for - question - climate crisis - climate models underestimate warming in some areas up to 4x - what is the REAL carbon budget if adjusted to the real situation?

question - climate crisis - climate models underestimate warming in some areas up to 4x<br /> - What is the REAL carbon budget if adjusted to the real situation? - If we have even less than 5 years remaining in our carbon budget, then how many years do we actually have to stay within 1.5 Deg C?

for - climate model - gaps - from - post - LinkedIn - Reality vs Climate Models - Kasper Benjamin Reimer BjørkskovKasper Benjamin Reimer Bjørkskov

from - post - Linked In - Reality vs Climate Models - Kasper Benjamin Reimer BjørkskovKasper Benjamin Reimer Bjørkskov - https://hyp.is/Dc_w8rM2Ee-I0VO9JwZKNg/www.linkedin.com/feed/update/urn:li:activity:7270232384455720960/?commentUrn=urn:li:comment:(activity:7270232384455720960,7270500962702655489)&dashCommentUrn=urn:li:fsd_comment:(7270500962702655489,urn:li:activity:7270232384455720960)

for - climate crisis - paper - models are underestimating by up to 4x - to - paper - Global emergence of regional heatwave hotspots outpaces climate model simulations Kornhuber et al, 2024 - https://hyp.is/9cS36LMtEe-2oL8C4AgQOQ/www.pnas.org/doi/10.1073/pnas.2411258121

for - climate crisis - models are underestimating as much as 4x - question - does our current remaining carbon budget of 5 years BAU need to be reduced?

for - climate crisis - Global Climate Models (GCM) limitation - do not allow alternative perspectives - unrealistic sensitivity to injection of fresh water into upper layers of the ocean - Jim Hansen

for - climate crisis - IPCC - warning - working group 3 - integrated assessment models - are basically economic models - with a bit of mythical technology - a bit of social science - Kevin Anderson

for - climate crisis - IPCC - warning - working group 3 - integrated assessment models - is just about reducing emissions - inherently political - Kevin Anderson

for - climate crisis - IPCC - warning - working group 3 - Integrated Assessment Models - Some good people here but - It's just Exxon in disguise - Kevin Anderson

for - quote - IPCC - climate models - are deeply colonial - Kevin Anderson

Implementing Trained AI Models in Mobile App Development is transforming app experiences by integrating machine learning into iOS and Android platforms. From AI-powered personalization to advanced analytics, trained models empower intelligent decision-making and enhanced functionality.

AI models collapse when trained on recursively generated data by Ilia Shumailov et al.

ᔥ[[Mathew Lowry]] in AI4Communities post - MyHub Experiments Wiki (accessed:: 2024-11-06 09:43:23)

https://metagov.org/projects/koi-pond

Metagov's KOI (Knowledge Organization Infrastructure) is a graph database that supports relationships between knowledge objects, users, and groups within Metagov. via JM

Introducing Llama 3.1: Our most capable models to date by [[Meta]]

Moderate people, not code by [[Ryan Barrett]]

https://research.ibm.com/blog/retrieval-augmented-generation-RAG

PK indicates that folks using footnotes in AI are using rag methods.

结论：预测结果，好于MOST（MO估计系统地低估了湍流通量的大小，改善了与观测值和减小与观测通量偏离的总幅度。）,不同地点的泛化能力 不足：不含物质通量，预测结果待提升，结果因稳定性而异常，不同季节的泛化能力，运用了不易获得的变量（找到最小观测集）

for - climate crisis - interview - Neil degrasse Tyson - Gavin Schmidt - 2023 record heat - NASA explanation

podcast details - title: How 2023 broke our climate models - host: Neil degrasse Tyson & Paul Mercurio - guest: NASA director, Gavin Schmidt - date: Jan 2024

summary - Neil degrasse and his cohost Paul Mercurio interview NASA director Gavin Schmidt to discuss the record-breaking global heating in 2023 and 2024. - Neil and Paul cover a lot in this short interview including: - NASA models can't explain the large jump in temperature in 2023 / 2024. Yes, they predicted incremental increases, but not such large jumps. Gavin finds this worrying. - PACE satellite launches this month, to gather important data on the state of aerosols around the planet. This infomration can help characterize more precisely the role aerosols are playing in global heating. - geoengineering with aerosols is not considered a good idea by Gavin, as it essentially means once started, and if it works to cool the planet, we would be dependent on them for centuries. - Gavin stresses the need for a cohesive collective solution, but that it's beyond him how we achieve that given all the denailsim and misinformation that influeces policy out there.

Hubinger, et. al. "SLEEPER AGENTS: TRAINING DECEPTIVE LLMS THAT PERSIST THROUGH SAFETY TRAINING". Arxiv: 2401.05566v3. Jan 17, 2024.

Very disturbing and interesting results from team of researchers from Anthropic and elsewhere.

• for: progress trap -AI, carbon footprint - AI, progress trap - AI - bias, progress trap - AI - situatedness

haha, china and russia and friends are shitting all over your "scientific models".<br /> the ONLY problem is "too many humans", aka overpopulation, caused by pacifism.<br /> these "save the world" policies are collective suicide for the 95% useless eaters. byee!

Introduction of the RoBERTa improved analysis and training approach to BERT NLP models.

(Chen, NeurIPS, 2021) Che1, Lu, Rajeswaran, Lee, Grover, Laskin, Abbeel, Srinivas, and Mordatch. "Decision Transformer: Reinforcement Learning via Sequence Modeling". Arxiv preprint rXiv:2106.01345v2, June, 2021.

Quickly a very influential paper with a new idea of how to learn generative models of action prediction using SARSA training from demonstration trajectories. No optimization of actions or rewards, but target reward is an input.

Wu, Prabhumoye, Yeon Min, Bisk, Salakhutdinov, Azaria, Mitchell and Li. "SPRING: GPT-4 Out-performs RL Algorithms byStudying Papers and Reasoning". Arxiv preprint arXiv:2305.15486v2, May, 2023.

Zecevic, Willig, Singh Dhami and Kersting. "Causal Parrots: Large Language Models May Talk Causality But Are Not Causal". In Transactions on Machine Learning Research, Aug, 2023.

"The Age of AI has begun : Artificial intelligence is as revolutionary as mobile phones and the Internet." Bill Gates, March 21, 2023. GatesNotes

Minda Zetlin. "Bill Gates Says We're Witnessing a 'Stunning' New Technology Age. 5 Ways You Must Prepare Now". Inc.com, March 2023.

Feng, 2022. "Training-Free Structured Diffusion Guidance for Compositional Text-to-Image Synthesis"

Shared and found via: Gowthami Somepalli @gowthami@sigmoid.social Mastodon > Gowthami Somepalli @gowthami StructureDiffusion: Improve the compositional generation capabilities of text-to-image #diffusion models by modifying the text guidance by using a constituency tree or a scene graph.

Original Paper for discussion of the Reinforcement Learning with Human Feedback algorithm.

GPT-2 Introduction paper

Language Models are Unsupervised Multitask Learners A. Radford, J. Wu, R. Child, D. Luan, D. Amodei, and I. Sutskever, (2019).

"Attention is All You Need" Foundational paper introducing the Transformer Architecture.

GPT-3 introduction paper

"Are Pre-trained Convolutions Better than Pre-trained Transformers?"

"LaMDA: Language Models for Dialog Application" Meta's introduction of LaMDA v1 Large Language Model.

Benyamin GhojoghAli Ghodsi. "Attention Mechanism, Transformers, BERT, and GPT: Tutorial and Survey"

• for: accelerating foundation models in non-human communication, non-human communication - anthropogenic impacts, species extinction - AI communication tools, conservation - AI communication tools

• comment

  • imagine the empathy we can realize to help slow down climate change and species extinction by communicating and listening to the feedback from other species about what they think of our species impacts on their world!

The data of the past is congruent with the current psychological trends that face the education system of today. Developmentalists have charted how children construct and revise intuitive theories. In turn, a variety of theories have developed because of the greater use of statistical information that supports probabilistic contingencies that help to better inform us of causal models and their distinctive cognitive functions. These studies investigate the physical, psychological, and social domains. In the case of intuitive psychology, or "theory of mind," developmentalism has traced a progression from an early understanding of emotion and action to an understanding of intentions and simple aspects of perception, to an understanding of knowledge vs. ignorance, and finally to a representational and then an interpretive theory of mind.

The mechanisms by which life evolved—from chemical beginnings to cognizing human beings—are central to understanding the psychological basis of learning. We are the product of an evolutionary process and it is the mechanisms inherent in this process that offer the most probable explanations to how we think and learn.

Bada, & Olusegun, S. (2015). Constructivism Learning Theory : A Paradigm for Teaching and Learning.

Title: Delays, Detours, and Forks in the Road: Latent State Models of Training Dynamics Authors: Michael Y. Hu1 Angelica Chen1 Naomi Saphra1 Kyunghyun Cho Note: This paper seems cool, using older interpretable machine learning models, graphical models to understand what is going on inside a deep neural network

Link: https://arxiv.org/pdf/2308.09543.pdf

LLAMA 2 Release Paper

Daniel Adiwardana Minh-Thang Luong David R. So Jamie Hall, Noah Fiedel Romal Thoppilan Zi Yang Apoorv Kulshreshtha, Gaurav Nemade Yifeng Lu Quoc V. Le "Towards a Human-like Open-Domain Chatbot" Google Research, Brain Team

Defined the SSI metric for chatbots used in LAMDA paper by google.

Bowen Baker et. al. (Open AI) "Video PreTraining (VPT): Learning to Act by Watching Unlabeled Online Videos" Arkiv, June 2022.

Introduction of VPT : New semi-supervied pre-trained model for sequential decision making on Minecraft. Data are from human video playthroughs but are unlabelled.

https://en.wikipedia.org/wiki/Spherical_cow

Raworth, Kate. Doughnut Economics: Seven Ways to Think Like a 21st-Century Economist. White River Junction, Vermont: Chelsea Green Publishing, 2017.

The Annotated S4 Efficiently Modeling Long Sequences with Structured State Spaces Albert Gu, Karan Goel, and Christopher Ré.

A new approach to transformers

Efficiently Modeling Long Sequences with Structured State SpacesAlbert Gu, Karan Goel, and Christopher R ́eDepartment of Computer Science, Stanford University

Bowman, Samuel R.. "Eight Things to Know about Large Language Models." arXiv, (2023). https://doi.org/https://arxiv.org/abs/2304.00612v1.

Abstract

The widespread public deployment of large language models (LLMs) in recent months has prompted a wave of new attention and engagement from advocates, policymakers, and scholars from many fields. This attention is a timely response to the many urgent questions that this technology raises, but it can sometimes miss important considerations. This paper surveys the evidence for eight potentially surprising such points: 1. LLMs predictably get more capable with increasing investment, even without targeted innovation. 2. Many important LLM behaviors emerge unpredictably as a byproduct of increasing investment. 3. LLMs often appear to learn and use representations of the outside world. 4. There are no reliable techniques for steering the behavior of LLMs. 5. Experts are not yet able to interpret the inner workings of LLMs. 6. Human performance on a task isn't an upper bound on LLM performance. 7. LLMs need not express the values of their creators nor the values encoded in web text. 8. Brief interactions with LLMs are often misleading.

Found via: Taiwan's Gold Card draws startup founders, tech workers | Semafor

Bowen Baker et. al. (Open AI) "Video PreTraining (VPT): Learning to Act by Watching Unlabeled Online Videos" Arkiv, June 2022.

New supervised pre-trained model for sequential decision making on Minecraft. Data are from human video playthroughs but are unlabelled.

reinforcement-learning foundation-models pretrained-models proj-minerl minecraft

This isn't true. The Stochastic Parrots paper outlines other avenues for reining in the harms of language models like GPT's.

Ganguli, Deep, Askell, Amanda, Schiefer, Nicholas, Liao, Thomas I., Lukošiūtė, Kamilė, Chen, Anna, Goldie, Anna et al. "The Capacity for Moral Self-Correction in Large Language Models." arXiv, (2023). https://doi.org/https://arxiv.org/abs/2302.07459v2.

Abstract

We test the hypothesis that language models trained with reinforcement learning from human feedback (RLHF) have the capability to "morally self-correct" -- to avoid producing harmful outputs -- if instructed to do so. We find strong evidence in support of this hypothesis across three different experiments, each of which reveal different facets of moral self-correction. We find that the capability for moral self-correction emerges at 22B model parameters, and typically improves with increasing model size and RLHF training. We believe that at this level of scale, language models obtain two capabilities that they can use for moral self-correction: (1) they can follow instructions and (2) they can learn complex normative concepts of harm like stereotyping, bias, and discrimination. As such, they can follow instructions to avoid certain kinds of morally harmful outputs. We believe our results are cause for cautious optimism regarding the ability to train language models to abide by ethical principles.

Google translation:

Every Egyptology student learns in their first semester that the Egyptian word pt (pronounced pet) means "heaven". The collection of documents in the dictionary archive comprises around 6,000 document slips. In the order of this material one learns that the Egyptian heaven has gates and ways, waters and banks, sides, pillars and chapels. This makes it tangible that the Egyptians had something completely different in mind when they heard the word "heaven" than modern Westerners do, namely a mythical space in which gods and spirits of the dead dwell.

This is a fantastic example of context creation for a dead language as well as for creating proper historical context.

In looking at the uses of and similarities between Wb and TLL, I can't help but think that these two zettelkasten represented the state of the art for Large Language Models and some of the ideas behind ChatGPT

Is there? By whom? Why industry only and not government, academia and civil society?

Bender, Emily M., Timnit Gebru, Angelina McMillan-Major, and Shmargaret Shmitchell. “On the Dangers of Stochastic Parrots: Can Language Models Be Too Big? 🦜” In Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency, 610–23. FAccT ’21. New York, NY, USA: Association for Computing Machinery, 2021. https://doi.org/10.1145/3442188.3445922.

Would the argument here for stochastic parrots also potentially apply to or could it be abstracted to Markov monkeys?

What are the long term effects of fast breaking changes and gradual changes for evolved entities?

// Insight Maker is used to model system dynamics and create agent based models by creating causal loop diagrams and allowing users to run simulations on those

https://en.wikipedia.org/wiki/Armillary_sphere

Could it be the sift from person to person (known in both directions) to massive broadcast that is driving issues with content moderation. When it's person to person, one can simply choose not to interact and put the person beyond their individual pale. This sort of shunning is much harder to do with larger mass publics at scale in broadcast mode.

How can bringing content moderation back down to the neighborhood scale help in the broadcast model?

Hartmut Grassl u.a. über den Unterschied von Klimaneutralität und Treibhausgasneutralität, über Suffizienz und Klimamodellierung.

Is LaMDA really able to "learn higher-level concepts" or is it just a large, straight-forward information theoretic-based prediction engine?

In the fraternity and other social spaces, how does one correct for a "bad first date", a botched meeting, or a lone bad day? Does statistical thermodynamics as a model provide clues? How would rank be determined here in an unbiased way? What about individual chemical affinities and how chemical interactions change and/or bias the samples?

https://ncase.me/polygons/

An interesting interactive model for segregation here. See also https://www.bloomberg.com/news/articles/2014-12-10/an-immersive-game-shows-how-easily-segregation-arises-and-how-we-might-fix-it for press coverage.

A term recommended by Eve regarding an interdisciplinary approach that accounts for multiple feedback loops within complex systems. Need to confer complex systems science to see if ADHD is already addressed in that domain.

"Talking About Large Language Models" by Murray Shanahan

This is probably true of average or poor software engineers given that GPT-3 can produce pretty reasonable code snippets

Noteworthy that the recommendations for quality prioritize 1) The granularity of documenting learning outcomes; and 2) that credentials use standards that can be independently verified and validated.

What constitutes TRUST?

Clearly, the LinkedIn recommendations use case is an example of one of these initiatives. It has not succeeded in creating strong social signals anchored in trust models. We are wise to consider what's missing from efforts like this. An even greater concern however, and one that I believe is an essential if we are to realize the transformative potential of digital credentials, is how to design social signals built on trust models that help all people. In a world long-governed by "it's not what you know, it's who you know," the social signals and trust models are overweighted in favor of people with connections to other people, organizations and brands that are all to some degree legacies of exclusionary and inequitable systems. We are likely to build new systems that perpetuate the same problems if we do not intentionally design them to function otherwise. For people (especially those from historically underserved populations) worthy of the recommendations but lacking in social connections, how do they access social recommendations built on trust models?

How can we frame the science of poverty with respect to the model of statistical mechanics?

Unemployment numbers have very little to do with levels of poverty. They definitely don't seem to be correlated with poverty levels, in fact perhaps inversely so. Many would say that people are lazy and don't want to work when the general reality is that they do want to work (for a variety of reasons including identity and self-esteem), but the amount of work they can find and the pay they receive for it are the bigger problems.

I went through some of the pieces in the collection. It is important to give a platform to the voices that are missing from the conversation usually.

Just a few similar initiatives that you might want to check out:

Storycorps - people can record their stories via an app

Project Voice - spoken word poetry

Living Library - sharing one's story

Freedom Writers - book and curriculum based on real-life stories

Their misleading templates really are misleading

{premise} Can that be paraphrased as "{hypothesis}"

{premise} Is this a sports news? {hypothesis}

although prompts seem to help NLP models improve their performance, the authors find that this performance is still present even when prompts are deliberately misleading which is a bit weird

This is a really excellent illustration of the difference in paradigm between "normal" text model fine tuning and prompt-based modelling

The authors use prompt training on LLMs to build a classifier that can identify statements that describe whether or not micro-organisms have antibiotic resistant genes in scientific papers.

Interesting metaphor for why humans are happy to trust outputs from generative models

https://threadreaderapp.com/thread/1590111416014409728.html

I'm slowly getting the feeling that Musk is a system one thinker who relies on others to do his system two thinking.

"On the Opportunities and Risks of Foundation Models" This is a large report by the Center for Research on Foundation Models at Stanford. They are creating and promoting the use of these models and trying to coin this name for them. They are also simply called large pre-trained models. So take it with a grain of salt, but also it has a lot of information about what they are, why they work so well in some domains and how they are changing the nature of ML research and application.

"TBD 🚀🚀🚀" is such a bad indication for the future of a product

https://www.youtube.com/watch?v=7s4xx_muNcs

Don't recommend unless you have 100 hours to follow up on everything here that goes beyond the surface.

Be aware that this is a gateway for what I'm sure is a relatively sophisticated sales funnel.

Motivational and a great start, but I wonder how many followed up on these techniques and methods, internalized them and used them every day? I've not read his book, but I suspect it's got the usual mnemonic methods that go back millennia. And yet, these things are still not commonplace. People just don't seem to want to put in the work.

As a result, they become a sales tool with a get rich quick (get smart quick) hook/scheme. Great for Kwik's pocketbook, but what about actual outcomes for the hundreds who attended or the 34.6k people who've watched this video so far?

These methods need to be instilled in youth as it's rare for adults to bother.

Acronyms for remembering things are alright, but not incredibly effective as most people will have issues remembering the acronym itself much less what the letters stand for.

There seems to be an over-fondness for acronyms for people selling systems like this. (See also Tiago Forte as another example.)

Forte is fairly good at contextualizing people and proving ethos for what he's about to present. Essentially saying, "these people are the smart, well-known geniuses, so let's imitate them".

Humans are already good at imitating. Are they even better at it or more motivated if the subject of imitation is famous?

See also his sections on Twyla Tharp and Taylor Swift...

link to : - lone genius myth: how can there be a lone genius when the majority of human history is littered with imitation?

...but we'll make up for it in volume.

For many tools and apps payment for privacy is becoming the norm.

Examples: - Kumu.io - Github for private repos - ...

pros: - helps to encourage putting things into the commons

cons: - Normalizes the idea of payment for privacy which can be a toxic tool.

discuss...

https://www.latimes.com/entertainment-arts/story/2022-05-02/neal-adams-crusty-bunkers-burbank

Adam Kucharski [@adamjkucharski]. (2021, October 26). Lots of useful international examples in the comments to this post 👇 And more from @cmmid_lshtm here: Https://github.com/cmmid [Tweet]. Twitter. https://twitter.com/adamjkucharski/status/1452905501684011008

Kai Kupferschmidt. (2021, December 1). @DirkBrockmann But these kinds of models do help put into context what it means when certain countries do or do not find the the variant. You can find a full explanation and a break-down of import risk in Europe by airport (and the people who did the work) here: Https://covid-19-mobility.org/reports/importrisk_omicron/ https://t.co/JXsYdmTnNP [Tweet]. @kakape. https://twitter.com/kakape/status/1466109304423993348

NetScience on Twitter. (n.d.). Twitter. Retrieved 15 February 2021, from https://twitter.com/net_science/status/1360990028168503297

ReconfigBehSci. (2022, January 26). RT @chrischirp: One consequence of the Omicron epidemic moving from.older people into children is dropping hospital admissions. Fewer adm… [Tweet]. @SciBeh. https://twitter.com/SciBeh/status/1486618430182731776

Learnings: - It's easy to assume people in the past didn't care or were stupid. But people do things for a reason. Not understanding the reason for how things are is a missed learning opportunity, and very likely leads to unintended consequences. - Similar to having a valid strong opinion, one must understand why things are as they are before changing them (except if the goal is only signaling).

Look at their life and how it can bring opportunities and then if you will be exposed and streched.

isolated nor learned in an isolated fashion, but hang together in a network of ideas, or “latticework of mental models” (Munger, 1994), it becomes easier to make sense of new information. That makes it easier not only to learn and remember, but also to retrieve the information later in the moment and context it is needed.

Our natural memories are limited in their capacities, but it becomes easier to remember facts when they've got an association to other things in our minds. The building of mental models makes it easier to acquire and remember new information. The down side is that it may make it harder to dramatically change those mental models and re-associate knowledge to them without additional amounts of work.

The mental work involved here may be one of the reasons for some cognitive biases and the reason why people are more apt to stay stuck in their mental ruts. An example would be not changing their minds about ideas of racism and inequality, both because it's easier to keep their pre-existing ideas and biases than to do the necessary work to change their minds. Similar things come into play with respect to tribalism and political party identifications as well.

This could be an interesting area to explore more deeply. Connect with George Lakoff.

As the result of online business models and SEO, most writing becomes about chasing clout and audience eyeballs rather than providing thought provoking insight and razor sharp analysis. The audience reaction has weakened with the anger reaction machines like Twitter.

We need better business models that aren't built on hype.

http://puck.news/who-is-puck/

Another platform play of journalists banding together to find a niche space of readers.

One of the most complex questions in journalism in the past decade or more is how can one best align editorial mission with the business model? This is particularly difficult because the traditional business model(s) have been shifting in the move to online.

Old business advice: find the rich and charge them a pretty penny for something they either think they need or fear they can't live without.

We are at a point in time where we need to reflect on our business practices. The question should be who are we serving and how do we show we are serving them.

In che modo si può applicare il modello OODA Loop nella vita quotidiana?

Semplicemente applicando ad ogni nostra decisione le fasi previste dal modello, rendendo questo processo una abitudine riusciremo ad essere sempre più veloci nell'eseguirlo e questo ci darà la velocità necessaria per sopravvivere e vincere.

Quale è il ruolo della velocità e della prevedibilità del nostro operato nel modello OODA Loop ?

Operare a velocità maggiore degli altri ci rende imprevedibili e questo ci fornisce un vantaggio competitivo, adatto all'OODA Loop che è per definizione un modello fluido.

In che modo la [[Seconda legge della termodinamica]] si applica all' #incertezza e come possiamo utilizzarla come parte del modello OODA Loop ?

Il principio afferma che all'interno di un sistema chiuso tutto tenderà sempre all'entropia. Per questo bisogna essere dei sistemi aperti acquisendo ogni volta informazioni dal contesto, così da evitare che la situazione diventi caotica.

In che modo si applica il [[principio di indeterminazione di Heisenberg]] nel modello [[OODA Loop]] ed in che modo ci aiuta ad affrontare l' #incertezza ?

Il principio afferma che è impossibile determinare in maniera specifica due proprietà fisiche allo stesso tempo.

Boyd estende questo concetto anche alla gestione delle informazioni, cercare di gestire al meglio due variabili informative diverse è troppo difficile ed all'atto pratico induce a maggiore incertezza

Quali sono i tre principi che possono aiutarci nella gestione dell'incertezza e parte integrante del modello OODA Loop?

• Il teorema di #Godel
• [[principio di indeterminazione di Heisenberg]]
• [[Seconda legge della termodinamica]]