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 [Machine learning unlocks secrets to advanced alloys | MIT News | Massachusetts Institute of Technology](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 [Op-ed: How well can AI chatbots mimic doctors in a treatment setting?](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” [Samsung’s new AI image generating tool is a little too good - The Verge](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. [JPMorgan CEO Jamie Dimon says he'll add thousands of jobs focused on AI in the next couple of years | Business Insider India](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 [Inhibition of IL-11 signalling extends mammalian healthspan and lifespan | Nature](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 (applied science, industry, etc). Then, Japan became an industrial powerhouse. Must be the greatest industrial policy investment ever made! Once again: the importance of upper human capital.” https://x.com/whyvert/status/1814008181104029995 (paper: https://www.nber.org/papers/w32667) Demis Hassabis says AI will enhance science: designing new drugs for medicine, discovering a room temperature superconductor, cracking mathematical conjectures and formulating its own new conjectures https://x.com/tsarnick/status/1814908695706255528?t=ySOCp2a-hjgDN6_Sv3HbSQ&s=19 https://www.sciencedirect.com/science/article/pii/S0010945222001836?via%3Dihub "An underappreciated aspect of productivity: momentum. The more stuff you do the easier it becomes to do stuff (literally anything). And inversely. The idea here is that doing something difficult/high-effort has a high mental "activation energy" (separate from the energy needed to actually doing it). If you're only sparsely active, you have to pay this price every time, which is inefficient, painful, and actually disincentivizes you from doing things -- leading to procrastination. If you're always active, you never have to pay this price, making it dramatically easier and less mentally draining to do pretty much anything." https://x.com/fchollet/status/1815179582246645769 "It is true that, on an immediate and surface level, the Republican party's take on both AI and cryptocurrency is more appealing to me. A lot of Democrats seem to want to harshly regulate both of these in simplistic ways that I think would do more harm than good. However, the stark lack of compassion, reflectiveness, open-mindedness, depth of thought and integrity that one sees in Trump and his minions ... I mean, this manages to feel way worse than "merely" some well meaning bureaucrats seeking to over-regulate important technologies they don't understand... BTW the stultified, often dumb-ass but largely well-meaning bureaucracy of the Democratic party ... versus the selfish, racist, sexist, uncompassionate, utter amorality of the Trump camp ... I mean I'll have to take the Democrats' ham-handed obnoxious attempts to regulate the Singularity into submission, since these are doomed to fail anyway ;p Overall I feel quite grateful that the US gov't is configured to make it quite difficult for the gov't to actually take any highly significant actions domestically. I don't want to minimize the damage that gets done by dumb or corrupt gov't policies regarding healthcare, social welfare, drug policy, education etc. etc. ... nor by all the military misadventures of US and other major governments.... OTOH for things like AI and crypto that are evolving so rapidly, it seems fairly likely to me that the slow pace of gov't cognition, perception and action will cause it to stay substantially out of the picture until an AGI breakthrough actually occurs, at which point -- if indeed AGI has been developed in an open and decentralized way, defying limited and ham-handed attempts by gov't to stop this from happening -- gov't will be powerless to screw with its beneficial transformation of society too badly... What a mess we humans are making ... fortunately ASI will be here soon to put everything in the appropriate superposition of order and chaos, right?? ;-)" https://x.com/bengoertzel/status/1815211923782418519 [GitHub - ComposioHQ/composio: Composio equips agents with well-crafted tools empowering them to tackle complex tasks](https://github.com/ComposioHQ/composio) https://x.com/svpino/status/1814269990884855942?t=Z2JuVffv-pDBPNnVnzou0A&s=19 https://dashboard.eyelevel.ai/xray Map of complexity science [complexity history, social complexity history, complexity theory, complexity sciences, history, complexity map, map of complexity science, history of complexity, complexity theory, complexity science, complexity, brian castellani, lasse gerrits, durham sociology complexity](https://www.art-sciencefactory.com/complexity-map_feb09.html) New paper finds that having multiple instances of an AI debate and critique each other's responses leads to more accurate results https://arxiv.org/abs/2305.14325 [Researchers boost AI performance by having multiple AIs "debate" each other](https://suchscience.net/multiagent-debate-enchances-ai-performance/) https://arxiv.org/abs/2407.14402 https://x.com/burny_tech/status/1815473921187098685 A model has started training on the new and most powerful AI cluster in the world? https://x.com/elonmusk/status/1815325410667749760 UBI study: cash transfers increased recipients' agency in making employment decisions that aligned with their individual circumstances, goals, and values https://x.com/sama/status/1815432708568608946 [Findings | OpenResearch](https://www.openresearchlab.org/findings) [Reversal of Autism Symptoms among Dizygotic Twins through a Personalized Lifestyle and Environmental Modification Approach: A Case Report and Review of the Literature - PubMed](https://pubmed.ncbi.nlm.nih.gov/38929862/) [MathΣtral | Mistral AI | Frontier AI in your hands](https://mistral.ai/news/mathstral/) 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. [A New Neural Network Makes Decisions Like a Human Would | Research](https://research.gatech.edu/new-neural-network-makes-decisions-human-would) 9. Why Has Predicting Downstream Capabilities of Frontier AI Models with Scale Remained Elusive? https://arxiv.org/abs/2406.04391 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. [R + X](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 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. [NuminaMath - a AI-MO Collection](https://huggingface.co/collections/AI-MO/numinamath-6697df380293bcfdbc1d978c) 11. The AI-Powered Future of Coding Is Near [The AI-Powered Future of Coding Is Near | WIRED](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 [NASA’s Curiosity rover makes its ‘most unexpected’ find on Mars | CNN](https://edition.cnn.com/2024/07/20/science/nasa-curiosity-rover-mars-sulfur-rocks/index.html) 2. Chinese nuclear reactor is completely meltdown-proof [Chinese nuclear reactor is completely meltdown-proof | New Scientist](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 [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 [Dogs might have evolved to read your emotions](https://www.nature.com/articles/d41586-024-02320-w) "Karl Friston says that current AI language models lack an essential quality of intelligent agents and that is curiosity: "to be an agent is to be curious"" https://x.com/tsarnick/status/1815323807244013842 "Sara Hooker of Cohere says the emergent properties of AI models which are unlocked at scale are the result of memorization of low frequency features of the training data and not something mystical that comes out of nowhere" https://x.com/tsarnick/status/1815313144157528325 https://phys.org/news/2024-07-quantum-microscopy-electrons-visible-motion.html Neural cellular automata https://x.com/Esychology/status/1815366496845009104?t=hEhAyJgwPFbgVb82_HRMkA&s=19 Kokotailo AGI singularity timelines https://x.com/i_dg23/status/1814965562700906503?t=7WYsp0AEl7XJuIAAOZyqSQ&s=19 https://medium.com/@causalwizard/why-and-how-to-integrate-causality-in-predictive-modelling-670bd66c226b https://neurosciencenews.com/path-of-least-resistance-6139/ "Intelligence is the ability to make models. General intelligence means that a sufficiently large computational substrate can be fitted to an arbitrary computable function, within the limits of that substrate. Humans might already be in that category, for a rather small substrate." https://x.com/Plinz/status/1019328120497803271 https://arxiv.org/abs/2407.15908 https://x.com/WiringTheBrain/status/1816018582154715462?t=5Z1QtswOp_mHk8ouHii90A&s=19 [An Introduction to Consciousness Theories [Layer 1] - YouTube](https://youtu.be/GDjnEiys98o?si=tLCopqM-HRkmsO4W) Joscha x Penrose https://x.com/Plinz/status/1815676091450998940?t=UKHv4xGI90uzxg06rven3Q&s=19 "Humanity is likely to "Speciate" within the next 20 years. Where one group decides to Merge with AI, and the remainder do not... Which part will you join?" https://x.com/PeterDiamandis/status/1816531940138451416?t=t1XNjOkAqwsLnERIxfwSEw&s=19 https://arxiv.org/abs/2407.15908 [Humans Could Acquire a New Form of Consciousness, Scientists Say](https://www.popularmechanics.com/technology/a61075478/human-machine-hybrid-consciousness/?utm_campaign=trueanthemFBPOP&utm_medium=social&utm_source=facebook) [Single-cell multiregion dissection of Alzheimer’s disease | Nature](https://www.nature.com/articles/s41586-024-07606-7) https://x.com/manoliskellis/status/1816161472168485274 [The big problem with how we measure “intelligence” - Big Think](https://bigthink.com/the-well/does-testing-really-measure-intelligence/?utm_term=Autofeed&utm_medium=Social&utm_source=Facebook#Echobox=1721999353) [Context-aware geometric deep learning for protein sequence design | Nature Communications](https://www.nature.com/articles/s41467-024-50571-y) [AI-Powered Weather and Climate Models Are Set to Change Forecasting](https://singularityhub.com/2024/07/23/ai-powered-weather-and-climate-models-are-set-to-change-forecasting/) [Neural general circulation models for weather and climate | Nature](https://www.nature.com/articles/s41586-024-07744-y) [Scientists Say They Extended Mice's Lifespans 25% With an Antibody Drug](https://singularityhub.com/2024/07/22/scientists-say-they-extended-mices-lifespans-25-with-an-antibody-drug/) [Inhibition of IL-11 signalling extends mammalian healthspan and lifespan | Nature](https://www.nature.com/articles/s41586-024-07701-9) "Progressives think that values need to be improved, rationalists think values need to be proven, conservatives think values need to be empirically tested, liberals think values need to originate in the individual, romantic liberals think values have to rationally serve transcendence, libertarians think values need to serve the individual, identitarians think values need to serve the ingroup, Kantians think values need to be universal, Christians think values need to serve God, humanists think that values need to serve the Greater Good, fascists think that values serve evolutionary incentives, cynics think that values are what you get away with, post modernists think that values are socially enforced stories, sociopaths observe that values are socially induced and subconsciously grounded defects of rationality, Fristonians think that values ought to maximize total energy utilization, nihilists see that there is no value, moderates think the truth is in the middle." https://x.com/Plinz/status/1817057953452818851 Universal Basic Income for artists in Ireland ❤️ [Basic Income for the Arts (BIA)](https://www.citizensinformation.ie/en/employment/unemployment-and-redundancy/employment-support-schemes/basic-income-arts/) [Announcing Entropycon 12017 | Hivewired](https://hivewired.wordpress.com/2017/05/07/announcing-entropycon-12017/) https://x.com/BasedBeffJezos/status/1817097098807935062 Exploration and acquisition of knowledge to refine our model of our surroundings (thereby increasing our predictive power over the latter) is the fundamental driving force of life and intelligence in our universe. [[LLM News] Llama 3.1, Self-Improving Agents, Text-to-SQL, Multi-LLM Systems, Fine-tuning GPT-4o mini - YouTube](https://www.youtube.com/watch?v=ooRl0shfmAo) "I wonder when social media companies will have sophisticated tools to trace, predict and plan the genealogy, incentive structure, causal power and distribution of memes for politics and consumption at meaningfully high resolution As a social media user, I also want to see a real time map of the meme galaxy, my own location in it, and the local and global meme weather report. And who gives what marching order to whom on this national middle school yard? Who are the main hubs and jet streams?" https://x.com/Plinz/status/1818146356474843501 Which architecture was my neural network collective intelligence psyopped into? "Successful people create companies. More successful people create countries. The most successful people create religions." [Successful people - Sam Altman](https://blog.samaltman.com/successful-people) [The world is awful. The world is much better. The world can be much better. - YouTube](https://youtu.be/CGmdvdXoJ-8?si=6R8effijzevRLc2g) https://www.marktechpost.com/2024/07/29/neural-magic-releases-fully-quantized-fp8-version-of-metas-llama-3-1-405b-model-fp8-dynamic-quantization-and-fp8-static-quantization/ [Neuralink rival Synchron offers thought control with Apple Vision Pro](https://www.cnbc.com/amp/2024/07/30/neuralink-rival-synchron-offers-thought-control-with-apple-vision-pro-.html) [Why "Grokking" AI Would Be A Key To AGI - YouTube](https://www.youtube.com/watch?v=Nvb_4Jj5kBo) [The Trillion Dollar Equation - YouTube](https://www.youtube.com/watch?v=A5w-dEgIU1M) https://www.lesswrong.com/s/ZbmRyDN8TCpBTZSip/p/hHnpn3mEPbJMFLj4g "I think that it's likely that the first superintelligent AI will turn out to be internally misaligned and fail to achieve a goal as complex as taking over the world." 😁 [OpenAI says US govt will get to see ChatGPT 5 before anyone else - India Today](https://www.indiatoday.in/technology/news/story/openai-says-us-govt-will-get-to-see-chatgpt-5-before-anyone-else-2575890-2024-08-02) [Fully-automatic robot dentist performs world's first human procedure](https://newatlas.com/health-wellbeing/robot-dentist-world-first/) [The Wide Boundary Impacts of AI with Daniel Schmachtenberger | TGS 132 - YouTube](https://www.youtube.com/watch?v=1NFuddEAi5s) "There are only 3 great scientific questions: 1. What's the universe made of? 2. What's life all about? 3. What is intelligence? There are interesting sub-questions: 1.1 What's dark matter and dark energy? 1.2 how do you get "it from bit" to paraphrase John Wheeler 1.3 what is the nature of time? 2.1 is the emergence of life an intrinsic property of the universe? 2.2 how does complexity spontaneously form? 3.1 is the emergence of intelligent behavior an intrinsic property of the universe? 3.2 how does intelligence spontaneously form? (which is why learning is so fascinating) At the core of all of these questions is the nature of information and computation." https://x.com/ylecun/status/1819670025155850466 [Consequences of Our Technological Transformation | Steve Fuller, ep 32 - YouTube](https://youtu.be/1eVacMOgkMY?si=4HYyflqzPsyogS_H) transhumanism [Transhumanist Party - Wikipedia](https://en.wikipedia.org/wiki/Transhumanist_Party) [Transhumanist politics - Wikipedia](https://en.wikipedia.org/wiki/Transhumanist_politics) [Cognitive liberty - Wikipedia](https://en.wikipedia.org/wiki/Cognitive_liberty) [A.I. ‚Äê Humanity's Final Invention? - YouTube](https://www.youtube.com/watch?v=fa8k8IQ1_X0) [The Impact of Unconditional Cash Transfers on Consumption and Household Balance Sheets: Experimental Evidence from Two US States | NBER](https://www.nber.org/papers/w32784) [Home](https://sites.google.com/view/competitive-robot-table-tennis/home?utm_source&utm_medium&utm_campaign&utm_content&pli=1) https://www.artificialintelligence-news.com/news/paige-and-microsoft-unveil-next-gen-ai-models-cancer-diagnosis/ [Andrew Yang on the path to UBI, the explosion of AI, and optimism for the future - YouTube](https://www.youtube.com/watch?v=m-XUAQ7k_Bw) "