Unlike artificial language models, which process long texts as a whole, the human brain creates a “summary” while reading, helping it understand what comes next.
In recent years, large language models (LLMs) like ChatGPT and Bard have revolutionized AI-driven text processing, enabling machines to generate text, translate languages, and analyze sentiment. These models are inspired by the human brain, but key differences remain.
A new Technion-Israel Institute of Technology study, published in Nature Communications, explores these differences by examining how the brain processes spoken texts. The research, led by Prof. Roi Reichart and Dr. Refael Tikochinski from the Faculty of Data and Decision Sciences. It was conducted as part of Dr. Tikochinski’s Ph.D., co-supervised by Prof. Reichart at Technion and Prof. Uri Hasson at Princeton University.
Researchers have developed small robots that can work together as a collective that changes shape and even shifts between solid and “fluid-like” states — a concept that should be familiar to anyone still haunted by nightmares of the T-1000 robotic assassin from “Terminator 2.”
A team led by Matthew Devlin of UC Santa Barbara described this work in a paper recently published in Science, writing that the vision of “cohesive collectives of robotic units that can arrange into virtually any form with any physical properties … has long intrigued both science and fiction.”
Otger Campàs, a professor at Max Planck Institute of Molecular Biology and Genetics, told Ars Technica that the team was inspired by tissues in embryos to try and design robots with similar capabilities. These robots have motorized gears that allow them to move around within the collective, magnets so they can stay attached, and photodetectors that allow them to receive instructions from a flashlight with a polarization filter.
Well, now the Ultra is officially been released A handful of Chinese media drivers have finally gotten behind the wheel for a review—both in the context of on-the-road driving and hammering it in more aggressive circumstances. Haoran Zhou, the former car PR person and F1 reporter, did a lead-follow of the SU7 Ultra on track.
I have to note that this is technically a step down from the full-race-ready track-prepped version that Xiaomi sent around the Nürburgring. The two cars still have the same 1,526 horsepower, but the lap-setting version has essentially a full carbon-fiber body, complete with huge brake ducts right into the side of the car. This version uses mostly the body of the standard SU7, although it does have a new aluminum hood.
Because of this, the SU7 Ultra is still as fully featured as the standard car. Zhou spent half of the video using Xiaomi’s driver assistance features. It appears to work as well as the standard SU7, but Zhou did remark that it was a little surreal to have a 1,500-horsepower car do some sort of autonomous driving. “I’m trying my best to find a positive use case for it,” he said, theorizing that these features would save wear and tear on the vehicle itself between track day use. “No normal human being would be driving like this in an SU7 Ultra,” he said.
We speak with Sakana AI, who are building nature-inspired methods that could fundamentally transform how we develop AI systems.
The guests include Chris Lu, a researcher who recently completed his DPhil at Oxford University under Prof. Jakob Foerster’s supervision, where he focused on meta-learning and multi-agent systems. Chris is the first author of the DiscoPOP paper, which demonstrates how language models can discover and design better training algorithms. Also joining is Robert Tjarko Lange, a founding member of Sakana AI who specializes in evolutionary algorithms and large language models. Robert leads research at the intersection of evolutionary computation and foundation models, and is completing his PhD at TU Berlin on evolutionary meta-learning. The discussion also features Cong Lu, currently a Research Scientist at Google DeepMind’s Open-Endedness team, who previously helped develop The AI Scientist and Intelligent Go-Explore.
SPONSOR MESSAGES: *** CentML offers competitive pricing for GenAI model deployment, with flexible options to suit a wide range of models, from small to large-scale deployments. Check out their super fast DeepSeek R1 hosting! https://centml.ai/pricing/
Tufa AI Labs is a brand new research lab in Zurich started by Benjamin Crouzier focussed on o-series style reasoning and AGI. They are hiring a Chief Engineer and ML engineers. Events in Zurich.
Rather than simply scaling up models with more parameters and data, they’re drawing inspiration from biological evolution to create more efficient and creative AI systems. The team explains how their Tokyo-based startup, founded in 2023 with $30 million in funding, aims to harness principles like natural selection and emergence to develop next-generation AI.
Summary: Concerns over potential negative impacts of AI have dominated headlines, particularly regarding its threat to employment. However, a closer examination reveals AI’s immense potential to revolutionize equal and high quality access to necessities such as education and healthcare, particularly in regions with limited access to resources. From India’s agricultural advancements to Kenya’s educational support, AI initiatives are already transforming lives and addressing societal needs.
The latest technology panic is over artificial intelligence (AI). The media is focused on the negatives of AI, making many assumptions about how AI will doom us all. One concern is that AI tools will replace workers and cause mass unemployment. This is likely overblown—although some jobs will be lost to AI, if history is any guide, new jobs will be created. Furthermore, AI’s ability to replace skilled labor is also one of its greatest potential benefits.
Think of all the regions of the world where children lack access to education, where schoolteachers are scarce and opportunities for adult learning are scant.
Blue ghost’s amazing view of the moon from 62 miles up.
IF THERE ARE COPYRIGHT ISSUES, PLEASE CONTACT: [email protected]. COPYRIGHT DISCLAIMER 7EVEN don’t fully own the material compiled in this video. It belongs to individuals or organizations that deserve respect.
We use under: Copyright disclaimer section 107 of the Copyright Act 1976. “fair use” is allowed for purposes such as criticism, comment, news reporting, teaching, scholarships and research.
0:00 Intro. 0:21 Rufo Guerreschi. 0:28 Contents. 0:41 Part 1: Why we have a governance problem. 1:18 From e-democracy to cybersecurity. 2:42 Snowden showed that international standards were needed. 3:55 Taking the needs of intelligence agencies into account. 4:24 ChatGPT was a wake up moment for privacy. 5:08 Living in Geneva to interface with states. 5:57 Decision making is high up in government. 6:26 Coalition for a Baruch plan for AI 7:12 Parallels to organizations to manage nuclear safety. 8:11 Hidden coordination between intelligence agencies. 8:57 Intergovernmental treaties are not tight. 10:19 The original Baruch plan in 1946 11:28 Why the original Baruch plan did not succeed. 12:27 We almost had a different international structure. 12:54 A global monopoly on violence. 14:04 Could expand to other weapons. 14:39 AI is a second opportunity for global governance. 15:19 After Soviet tests, there was no secret to keep. 16:22 Proliferation risk of AI tech is much greater? 17:44 Scale and timeline of AI risk. 19:04 Capabilities of security agencies. 20:02 Internal capabilities of leading AI labs. 20:58 Governments care about impactful technologies. 22:06 Government compute, risk, other capabilities. 23:05 Are domestic labs outside their jurisdiction? 23:41 What are the timelines where change is required? 24:54 Scientists, Musk, Amodei. 26:24 Recursive self improvement and loss of control. 27:22 A grand gamble, the rosy perspective of CEOs. 28:20 CEOs can’t really say anything else. 28:59 Altman, Trump, Softbank pursuing superintelligence. 30:01 Superintelligence is clearly defined by Nick Bostrom. 30:52 Explain to people what “superintelligence” means. 31:32 Jobs created by Stargate project? 32:14 Will centralize power. 33:33 Sharing of the benefits needs to be ensured. 34:26 We are running out of time. 35:27 Conditional treaty idea. 36:34 Part 2: We can do this without a global dictatorship. 36:44 Dictatorship concerns are very reasonable. 37:19 Global power is already highly concentrated. 38:13 We are already in a surveillance world. 39:18 Affects influential people especially. 40:13 Surveillance is largely unaccountable. 41:35 Why did this machinery of surveillance evolve? 42:34 Shadow activities. 43:37 Choice of safety vs liberty (privacy) 44:26 How can this dichotomy be rephrased? 45:23 Revisit supply chains and lawful access. 46:37 Why the government broke all security at all levels. 47:17 The encryption wars and export controls. 48:16 Front door mechanism replaced by back door. 49:21 The world we could live in. 50:03 What would responding to requests look like? 50:50 Apple may be leaving “bug doors” intentionally. 52:23 Apple under same constraints as government. 52:51 There are backdoors everywhere. 53:45 China and the US need to both trust AI tech. 55:10 Technical debt of past unsolved problems. 55:53 Actually a governance debt (social-technical) 56:38 Provably safe or guaranteed safe AI 57:19 Requirement: Governance plus lawful access. 58:46 Tor, Signal, etc are often wishful thinking. 59:26 Can restructure incentives. 59:51 Restrict proliferation without dragnet? 1:00:36 Physical plus focused surveillance. 1:02:21 Dragnet surveillance since the telegraph. 1:03:07 We have to build a digital dog. 1:04:14 The dream of cyber libertarians. 1:04:54 Is the government out to get you? 1:05:55 Targeted surveillance is more important. 1:06:57 A proper warrant process leveraging citizens. 1:08:43 Just like procedures for elections. 1:09:41 Use democratic system during chip fabrication. 1:10:49 How democracy can help with technical challenges. 1:11:31 Current world: anarchy between countries. 1:12:25 Only those with the most guns and money rule. 1:13:19 Everyone needing to spend a lot on military. 1:14:04 AI also engages states in a race. 1:15:16 Anarchy is not a given: US example. 1:16:05 The forming of the United States. 1:17:24 This federacy model could apply to AI 1:18:03 Same idea was even proposed by Sam Altman. 1:18:54 How can we maximize the chances of success? 1:19:46 Part 3: How to actually form international treaties. 1:20:09 Calling for a world government scares people. 1:21:17 Genuine risk of global dictatorship. 1:21:45 We need a world /federal/ democratic government. 1:23:02 Why people are not outspoken. 1:24:12 Isn’t it hard to get everyone on one page? 1:25:20 Moving from anarchy to a social contract. 1:26:11 Many states have very little sovereignty. 1:26:53 Different religions didn’t prevent common ground. 1:28:16 China and US political systems similar. 1:30:14 Coming together, values could be better. 1:31:47 Critical mass of states. 1:32:19 The Philadelphia convention example. 1:32:44 Start with say seven states. 1:33:48 Date of the US constitutional convention. 1:34:42 US and China both invited but only together. 1:35:43 Funding will make a big difference. 1:38:36 Lobbying to US and China. 1:38:49 Conclusion. 1:39:33 Outro
