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.

Links for 2024-07-14

AI:

1. OpenAI is working on new reasoning technology under the code name ‘Strawberry’ (formerly known as Q*) to perform long-horizon tasks. Planning ahead will enable it to navigate the internet autonomously and reliably to perform “deep research.” In an internal all-hands meeting, OpenAI showed a demo of a research project that it claimed had new human-like reasoning skills. https://www.reuters.com/technology/artificial-intelligence/openai-working-new-reasoning-technology-under-code-name-strawberry-2024-07-12/ [archived version: https://archive.is/cnCrI]

2. "Standard deep learning (is) slow and power-hungry..we introduce..an analog electronic network (which) learns tasks unachievable in linear systems, (is) robust to damage, retrainable in seconds, & performs.. in microseconds..dissipating only picojoules" https://www.pnas.org/doi/10.1073/pnas.2319718121

3. The Making of Devin [AI software agent] by Cognition AI: Scott Wu https://www.youtube.com/watch?v=T7NWjoD_OuY

4. “Many AI researchers believe that deep learning alone is not enough; there must be more than naive scaling to get to human level AGI. If you’re in this plurality, I have some questions” https://evjang.com/2024/07/11/arc.html

5. OpenDiLoCo: Enabling globally distributed AI model training. https://www.primeintellect.ai/blog/opendiloco

6. MambaVision: A Hybrid Mamba-Transformer Vision Backbone https://arxiv.org/abs/2407.08083

7. SEED-Story: Multimodal Long Story Generation with Large Language Model https://arxiv.org/abs/2407.08683

8. “Fast, robust, reactive, direct-from-sensor grasp-anything policies. RL really works, and it’s going to transform the entire robotics economy.” https://arxiv.org/abs/2407.02274

9. Machine Learning Can Predict Shooting Victimization Well Enough to Help Prevent It — “Out-of-sample accuracy is strikingly high: of the 500 people with the highest predicted risk, almost 13 percent are shot within 18 months, a rate 128 times higher than the average Chicagoan.” https://www.nber.org/papers/w30170

10. AI system achieves 96% accuracy in determining sex from dental X-rays https://www.psypost.org/ai-system-achieves-96-accuracy-in-determining-sex-from-dental-x-rays/

11. Helsing, a startup developing AI software for defense, raises €450 million to expand its presence in European nations bordering Russia https://www.bloomberg.com/news/articles/2024-07-11/defense-startup-helsing-nets-5-billion-valuation-plans-eastern-flank-expansion [no paywall: https://archive.is/Ns43P]

12. Scaling Law in Neural Data: Non-Invasive Speech Decoding with 175 Hours of EEG Data. https://arxiv.org/abs/2407.07595

13. A.I. Helped Spot a Copper Bonanza. It Could Transform More Than Mining. https://www.nytimes.com/2024/07/11/climate/kobold-zambia-copper-ai-mining.html [no paywall: https://archive.is/smmDA]

14. How AI Revolutionized Protein Science, but Didn’t End It https://www.quantamagazine.org/how-ai-revolutionized-protein-science-but-didnt-end-it-20240626/

15. A full highly accurate radiance field - where you can choose to see a 3D scene from any point of view (effectively a 5D representation) - compresses the scene to roughly the size of any single training picture. https://www.youtube.com/watch?v=CRlN-cYFxTk

16. Reasoning through arguments against taking AI safety seriously https://yoshuabengio.org/2024/07/09/reasoning-through-arguments-against-taking-ai-safety-seriously/

Miscellaneous:

1. Food without agriculture: Food from CO2, biomass and hydrocarbons to secure humanity's food supply against global catastrophe https://www.sciencedirect.com/science/article/abs/pii/S0924224424002851

2. Ice: The Penultimate Frontier — “I argue here that preventing a large iceberg from melting is absurdly cheap per unit area compared to just about any other way of making new land, and it's kind of crazy to spend money on space exploration and colonization before colonizing the oceans with floating ice-islands.” https://www.lesswrong.com/posts/gthjxPDywrMTs3p2j/ice-the-penultimate-frontier

Links for 2024-07-17

AI:

1. Trump allies draft AI order to launch ‘Manhattan Projects’ for defense to compete with China https://www.washingtonpost.com/technology/2024/07/16/trump-ai-executive-order-regulations-military [no paywall: https://archive.is/LFseW]

2. Mistral launches Mathstral 7B and Codestral Mamba 7B. On the MATH benchmark, Mathstral 7B obtains 56.6% pass@1, outperforming Minerva 540B by more than 20%. Mathstral scores 68.4% on MATH with majority voting@64, and 74.6% using a reward model. https://mistral.ai/news/mathstral/ Codestral Mamba is one of the first open source models with a Mamba 2 architecture. It is the best 7B code model available, and is trained with a context length of 256k tokens. https://mistral.ai/news/codestral-mamba/

3. SpreadsheetLLM: Encoding Spreadsheets for Large Language Models https://arxiv.org/abs/2407.09025

4. Human-like Episodic Memory for Infinite Context LLMs https://arxiv.org/abs/2407.09450

5. Generating Games Via Evolution and Language Models https://arxiv.org/abs/2407.09388

6. Foundational Autoraters: Taming Large Language Models for Better Automatic Evaluation https://arxiv.org/abs/2407.10817

7. Learning Multiple Concepts from a Single Image — Unsupervised Concept Extraction (UCE) is a new task that extracts and recreates multiple concepts from a single image without any human annotations. https://haoosz.github.io/ConceptExpress/

8. AI method radically speeds predictions of materials’ thermal properties https://news.mit.edu/2024/ai-method-radically-speeds-predictions-materials-thermal-properties-0716

9. Artificial intelligence outperforms clinical tests at predicting progress of Alzheimer’s disease https://www.cam.ac.uk/research/news/artificial-intelligence-outperforms-clinical-tests-at-predicting-progress-of-alzheimers-disease

10. Does GPT-4 have Theory of Mind? “Across the battery of theory of mind tests, we found that GPT-4 models performed at, or even sometimes above, human levels” https://www.nature.com/articles/s41562-024-01882-z

Robotics:

1. “UMI on Legs is a framework for combining real-world human demonstrations with simulation trained whole-body controllers, providing a scalable approach for manipulation skills on robot dogs with arms.” https://umi-on-legs.github.io/

2. Surgical Robot Transformer🪡: Automating delicate surgical tasks with end-to-end imitation learning. https://surgical-robot-transformer.github.io/

Biotechnology:

1. Multiplex Gene Editing: Where Are We Now? https://www.lesswrong.com/posts/oSy5vHvwSfnjmC7Tf/multiplex-gene-editing-where-are-we-now

2. Disruptive and innovative approach to drug discovery using high throughput in vivo screening. https://www.gordian.bio/blog/the-in-vivo-screening-revolution/

3. Genomic Language Models: Opportunities and Challenges https://arxiv.org/abs/2407.11435

4. Wet-lab innovations will lead the AI revolution in biology https://www.abhishaike.com/p/wet-lab-innovations-will-lead-the

Miscellaneous:

1. New quantum computer smashes 'quantum supremacy' record by a factor of 100 — and it consumes 30,000 times less power https://www.livescience.com/technology/computing/new-quantum-computer-smashes-quantum-supremacy-record-by-a-factor-of-100-and-it-consumes-30000-times-less-power

2. A new meta-analysis of group differences in measured IQs in Britain https://www.emilkirkegaard.com/p/the-ethnic-meritocracy-in-the-united

3. Study reveals how an anesthesia drug induces unconsciousness https://news.mit.edu/2024/study-reveals-how-anesthesia-drug-induces-unconsciousnes-0715

4. Simulation arguments, a research paper in philosophy. https://jc.gatspress.com/pdf/simulation_arguments_revised.pdf

South Korea:

1. More South Koreans want Seoul to have its own nuclear weapons https://www.ft.com/content/0a7b8855-5682-4fbf-be42-156811d4d578 [no paywall: https://archive.is/2v3Mb]

2. South Korea to mass produce lasers that can take out drones at $1.50 a hit https://edition.cnn.com/2024/07/11/asia/south-korea-antidrone-lasers-intl-hnk-ml/index.html

Links for 2024-07-19

AI:

1. Implicit meta-learning may lead language models to trust more reliable sources — “Our results suggest that during training, LLMs better internalize text that appears useful for predicting other text (e.g. seems reliable).” https://arxiv.org/abs/2310.15047

2. Weak-to-Strong Reasoning: A progressive learning framework that enables the strong model to autonomously refine its training data, without requiring input from either a more advanced model or human-annotated data. https://arxiv.org/abs/2407.13647

3. OpenAI: “We trained advanced language models to generate text that weaker models can easily verify, and found it also made these texts easier for human evaluation.” https://openai.com/index/prover-verifier-games-improve-legibility/

4. Towards intelligence too cheap to meter: OpenAI announced GPT-4o Mini. A powerful, lightweight, and cost-efficient model. 15 cents per million input tokens, 60 cents per million output tokens. https://openai.com/index/gpt-4o-mini-advancing-cost-efficient-intelligence/

5. “…it is possible to find multiple steering vectors in a language model that activate very similar behaviors while all being orthogonal.” https://www.lesswrong.com/posts/CbSEZSpjdpnvBcEvc/i-found-greater-than-800-orthogonal-write-code-steering

6. Goldfish: Vision-Language Understanding of Arbitrarily Long Videos https://arxiv.org/abs/2407.12679

7. “This method demonstrates significant improvements over traditional multimodal training on image-text pairs, while reducing training costs by approximately 95%.” https://arxiv.org/abs/2407.12580

8. Machine learning unlocks secrets to advanced alloys https://news.mit.edu/2024/machine-learning-unlocks-secrets-advanced-alloys-0718

9. What Could Conquer the Superweeds? Bayer and Others Turn to AI https://www.wsj.com/science/environment/super-weed-killer-ai-8105de6a [no paywall: https://archive.is/X5heZ]

10. How well can AI chatbots mimic doctors in a treatment setting? We put 5 to the test https://www.cnbc.com/2024/07/18/op-ed-how-well-can-ai-chatbots-mimic-doctors.html

11. Claude 3.5 system prompt for coding https://www.reddit.com/r/ClaudeAI/comments/1dwra38/sonnet_35_for_coding_system_prompt/

12. “Samsung’s new image-generating AI tool is a little too good” https://www.theverge.com/2024/7/17/24199005/samsung-galaxy-ai-z-fold-6-sketch-to-image

13. JPMorgan CEO Jamie Dimon says he’ll add thousands of jobs focused on AI in the next couple of years. https://www.businessinsider.in/artificial-intelligence/news/jpmorgan-ceo-jamie-dimon-says-hell-add-thousands-of-jobs-focused-on-ai-in-the-next-couple-of-years/articleshow/111823636.cms

14. Meta won't offer future multimodal AI models in EU, citing regulatory uncertainty. https://www.axios.com/2024/07/17/meta-future-multimodal-ai-models-eu

15. “Donald Trump says America is on the cusp of a new golden age which will require tremendous energy investments to power AI” https://x.com/tsarnick/status/1814149823765086384

Miscellaneous:

1. “New study found that relative reproductive success (RLRS) is higher for people with high ADHD polygenic scores and lower for people with high education attainment and cognitive polygenic scores. People are becoming genetically more ADHD and genetically lower IQ” https://x.com/BronskiJoseph/status/1813571969536630999 (paper: https://link.springer.com/article/10.1007/s10519-024-10189-8]

2. New anti-ageing therapy extends life of mice by 25%, study finds https://www.nature.com/articles/s41586-024-07701-9

3. “We find no evidence for a negative association between COVID-19 infection and subsequent measures of cognitive functioning. The associations found in earlier studies may at least partly reflect reverse causation.” https://marginalrevolution.com/marginalrevolution/2024/07/good-news-on-covid-and-your-brain.html

4. “The Meiji government translated 10,000 technical books...Then, Japan became an industrial powerhouse. Must be the greatest industrial policy investment ever made!” https://x.com/whyvert/status/1814008181104029995 (paper: https://www.nber.org/papers/w32667)

Links for 2024-07-21

1. Interfacing an LLM with a reliable symbolic system (Prolog) raises math performance near ceiling: Tested on an *entirely new* collection of math word problems, the Non-Linear (NLR) reasoning dataset, to ensure all were outside the LLM training set. GPT fails completely. But GPT writing prolog code succeeds near ceiling. https://arxiv.org/abs/2407.11373

2. How can informal reasoning improve formal theorem proving? Lean-STaR: A framework for learning to interleave informal thoughts with steps of formal proving. Training language models to produce informal thoughts prior to each step of a proof, thereby improving the model’s formal theorem-proving capabilities. https://arxiv.org/abs/2407.10040

3. Adding self-modeling to artificial networks causes a significant reduction in network complexity. When artificial networks learn to predict their internal states as an auxiliary task, they change in a fundamental way. https://arxiv.org/abs/2407.10188

4. A system that incorporates both natural language pre-training and reinforcement learning from the start. https://arxiv.org/abs/2308.01399

5. Georgia Tech researchers have developed a neural network, RTNet, that mimics human decision-making processes, including confidence and variability, improving its reliability and accuracy in tasks like digit recognition. https://research.gatech.edu/new-neural-network-makes-decisions-human-would

6. R+X: Retrieval and Execution from Everyday Human Videos — By using a VLM for retrieval and in-context IL for execution, robots can now learn from unlabelled videos of humans performing tasks. https://www.robot-learning.uk/r-plus-x

7. “We trained GPT2 to predict the product of two numbers up to 🌟20🌟 digits w/o intermediate reasoning steps, surpassing our previous 15-digit demo! How does a 12-layer LM solve 20-digit multiplication w/o CoT?🤯” https://arxiv.org/abs/2405.14838

8. AI AI Bias: Large Language Models Favor Their Own Generated Content https://arxiv.org/abs/2407.12856

9. Why Has Predicting Downstream Capabilities of Frontier AI Models with Scale Remained Elusive? https://arxiv.org/abs/2406.04391

10. NuminaMath datasets: the largest collection of ~1M math competition problem-solution pairs, ranging in difficulty from junior challenge to Math Olympiad preselection. These datasets were used to win the 1st Progress Prize of the AI Math Olympiad. https://huggingface.co/collections/AI-MO/numinamath-6697df380293bcfdbc1d978c

11. The AI-Powered Future of Coding Is Near https://www.wired.com/ai-powered-coding/ [no paywall: https://archive.is/BRBCw]

12. OpenAI employee: a 60% probability that AGI will have been built in the next 3 years, and 90% in the next 5 years. https://x.com/TolgaBilge_/status/1814828193985003666

Miscellaneous:

1. Accidentally exposed yellowish-green crystals reveal ‘mind-blowing’ finding on Mars, scientists say https://edition.cnn.com/2024/07/20/science/nasa-curiosity-rover-mars-sulfur-rocks/index.html

2. Chinese nuclear reactor is completely meltdown-proof https://www.newscientist.com/article/2440388-chinese-nuclear-reactor-is-completely-meltdown-proof/ [no paywall: https://archive.is/OPcrZ]

3. Your brain on shrooms — how psilocybin resets neural networks https://www.nature.com/articles/d41586-024-02275-y

4. Dogs might have evolved to read your emotions https://www.nature.com/articles/d41586-024-02320-w

Links for 2024-07-25

AI:

1. Mistral: “Today, we release Mistral Large 2, the new version of our largest model. Mistral Large 2 is a 123B-parameter model with a 128k context window. On many benchmarks (notably in code generation and math), it is superior or on par with Llama 3.1 405B.” https://mistral.ai/news/mistral-large-2407/ (Try it here: https://chat.mistral.ai/chat)

2. Meta: “Introducing Llama 3.1: Our most capable models to date” https://ai.meta.com/blog/meta-llama-3-1/ (Try it here: https://huggingface.co/chat/)

3. Mark Zuckerberg: “Open Source AI Is the Path Forward” https://about.fb.com/news/2024/07/open-source-ai-is-the-path-forward/

4. OpenAI: “We’ve developed Rule-Based Rewards (RBRs) to align AI behavior safely without needing extensive human data collection, making our systems safer and more reliable for everyday use.” https://openai.com/index/improving-model-safety-behavior-with-rule-based-rewards/

5. System-1.x: Learning to Balance Fast and Slow Planning with Language Models https://arxiv.org/abs/2407.14414

6. Study suggests that models may have more nuanced and generalizable understandings of truth and falsity than previously thought. Authors find a two-dimensional subspace within an LLM's representations: a general truth direction and a polarity-sensitive one. https://arxiv.org/abs/2407.12831

7. Targeted Latent Adversarial Training Improves Robustness to Persistent Harmful Behaviors in LLMs https://arxiv.org/abs/2407.15549

8. Cross Anything: General Quadruped Robot Navigation through Complex Terrains https://arxiv.org/abs/2407.16412

9. MovieDreamer: Hierarchical Generation for Coherent Long Visual Sequence https://arxiv.org/abs/2407.16655

10. Generating all tokens at once with flow matching https://arxiv.org/abs/2407.15595

11. An open-source platform that enables development teams to create AI agents that monitor issues, manage bugs, and handle various aspects of the software lifecycle—all through natural language interactions. https://venturebeat.com/ai/google-brings-ai-agent-platform-project-oscar-open-source/

Neuroscience:

1. “Our results shed light on the possible rationale for the brain’s modularity and suggest that artificial systems can use this insight from neuroscience to improve learning and generalization in natural tasks.” [PDF] https://www.science.org/doi/pdf/10.1126/sciadv.adk1256

2. “If you want to unlock unlimited intelligence you need to increase density of microtubules in the PFC” https://x.com/SterlingCooley/status/1757802109293261245

3. Rice neuroscientists to build state-of-the-art neural recording system https://news.rice.edu/news/2024/rice-neuroscientists-build-state-art-neural-recording-system

Drones:

1. The world’s smallest and lightest solar-powered drone, weighing just 4.21g with a 200mm wingspan. It can fly non-stop during daylight. https://www.tomshardware.com/tech-industry/new-chinese-drone-can-fly-as-long-as-the-sun-shines-solar-powered-device-with-200mm-wingspan-weighs-record-breaking-421g

2. Bug brains could help drone swarms find their way home https://www.popsci.com/technology/drones-ants-memory/

Miscellaneous:

1. "We estimate that the decline in Nuclear power Plants caused by Chernobyl led to the loss of approximately 141 million expected life years in the U.S., 33 in the U.K. and 318 million globally" [PDF] https://conference.nber.org/conf_papers/f205791.pdf

2. HKUST Engineering Researchers Discover a “Secret” Hidden Structure that Paves New Way of Making More Efficient and Stable Perovskite Solar Cells https://seng.hkust.edu.hk/news/20240719/hkust-engineering-researchers-discover-secret-hidden-structure-paves-new-way-making-more-efficient-and-stable-perovskite-solar-cells

3. Free book: Linear Algebra for Data Science https://kyunghyuncho.me/linear-algebra-for-data-science/

4. “China appears to be stockpiling materials at a rapid pace…when commodities are expensive” https://www.economist.com/finance-and-economics/2024/07/23/why-is-xi-jinping-building-secret-commodity-stockpiles [no paywall: https://archive.is/XiIMu]