Links for 2024-06-25

AI:

1. Neural Network Potentials for Enabling Advanced Small-Molecule Drug Discovery and Generative Design https://www.liebertpub.com/doi/10.1089/genbio.2024.0011

2. “ESM3 is a generative language model for programming biology. In experiments, we found ESM3 can simulate 500M years of evolution to generate new fluorescent proteins.” https://www.evolutionaryscale.ai/blog/esm3-release

3. RL on Incorrect Synthetic Data Scales the Efficiency of LLM Math Reasoning by Eight-Fold https://arxiv.org/abs/2406.14532

4. LongRAG: Enhancing Retrieval-Augmented Generation with Long-context LLMs https://arxiv.org/abs/2406.15319

5. Microsoft unveiled Florence-2, a new generalist AI vision model. It uses prompt-based instructions and excels in tasks like captioning, object detection, OCR, and segmentation. Most notably, it's 100x smaller than similar models and open-sourced. https://huggingface.co/microsoft/Florence-2-large

6. Microsoft AI CEO Mustafa Suleyman says it won't be until GPT-6 in 2 years time that AI models will be able to follow instructions and take consistent action https://youtu.be/cNzRviY4Ei8?si=2ELY9k7ozbs7SQc4&t=1331

7. Frontier AI systems could pose increasing risks to public safety and security. But what level of risk is acceptable? https://arxiv.org/abs/2406.14713

8. LLMs may be fundamentally incapable of fully general reasoning, and if so, short timelines are less plausible. https://www.lesswrong.com/posts/k38sJNLk7YbJA72ST/llm-generality-is-a-timeline-crux

9. “In 2021, researchers were asked how smart AI would be in 2022. They predicted 12% pass-rate on the MATH dataset; AI outperformed significantly, scoring 50%. Now it's at 90%.” https://x.com/Jake_Schwartz/status/1804974011434344631

Neuroscience:

1. New Computational Model of Real Neurons Could Lead to Better AI — “The new model developed by researchers at the Flatiron Institute proposes that biological neurons have more control over their surroundings than previously thought, something that could be replicated in the artificial neural networks used in machine learning.” https://www.simonsfoundation.org/2024/06/24/new-computational-model-of-real-neurons-could-lead-to-better-ai/

2. Brain-Like Language Processing via a Shallow Untrained Multihead Attention Network https://arxiv.org/abs/2406.15109

3. Reconstructing seen movies from mouse visual cortex🐭🧠https://www.biorxiv.org/content/10.1101/2024.06.19.599691v1 (video: https://www.youtube.com/watch?v=TA6Oi5NfuMs)

4. A novel treatment has been proven to effectively treat cognitive decline in mice with Alzheimer’s disease. https://www.oist.jp/news-center/news/2024/6/20/damage-synapses-caused-alzheimers-disease-reversed

5. The Path Towards Mammalian Whole-Brain Circuit Mapping https://www.youtube.com/watch?v=QZhvsJw7P_Q

6. Neurotechnology: Past, Present, and Future https://www.youtube.com/watch?v=sNP4_3cbLxA

Technology:

1. Northrop Grumman's solution to drones: a MACE sensor tower, a lightweight 30mm XM914 chaingun and XM1211 proximity-fuzed HE rounds, all working together off the back of two trucks. https://www.youtube.com/watch?v=rr7ym1zkda8

2. Terahertz Waves Supercharged: A Breakthrough With Magnetic Materials https://www.wpi-aimr.tohoku.ac.jp/en/achievements/press/2024/20240610_001808.html

Miscellaneous:

1. NASA, Global Astronomers Await Rare Nova Explosion – so bright it will be visible on Earth with the naked eye – is poised to occur. https://www.nasa.gov/centers-and-facilities/marshall/nasa-global-astronomers-await-rare-nova-explosion/

2. Mistakes people make when thinking about units https://www.lesswrong.com/posts/5vfSNLb92eyXKkQax/mistakes-people-make-when-thinking-about-units

Firework jellyfish: https://en.m.wikipedia.org/wiki/Halitrephes

Links for 2024-06-28

AI:

1. Finding GPT-4’s mistakes with GPT-4 — CriticGPT, a model based on GPT-4, writes critiques of ChatGPT responses to help human trainers spot mistakes during RLHF https://openai.com/index/finding-gpt4s-mistakes-with-gpt-4/

2. Meta Large Language Model Compiler: Foundation Models of Compiler Optimization https://ai.meta.com/research/publications/meta-large-language-model-compiler-foundation-models-of-compiler-optimization/

3. “LangGraph helps you build reliable agents that actually work. Today, we've launched LangGraph Cloud, our new infrastructure to run fault-tolerant LangGraph agents at scale.” https://www.youtube.com/watch?v=l4sMKF1dTDM

4. LLM-based validators that *automatically improve* in response to human feedback. https://www.youtube.com/watch?v=3gCTa0Li4ew

5. ICAL: Continual Learning of Multimodal Agents by Transforming Trajectories into Actionable Insights https://ical-learning.github.io/

6. Dreamitate: Real-World Visuomotor Policy Learning via Video Generation https://dreamitate.cs.columbia.edu/

7. Flow of Reasoning: Efficient Training of LLM Policy with Divergent Thinking https://yu-fangxu.github.io/FoR.github.io/

8. 600% Boost: Scientists Develop Game-Changing AI Chip With Impressive Energy Efficiency https://today.oregonstate.edu/news/new-computer-chips-show-promise-reducing-energy-footprint-artificial-intelligence

9. Study Finds Self-Driving Cars Are Actually Safer Than Humans in Many (But Not All) Situations https://singularityhub.com/2024/06/24/study-finds-self-driving-cars-are-actually-safer-than-humans-in-most-situations/

10. The A.I. Boom Has an Unlikely Early Winner: Wonky Consultants https://www.nytimes.com/2024/06/26/technology/ai-consultants.html [no paywall: https://archive.is/BAjal]

11. “We found that 94% of our AI submissions were undetected. The grades awarded to our AI submissions were on average half a grade boundary higher than that achieved by real students.” https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0305354

12. Interview with Carl Shulman: why economists get AGI mostly wrong; how output might double in 3 months: how incomes could grow 100x or more https://80000hours.org/podcast/episodes/carl-shulman-economy-agi/

Protegenneurotechnoengineering:

1. Protein design for growing semiconductors https://www.biorxiv.org/content/10.1101/2024.06.24.600095v1

2. Programmable RNA-guided enzymes for next-generation genome editing. https://arcinstitute.org/news/blog/bridge (Summary by Claude: Genome Design: The Bridge to Our Biological Future https://x.com/patrickc/status/1805996143228375263)

3. “What's actually impressive is the all-in-one approach that ESM3 took — combining structure, sequence, and function in a model with 98 billion parameters.” https://www.abhishaike.com/p/a-primer-on-gfp-and-esmgfp

4. Rat Neurons Repair Mouse Brains That Lack a Sense of Smell https://sitn.hms.harvard.edu/flash/2024/researchers-create-interspecies-brain-chimeras-from-mice-and-rats/

5. Detecting Genetically Engineered Viruses With Metagenomic Sequencing https://www.lesswrong.com/posts/iaPhjYhhp7PP6BWp9/detecting-genetically-engineered-viruses-with-metagenomic

6. Scientists use computational modeling to guide a difficult chemical synthesis https://news.mit.edu/2024/scientists-use-computational-modeling-for-difficult-chemical-synthesis-0627

7. Incredible New Technique Measures Forces As Small as a Virus With Unprecedented Precision https://phys.org/news/2024-06-advances-nanoscale-doors-unprecedented-biological.html

Miscellaneous:

1. Ancient Sanskrit carved in stone in Egypt! Over 2000 years old! https://www.smithsonianmag.com/history/hidden-ancient-egyptian-port-reveals-180984485/

2. JWST’s ‘Little Red Dots’ Offer Astronomers the Universe’s Weirdest Puzzle https://www.scientificamerican.com/article/jwsts-little-red-dots-offer-astronomers-the-universes-weirdest-puzzle/ [read without registration: https://archive.is/GHAi0]

Dario Amodei (Anthropic/Claude) says AI models "better than most humans at most things" are 1-3 years away

Just took this smartphone shot of noctilucent clouds: https://en.m.wikipedia.org/wiki/Noctilucent_cloud

Figure + BMW Group's Spartanburg Plant

→ Fully autonomous
→ AI-driven vision model
→ Neural Networks for all grasps

https://www.prnewswire.com/news-releases/figure-announces-commercial-agreement-with-bmw-manufacturing-to-bring-general-purpose-robots-into-automotive-production-302036263.html

Gen-3 Alpha Text to Video is now available to everyone.

A new frontier for high-fidelity, fast and controllable video generation.

Try it now at runwayml.com

Links for 2024-07-02

AI:

1. Scaling Synthetic Data Creation with 1,000,000,000 Personas — Massive gains on MATH: 49.6 ->64.9 https://github.com/tencent-ailab/persona-hub

2. “To build the next generation of intelligent agents, developing efficient world models is essential. We introduce Δ-IRIS, an agent that learns behaviors by imagining millions of trajectories in its world model.” https://github.com/vmicheli/delta-iris

3. GraphReader: Building Graph-based Agent to Enhance Long-Context Abilities of Large Language Models https://arxiv.org/abs/2406.14550v1

4. Babies use ‘helpless’ infant period to learn powerful foundation models, just like ChatGPT https://www.tcd.ie/news_events/articles/2024/infant-helplessness/

5. LLaRA: Supercharging Robot Learning Data for Vision-Language Policy https://arxiv.org/abs/2406.20095

6. PoliFormer: On-Policy RL with Transformers Results in Masterful Navigators https://poliformer.allen.ai/

7. Meta 3D Gen: A new system for end-to-end generation of 3D assets from text in <1min. https://ai.meta.com/research/publications/meta-3d-gen/

8. AI-assisted development is now the norm - 78% of survey respondents currently use AI in software development or plan to in the next two years, up from 64% in 2023. https://www.zdnet.com/article/ai-accelerates-software-development-to-breakneck-speeds-but-measuring-that-is-tricky/

Technology:

1. A new neuroprosthetic interface developed by researchers in the K. Lisa Yang Center for Bionics is driven by the nervous system and helps people with amputation walk naturally. https://mcgovern.mit.edu/2024/07/01/a-prosthesis-driven-by-the-nervous-system-helps-people-with-amputation-walk-naturally/

2. Electricity-free mechanical computer goes beyond binary data storage https://www.science.org/doi/10.1126/sciadv.ado6476

3. “This electric car battery takes less than 5 minutes to charge” https://edition.cnn.com/2024/07/01/cars/electric-car-battery-charge/index.html

4. Breakthrough Computational Warp Drive Design Without Needing Negative Energy https://www.nextbigfuture.com/2024/06/breakthrough-computational-warp-drive-design-without-needing-negative-energy.html

Archeology:

1. A lost civilization’s partial alphabet was discovered in a social media post https://www.sciencenews.org/article/lost-civilization-alphabet-social-media

2. Archaeological evidence of an ethnographically documented Australian Aboriginal ritual dated to the last ice age https://www.nature.com/articles/s41562-024-01912-w

Politics:

1. "South Korea is trending toward a fertility rate of just 0.68 births per woman in 2024 … was at 1.24 … in 2015, … Chile is projected to have a TFR of just 0.88 … in 2024, … 1.78 just in 2015. Turkey’s fertility was just 1.51 in 2023, having been 2.16 in 2015." https://x.com/MoreBirths/status/1807509085732106420

2. “The Indiana pi bill was bill 246 of the 1897 sitting of the Indiana General Assembly, one of the most notorious attempts to establish mathematical truth by legislative fiat.” https://en.wikipedia.org/wiki/Indiana_pi_bill

Real video of a Falcon 9 launch.

"Falcon launched 67 missions in the first 6 months of 2024, delivering nearly 900 metric tons to orbit so far this year"

Links for 2024-07-07

AI:

1. AI Mathematical Olympiad: It appears that the winning program correctly answered 29/50 of the private test questions. — “Maybe what's even more impressive about this competition, beside the level of math these models are already capable of is how ressource contraint the participants were actually, having to run inference in a short amont of time on T4 which only let us imagine how powerful these models will become in the coming months.” https://x.com/Thom_Wolf/status/1809895886899585164

2. Learning Formal Mathematics From Intrinsic Motivation https://arxiv.org/abs/2407.00695

3. “This means the relationship between changes in underlying model capabilities and changes in real world impact can be unintuitive. If stepwise accuracy goes from 99% to 99.99%, a 200 step task goes from failing most of the time to succeeding almost always” https://x.com/RatOrthodox/status/1809055334536786130 (Paper: Rethinking AI agent benchmarking and evaluation https://www.aisnakeoil.com/p/new-paper-ai-agents-that-matter)

4. Gradually, then Suddenly: What often matters is when technologies pass certain thresholds of capability. https://www.oneusefulthing.org/p/gradually-then-suddenly-upon-the

5. OmniJARVIS: Unified Vision-Language-Action Tokenization Enables Open-World Instruction Following Agents https://omnijarvis.github.io/

6. Introducing ReSearch: An iterative self-reflection algorithm that enhances LLM's self-restraint abilities. Encouraging abstention when uncertain. Producing accurate, informative content when confident. https://arxiv.org/abs/2405.13022

7. Natural Language Embedded Programs for Hybrid Language Symbolic Reasoning https://arxiv.org/abs/2309.10814

8. Diffusion Forcing combines the strength of full-sequence diffusion models and next-token models, acting as either or a mix at sampling time for different applications without retraining. https://boyuan.space/diffusion-forcing/

9. Improving retrieval with LLM-as-a-judge https://blog.vespa.ai/improving-retrieval-with-llm-as-a-judge/

10. “This is an interim report on reverse-engineering Othello-GPT, an 8-layer transformer trained to take sequences of Othello moves and predict legal moves. We find evidence that Othello-GPT learns to compute the board state using many independent decision rules that are localized to small parts of the board.” https://www.lesswrong.com/posts/gcpNuEZnxAPayaKBY/othellogpt-learned-a-bag-of-heuristics-1

Engineering:

1. New Multi-Material “Laser” 3D Printer Can Create Complex Devices With Just a Single Machine https://engineering.missouri.edu/2024/no-assembly-required/

2. Desalinating Water Is Becoming “Absurdly Cheap” https://humanprogress.org/desalinating-water-is-becoming-absurdly-cheap/

3. Open-TeleVision: Teleoperation with Immersive Active Visual Feedback https://robot-tv.github.io/

4. “Britain should reclaim an area the size of Wales from Dogger Bank, the area of the North Sea where the sea is only 15-40m deep. We could do it for less than £100bn.” https://model-thinking.com/p/a-new-atlantis

Miscellaneous:

1. BB(5) is now known to equal 47176870, thanks to a collaboratively-made Coq proof that decides the halting problem for all 5-state Turing machines by case analysis of ~180 million equivalence classes, which coqc can check in ~10 hours of wall-clock time. https://www.quantamagazine.org/amateur-mathematicians-find-fifth-busy-beaver-turing-machine-20240702/

2. “Our results imply that being genetically predisposed to be smarter causes left-wing beliefs.” https://www.sciencedirect.com/science/article/abs/pii/S0160289624000254

3. “…we show that inattentionally blind participants can successfully report the location, color and shape of the stimuli they deny noticing.” https://www.biorxiv.org/content/10.1101/2024.05.18.593967v1

Harmonic is continuing to make progress toward mathematical superintelligence: https://www.harmonic.fun/news

HD Atlas Manipulates | Boston Dynamics

Links for 2024-07-11

AI:

1. AI Math Olympiad winner is now Open Source! https://huggingface.co/AI-MO/NuminaMath-7B-TIR (Demo: https://huggingface.co/spaces/AI-MO/math-olympiad-solver)

2. “Transformers can display surprising “length generalization” capabilities on many algorithmic tasks: addition, multiplication, and even in-context simulation of SGD!” https://arxiv.org/abs/2407.03310

3. Mixture of A Million Experts https://arxiv.org/abs/2407.04153

4. Internet of Agents: Weaving a Web of Heterogeneous Agents for Collaborative Intelligence https://arxiv.org/abs/2407.07061

5. “Our 1B parameter model xLAM-1B is now the best micro model for function calling, outperforming models 7x its size, including GPT-3.5 & Claude. On-device agentic AI is here.” https://apigen-pipeline.github.io/

6. Pantheon Interface: 1. A human user “thinks out loud” by typing out their thoughts one at a time. This leaves a text trace of their stream of thought. 2. AI characters (called daemons) read this trace, and interact with the user by responding asynchronously with comments and questions. https://www.lesswrong.com/posts/JHsfMWtwxBGGTmb8A/pantheon-interface

7. How Google Project Zero got 20x improvements on having models exploit buffer overflows and memory corruption https://googleprojectzero.blogspot.com/2024/06/project-naptime.html

8. “Companies spend huge amounts of money on training runs, and feel secure doing so, because they know that you get out what you put in, without surprises in either direction.” https://nostalgebraist.tumblr.com/post/741247180226052096/i-dont-think-youre-drawing-the-right-lesson-from

9. The Chinese government is going all-in on autonomous vehicles https://www.technologyreview.com/2024/07/10/1094811/chinese-government-policy-autonomous-vehicles/ [no paywall: https://archive.is/ph0q9]

AI War:

1. “The guidance system provides optical lock-on: the operator identifies the target and flags it for the autopilot while the drone is well outside jamming range. Then it can carry on through the ‘jamming bubble’.” https://www.forbes.com/sites/davidhambling/2024/07/10/destroying-russian-tanks-is-just-the-start-for-us-ai-drone-autopilot/

2. He created Oculus headsets as a teenager. Now he makes AI weapons for Ukraine https://www.npr.org/2024/07/09/nx-s1-4985981/oculus-ai-weapons-ukraine-palmer-luckey

AI Education:

1. Free book: Understanding Deep Learning https://udlbook.github.io/udlbook/

2. A visual and intuitive guide to understanding how transformers work https://jalammar.github.io/illustrated-transformer/

3. How AlphaFold3 works. A visual walkthrough. https://elanapearl.github.io/blog/2024/the-illustrated-alphafold/

Biotech:

1. Why haven't biologists cured cancer? Slow feedback loops. https://www.writingruxandrabio.com/p/why-havent-biologists-cured-cancer

2. Inside the Laboratory for Extraordinary Microbes https://press.asimov.com/articles/cultivarium

Computer Science:

1. To understand quantum computers avoid falling for overly simple explanations. https://www.quantamagazine.org/why-is-quantum-computing-so-hard-to-explain-20210608/

2. The Zombie Misconception of Theoretical Computer Science https://scottaaronson.blog/?p=8106

3. A Trustworthy, Free (Libre), Linux Capable, Self-Hosting 64bit RISC-V Computer https://x.com/karpathy/status/1811097021539045582 (Project page: https://www.contrib.andrew.cmu.edu/~somlo/BTCP/)

Astronomy:

1. Astronomers find surprising ice world in the habitable zone with JWST data https://news.umich.edu/astronomers-find-surprising-ice-world-in-the-habitable-zone-with-jwst-data/

2. Primon gas: a theoretical gas where there's one kind of particle for each prime number. https://mathstodon.xyz/@johncarlosbaez/112762809456139367

Politics:

1. How Wikipedia Admin David Gerard Launders His Grudges Into the Public Record https://www.tracingwoodgrains.com/p/reliable-sources-how-wikipedia-admin

2. History is written by the losers https://scholars-stage.org/history-is-written-by-the-losers/

"This is a real-time video of extremely precise deposition of 1 nanoliter to 1 microliter droplets inside of 96 well plates.

Equivalent to developing photolithography but for life sciences. Absolutely incredible tech tree unlock. We live in an age of miracles.

Anyone that's worked in life sciences knows - pipetting is a nightmare, and variance in reactant volumes can absolutely destroy your experiment. The ability to massively mulitplex - by an factor of 100 - 1000x, the number of experiments that can occur in a standard well plate…" (Description by Andrew Côté)

Read more: https://www.m2-automation.com/en/

Shots fired at the Trump rally in Butler, Pennsylvania.