AI:
Transformers for amplitudes, a first step towards using symbolic language models in theoretical physics https://iopscience.iop.org/article/10.1088/2632-2153/ad743e
A learned simulator for general purpose robotics. https://www.1x.tech/discover/1x-world-model
“In this paper, we introduce AgentTorch -- a framework that scales ABMs to millions of agents while capturing high-resolution agent behavior using LLMs…Using the COVID-19 pandemic as a case study, we demonstrate how AgentTorch can simulate 8.4 million agents representing New York City, capturing the impact of isolation and employment behavior on health and economic outcomes.” https://arxiv.org/abs/2409.10568
Jensen Huang says technology has now reached a positive feedback loop where AI is designing new AI and is now advancing at the pace of "Moore's Law squared", meaning that the progress we will see in the next year or two will be "spectacular and surprising" https://www.youtube.com/watch?v=kfe3ajUYSdc
OpenAI's Noam Brown says that while AI model performance scales roughly equivalently with more training or inference compute, the cost of inference is on the order of 100 billion times cheaper https://www.youtube.com/watch?v=eaAonE58sLU
Former OpenAI board member Helen Toner tells a US Senate Hearing on AI that human-level AI or artificial general intelligence could be as little as 1-3 years away and there is a disconnect with how prepared the public is for this https://www.youtube.com/live/WVU7Awba3VM?si=CWKGFxZb40nyI-0M&t=964
A breakthrough in wildfire detection: How a new constellation of satellites can detect smaller wildfires earlier https://blog.google/outreach-initiatives/sustainability/google-ai-wildfire-detection/
“About 27% of firms using AI report replacing worker tasks, but only about 5% experience employment change due to AI use. These rates are expected to increase to nearly 35% and 12%, respectively, in the near future.” https://www.sciencedirect.com/science/article/abs/pii/S0165176524004555
“Google’s NotebookLM is crazy good. My mind is blown by the accuracy of the summary, the human voices, the banter, and the conversation.” https://x.com/rohitdotmittal/status/1836020602949304809
Towards a universal scientific data importer https://www.sphinxbio.com/post/universal-scientific-data-import
“Otto lets you use AI agents to help enrich lists, research companies, or read hundreds of documents in minutes, all through a native table interface.” https://ottogrid.ai/
“Is OpenAI's o1 a good calculator? We tested it on up to 20x20 multiplication—o1 solves up to 9x9 multiplication with decent accuracy, while gpt-4o struggles beyond 4x4. For context, this task is solvable by a small LM using implicit CoT with stepwise internalization. Interestingly, the number of private reasoning tokens grows sublinearly with problem size…” https://x.com/yuntiandeng/status/1836114401213989366
Nvidia presents NVLM: Open Frontier-Class Multimodal LLMs https://arxiv.org/abs/2409.11402
“It’s not that AIs will do something scary and then we ignore it. It’s that nothing will ever seem scary after a real AI does it.” https://www.astralcodexten.com/p/sakana-strawberry-and-scary-ai
The Future is Now:
Imagine reading the following in 2014: "We’ve unlocked Llama's inner workings, enabling direct modification of its internal concepts (or 'features'). By adjusting these features, we can guide Llama to claim consciousness." Fast-forward to 2024, and this is now the realm of practical research.
When I was young, the 2020s seemed like a far-off era of science fiction. Now, as I live through it, it actually *feels* like I'm living in a science fiction story!
In 2024, we have:
1. Chatbots capable of emotive conversation on topics as complex as economics and the time complexity of algorithms. ✅
2. Artificial intelligence that can generate code and solve advanced mathematical problems with increasing proficiency. ✅
3. Machine learning techniques that are helping scientists predict protein structures and design new computer chips, creating a positive feedback loop for innovation. ✅
4. Simulations of hundreds of autonomous agents collaborating in virtual worlds, with emergent economy, culture, religion, and government. ✅
5. AI that can generate art in various forms—poetry, paintings, music, and even videos. ✅
6. Robots that can jump, dance, play table tennis, and follow natural language commands in the real world. ✅
7. Speech transcription and translation technologies that are approaching human-level accuracy and robustness. ✅
8. Reusable rockets that land themselves and private spaceflight companies offering services affordable enough to create a commercial satellite internet infrastructure. ✅
9. Affordable whole-genome sequencing kits, empowering hobbyists to conduct genetic research on group populations. ✅
10. Brain-computer interfaces that let people control computers and even play games with their thoughts alone. ✅
11. Virtual and augmented reality devices that are portable, fun, and increasingly affordable. ✅
12. Portable supercomputers—smartphones—that offer instant access to the world’s knowledge, provide real-time location tracking, detailed global maps, and seamless video communication with anyone, anywhere. ✅
13. Millions of 3D-printed drones being deployed in large-scale warfare ✅
14. Flying cars. ❌
Compute:
Distributed computing with bacteria reaches a new level 👉🏻 Multicellular artificial neural network-type architectures demonstrate computational problem-solving. Mixtures of engineered bacteria were able to: Identify if a number is prime; Check if a letter in a string is a vowel; and Determine the max number of pieces of a pie obtained from n straight cuts. Answers are printed by expressing fluorescent proteins in different patterns. https://www.nature.com/articles/s41589-024-01711-4
14-bit breakthrough: Brain-like device hits massive 4.1 tera-operations per second/watt https://www.ul.ie/news/ul-researchers-discover-building-blocks-that-could-revolutionise-computing
Intel Solidifies $3.5 Billion Deal to Make Chips for Military https://www.bloomberg.com/news/articles/2024-09-13/intel-solidifies-3-5-billion-deal-to-make-chips-for-us-military [no paywall: https://archive.is/K9BuZ]
Miscellaneous:
Differentiable simulation enables large-scale training of detailed biophysical models of neural dynamics https://www.biorxiv.org/content/10.1101/2024.08.21.608979v1
Logistic regression is not fucked https://web.archive.org/web/20180319133405/http://jakewestfall.org/blog/index.php/2018/03/12/logistic-regression-is-not-fucked/
Is there are complete, consistent, physical systems that exhibit symmetry breaking at arbitrarily high temperatures? https://arxiv.org/abs/2409.10606
New PIXHELL Attack Exploits LCD Screen Noise to Exfiltrate Data from Air-Gapped Computers https://thehackernews.com/2024/09/new-pixhell-attack-exploits-screen.html
How to access archived web pages in Google https://www.tomsguide.com/computing/social-media/how-to-access-archived-web-pages-in-google
Politics:
Walkie-talkie explosions reported in Lebanon after deadly pager attack https://www.axios.com/2024/09/18/israel-detonates-hezbollah-walkie-talkies-second-wave-after-pager-attack
Microsoft: Russian election interference efforts focus on the Harris-Walz campaign https://blogs.microsoft.com/on-the-issues/2024/09/17/russian-election-interference-efforts-focus-on-the-harris-walz-campaign/
We're witnessing Georgia fall into Russia's hands in real time. The country that bravely resisted the 2008 Russian invasion now has its top leaders saying Georgia should apologize to Russia for its "aggression." https://x.com/Mylovanov/status/1836119163129008421
Ukraine:
In the Tver region of Russia, the destruction of an ammunition depot by Ukrainian drones was so large that it was registered as an earthquake of magnitude 2.7. Spectacular footage shows something almost resembling a nuclear explosion.
The depot contained up to 30,000 tons of various ammunition. It is claimed that missiles for "Grad" and S-400, as well as "Iskanders" were stored there. North Korean KN23s were also allegedly being assembled there. The depot was built to withstand nuclear attacks.
The governor of the region in Russia where the ammunition depot was located tells the press that all the drones were shot down and the fires were caused only by debris. At the same time, you can still hear the explosions in the background.
Ukrainska Pravda reports that the attack on warehouse in Toropets was conducted with UAVs from Ukrainian SBU, GUR, and SSO. New Voice of Ukraine reported that more than 100 UAVs were used in the attack.
More about Ukraine:
A strike, possibly by a Ukrainian Palianytsia rocket drone, hit the Russian base of the 856th self-propelled artillery regiment (unit 23857) of the 144th Guards Motorized Rifle Division. [note that the description wrongly states it is the ammunition depot] https://x.com/wartranslated/status/1836396708235768135
Navy of the Armed Forces of Ukraine report about the destruction of the Russian warehouses with ammunition near Mariupol as a result of a missile attack. Which means Neptune missile was most likely used. https://x.com/bayraktar_1love/status/1836075634591674371
225th Brigade showed footage of a border breakthrough with Russia in the direction of Hlushkove a few days ago. https://x.com/wartranslated/status/1836379853311926644
Russia's Remaining Towed Artillery - Approaching Crisis Point https://www.youtube.com/watch?v=eVKsoUCiGYc
“The enemy column moving towards our positions was spotted in time by the pilots of the Third Assault Brigade. The brigade units immediately joined the liquidation.” https://x.com/BBS3AB/status/1836325727618154915
“There are brothers lying everywhere, there is simply not a single living place anywhere!” https://t.me/ButusovPlus/13683
Thanks for highlighting NotebookLM. Excuse this self-plug of my own experience with it, asking it to make a podcast off from my first eleven Substack essays and stories.
https://mflood.substack.com/p/ai-podcast-discusses-my-essays