Links for 2024-05-22
AI:
If true, this might very well be the most important statement ever made: “we are nowhere near the point of diminishing marginal returns on how powerful we can make AI models as we increase the scale of compute” https://x.com/XiXiDu/status/1793046291217318160
BioInformatics Agent (BIA): Unleashing the Power of Large Language Models to Reshape Bioinformatics Workflow https://www.biorxiv.org/content/10.1101/2024.05.22.595240v1
Diffusion for World Modeling: Visual Details Matter in Atari https://arxiv.org/abs/2405.12399
Aggregation of Reasoning: A Hierarchical Framework for Enhancing Answer Selection in Large Language Models https://arxiv.org/abs/2405.12939
Adobe introduced generative removal in Lightroom. It allows you to remove people, distractions, unwanted objects, and more from images. https://www.youtube.com/watch?v=9k2LROVsVjg
Photoshop is testing out a new feature called "Reference Image." https://x.com/Pinsky/status/1791123362216894691
Octo: An Open-Source Generalist Robot Policy https://octo-models.github.io/
“I built my own omni assistant using Gemini 1.5 Flash to guide me through Super Mario 64. Gemini can see what I do on my screen and communicate with me in real time via voice, and thanks to the long 1M context, it has a memory of everything we do together.” https://x.com/skirano/status/1792948429754151293
The Zhousidun dataset, a Chinese-originated dataset that exhaustively labels critical components on American and Allied destroyers https://arxiv.org/abs/2405.12167v1
AI companies:
ChatGPT on Track for 2 Billion Visits in May, After Topping 100 Million Daily Visits Twice Last Week https://www.similarweb.com/blog/insights/ai-news/chatgpt-two-billion/
“Our calling is to build the data foundry for AI, and with today’s funding, we’re moving into the next phase of that journey – accelerating the abundance of frontier data that will pave our road to AGI.” https://scale.com/blog/scale-ai-series-f
H company founded ex-DeepMind employees and headquartered in France has secured 220M$ in a seed round. https://x.com/laurentsifre/status/1793045816740819161
AI safety:
Mapping the Mind of a Large Language Model — “Today we report a significant advance in understanding the inner workings of AI models. We have identified how millions of concepts are represented inside Claude Sonnet, one of our deployed large language models. This is the first ever detailed look inside a modern, production-grade large language model. This interpretability discovery could, in future, help us make AI models safer.” https://www.anthropic.com/research/mapping-mind-language-model (Review: https://www.lesswrong.com/posts/pH6tyhEnngqWAXi9i/eis-xiii-reflections-on-anthropic-s-sae-research-circa-may)
Using Degeneracy in the Loss Landscape for Mechanistic Interpretability https://arxiv.org/abs/2405.10927
The Local Interaction Basis: Identifying Computationally-Relevant and Sparsely Interacting Features in Neural Networks https://arxiv.org/abs/2405.10928
New in Science: Managing extreme AI risks amid rapid progress. Co-authored by 3 Turing Award winners, the most-cited scholar in computer security, a former President of Baidu, and many others. https://www.science.org/doi/10.1126/science.adn0117
Historic first as companies spanning North America, Asia, Europe and Middle East agree safety commitments on development of AI https://www.gov.uk/government/news/historic-first-as-companies-spanning-north-america-asia-europe-and-middle-east-agree-safety-commitments-on-development-of-ai
Some arguments that OpenAI is making, simultaneously https://forum.effectivealtruism.org/posts/9iLgAbu9KtpLupitk/what-s-going-on-with-openai-s-messaging
“RIP California. This law includes a standard that can not reasonably be met: if any expert can modify a model to become hazardous, the model is considered to be hazardous.” https://x.com/xlr8harder/status/1793139820354740597
Neuroscience and Neurotech:
No effect of additional education on long-term brain structure: a preregistered natural experiment in over 30,000 individuals. https://www.biorxiv.org/content/10.1101/2024.05.17.594682v1
Bionic Eye Gets a New Lease of Life — Ex-Neuralink exec Max Hodak's new company rescues Pixium technology https://spectrum.ieee.org/pixium-bionic-eye-tech-rescued
Volatile working memory representations crystallize with practice https://www.nature.com/articles/s41586-024-07425-w
Miscellaneous:
The day when human sperm and eggs can be grown in the laboratory has inched a step closer https://www.nature.com/articles/d41586-024-01404-x
Andrej Bauer: Constructive Mathematics - How to not believe in the Law of Excluded Middle https://www.youtube.com/watch?v=96iHUx0aGDs
“In the '90s, my father was part of the US’s $100-200B effort to mitigate the Y2K bug, which was an almost complete success, with the result that nowadays most people think it was a myth, hoax, or urban legend. This is the best-case scenario for AI safety.” — davidad https://x.com/davidad/status/1792984074182689098
Politics:
NATO Deputy SecGen: "Agencies have warned of Russian hackers, linked to Russia's GRU, targeting infrastructure, including water treatment plants in Texas and Indiana, Poland and France. Recently, there was a sustained attack on a European port" https://www.nato.int/cps/en/natohq/opinions_225387.htm
Finland's president responds to reports that Russia is seeking to redraw Baltic maritime borders near Lithuania/Finland https://x.com/alexstubb/status/1793192475965686066
Janet Yellen urges EU to join US in curbs on cheap Chinese exports https://www.theguardian.com/business/article/2024/may/21/janet-yellen-eu-us-chinese-exports-tariffs-electric-vehicles
Wang Wenbin: "China will and must achieve reunification. No one or force can hold it back. Our message to some in the US: The trend of the world is surging forward. Those who follow the trend will prosper; those who go against it will perish." https://www.fmprc.gov.cn/mfa_eng/xwfw_665399/s2510_665401/2511_665403/202405/t20240520_11308199.html
Ukraine:
UK defence minister says intelligence has evidence of Chinese lethal aid to Russia https://www.reuters.com/world/uk-defence-minister-says-intelligence-has-evidence-chinese-lethal-aid-russia-2024-05-22/ [no paywall: https://archive.is/ywk8V]
“Operators of the 3rd SSO regiment destroyed half a division of Russian Buk anti-aircraft missile systems in one of the operational areas in just 3 weeks. Three SAMs were destroyed in late April and early May. The fourth Buk was destroyed recently.” https://x.com/NOELreports/status/1792940977604182328
“Recent work of the 79th Air Assault Brigade working in the Novomykhailiivka direction. Among Russian losses are 4 tanks, 6 BMP's, 1 2S1 Gvozdika, 1 Tor AD and 1 OSA air defense systems, 1 ZALA UAV, 2 LANCET kamikaze drones and roughly 100 men personnel.” https://x.com/NOELreports/status/1792988669533421995
“The occupiers continue to use scorched earth tactics in Donbas In the video, you can see Russian shelling in the area of the village of Bilogorivka, Luhansk region, using a large number of Grad anti-aircraft missiles and Solntsepok rocket launchers (you can see the level of intensity yourself)” https://x.com/small10space/status/1792896188942815660
“Horrible footage from Vovchansk, which is being attacked heavily by Russian forces. Killed civilians can be seen lying in the streets, fires and smoke everywhere and most houses are either badly damaged or destroyed.” https://x.com/NOELreports/status/1793224485849502006
“A russian propagandist Sladkov proudly posts a video of russian soldiers in Robotyne sitting under a tank, which they call their “position.” They need masks to help with corpse poison from their comrades.” https://x.com/wartranslated/status/1793198943561478298
10 years ago, Russian tanks entered Luhansk. https://x.com/phqqqq/status/1792839270752731540


On Politics 4: The problem with committing yourself to something, saying that it "must and will" happen is that it makes your opponents wonder what will happen to you if you, let's say, fail to achieve it. Catastrophically.
https://www.csis.org/analysis/first-battle-next-war-wargaming-chinese-invasion-taiwan