour to slice a onion