Links for 2024-08-08

AI:

1. Empirical data on how useful AI agents are currently compared to humans: They can't do everything, but they can do a decent chunk of what humans can do, and they can do it significantly cheaper/faster. For example, the Claude 3.5 Sonnet agent fixed bugs in an object-relational mapping library using approximately 382,000 tokens (costing less than $2), whereas our human baseline took over two hours. https://metr.org/blog/2024-08-06-update-on-evaluations/

2. “Can LLMs predict results of social science experiments? Across 70 studies, we find striking alignment (r = .85) between simulated and observed effects. Overall our results show high accuracy of LLM-derived predictions for experiments with human participants, generally greater accuracy than samples of lay and expert humans.” https://docsend.com/view/qeeccuggec56k9hd

3. “LLaVA-OneVision allows strong transfer learning across different modalities/scenarios, yielding new emerging capabilities. In particular, strong video understanding and cross-scenario capabilities are demonstrated through task transfer from images to videos.” https://arxiv.org/abs/2408.03326

4. Key-Point-Driven Mathematical Reasoning Distillation of Large Language Model https://arxiv.org/abs/2407.10167

5. Benchmarking LLMs for Optimization Modeling and Enhancing Reasoning via Reverse Socratic Synthesis https://arxiv.org/abs/2407.09887

6. "Transformers are Universal In-context Learners": in this paper, we show that deep transformers with a fixed embedding dimension are universal approximators for an arbitrarily large number of tokens. https://arxiv.org/abs/2408.01367

7. “How can we prevent LLM safeguards from being simply removed with a few steps of fine-tuning? We show it's surprisingly possible to make progress on creating safeguards that are tamper-resistant, reducing malicious use risks of open-weight models.” https://arxiv.org/abs/2408.00761

8. Diffusion Models as Data Mining Tools https://arxiv.org/abs/2408.02752

9. Hierarchical Conditioning of Diffusion Models Using Tree-of-Life for Studying Species Evolution https://arxiv.org/abs/2408.00160

10. Google announces Scaling LLM Test-Time Compute Optimally can be More Effective than Scaling Model Parameters — Test-time compute can be used to outperform a 14× larger model https://arxiv.org/abs/2408.03314

11. [Open Source] Unitree First View Teleoperation for Humanoid Robots to advance the convenience of data collection for humanoid robots https://github.com/unitreerobotics/avp_teleoperate

12. A New Study Says AI Models Encode Language Like the Human Brain Does https://singularityhub.com/2024/08/07/a-new-study-says-ai-models-encode-language-like-the-human-brain-does/

13. A.I. ‐ Humanity's Final Invention? https://www.youtube.com/watch?v=fa8k8IQ1_X0

14. AI “godfather” Yoshua Bengio has joined a UK project to prevent AI catastrophes https://www.technologyreview.com/2024/08/07/1095879/ai-godfather-yoshua-bengio-joins-uk-project-to-prevent-ai-catastrophes/ [no paywall: https://archive.is/wcpgo]

https://x.com/Simeon_Cps/status/1821230833434554449

“Figure is giving artificial intelligence a body” — Brett Adcock [Easily one of the top 5 most important companies in the world.]

https://x.com/RichardSocher/status/1820544142256075033

Miscellaneous:

1. “We're using ultrasound to safely and non-invasively measure and modulate brain activity at high resolution” https://quintinfrerichs.xyz/nudge

2. Japanese scientists develop simplified EUV scanner that can make production of chips considerably cheaper https://www.tomshardware.com/tech-industry/japanese-scientists-develop-simplified-euv-scanner-that-can-make-production-of-chips-considerably-cheaper

3. Tiny arm bone belonged to smallest ancient human ever found https://www.nature.com/articles/d41586-024-02548-6

4. “The implications for life in the liquid water oceans, under the surface of icy moons, are obvious, and enormous.  So I'm going to predict now, with medium confidence (and a couple of caveats, to follow) that we may well ultimately discover similar polymetallic nodules, producing oxygen through similar chemical processes, on the warm seafloors of the liquid water oceans under the frozen crusts of icy moons.” https://theeggandtherock.com/p/the-deep-ocean-floor-is-covered-in

5. Feasibility of keeping Mars warm with nanoparticles https://www.science.org/doi/10.1126/sciadv.adn4650

6. “When that enormous magnitude-9 earthquake hit Japan in 2011, it caused waves 1.5 meters high in some lakes in NORWAY!” https://mathstodon.xyz/@johncarlosbaez/112920894947197795

Politics:

1. ‘Sky’s the limit’: Fort Stewart soldiers prepare for the modern battlefield by building small drones from scratch https://www.stripes.com/branches/army/2024-08-06/army-soldiers-building-drones-fort-stewart-14761022.html

2. What can we say about the "far right" riots? https://www.aporiamagazine.com/p/what-can-we-say-about-the-far-right

Ukraine:

1. Ukraine has launched a special military operation against Russia. Here you can see about 40-60 Russian soldiers captured by Ukrainian forces in the Kursk region of Russia. I have no idea if this is worthwhile for Ukraine, but it is certainly a massive embarrassment for Russia to be invaded by the country it tried to overrun two and a half years ago. Imagine Mexico invading the United States years after attempting regime change there. https://x.com/Danspiun/status/1821218979811119558

2. “The ongoing Ukrainian offensive in Kursk oblast has begun successfully. In less than two days, Ukraine has achieved a breakthrough, pushing at least 12 kilometers deep, through two lines of Russian fortifications.” https://x.com/emilkastehelmi/status/1821262611465564283

3. “Russia failed to identify this assault, showing a significant improvement in Ukrainian counterintelligence measures. Despite advanced ISR capabilities, Russian forces failed to interpret the concentration of Ukrainian forces as an offensive maneuver.” https://x.com/Tatarigami_UA/status/1821300140755271836

4. A new video from the area near Zelenyi Shlyakh, taken by a driver trying to leave, shows dead Russian soldiers and their rifles on the road. https://x.com/NOELreports/status/1821196943152075001

5. This is how Ukraine is now intercepting Russian reconnaissance drones at low cost. Instead of launching jets or using expensive anti-aircraft missiles, drones are shot down with drones. https://x.com/bayraktar_1love/status/1821113003724394600

6. Ukrainian FPV drones vs Russian helicopters. https://x.com/wartranslated/status/1821108479383294041

7. “A video from Ukrainian SOF unit has appeared in the media, as said, it shows how Russians on the Kherson direction began to throw out the bodies of their comrades after noticing a Ukrainian drone.” https://x.com/bayraktar_1love/status/1821170469497688120