The hottest Modeling Substack posts right now

And their main takeaways
Category
Top Climate & Environment Topics

Everyone Loves The Idea Of AI, But Not The Reality

High ROI Data Science • 615 implied HN points • 06 Oct 24
đź•ą Technology Modeling
  1. Many businesses love the idea of AI but find it hard to put into practice. It often looks easy on paper, but the reality is very different when trying to make it work.
  2. Data is really important for AI to work well. Companies need good data to build effective AI products, and often, they realize this too late after facing challenges.
  3. AI projects often fail because businesses don’t fully understand what they need to achieve. Companies should focus on solving real problems rather than just using the latest technology.

Entrevista a Grady Booch

Érase una vez un algoritmo... • 39 implied HN points • 27 Oct 24
đź•ą Technology Modeling
  1. Grady Booch is a key figure in software engineering, known for creating UML, which helps developers visualize software systems. His work has changed how we think about software design.
  2. He emphasizes the ongoing evolution in software engineering due to changes like AI and mobile technology. Adaptation and continuous learning are essential for success in this field.
  3. Booch advocates for ethics in technology development, stressing the need for education and accountability among tech leaders to ensure responsible use of AI and other emerging technologies.

N-1 Block A

Soviet Space Substack • 178 implied HN points • 12 Oct 24
đź•ą Technology Modeling
  1. The N1-3L rocket has a complex engine system, with different engines numbered for clarity. Understanding these details is crucial for analyzing the rocket's design and performance.
  2. Grid fins are an important feature of the N1 rocket, providing enhanced control during high-speed flights. Their design has evolved over time to improve stability and effectiveness.
  3. There were various design changes made to the Block A of the N1 rocket to improve its function and control. These updates were likely based on lessons learned from previous flight tests.

The German Grid Montage

Gordian Knot News • 102 implied HN points • 02 Mar 26
🌞 Climate & Environment Modeling
  1. The dominant technology depends heavily on nuclear overnight cost: if nuclear is cheaper than about $3,000/kW (2020 USD) you get low-cost, low-CO2 grids dominated by nuclear, but if nuclear is much more expensive the model shifts to coal or big wind/solar builds with much higher emissions.
  2. Dispatchable generation like nuclear reduces the need for massive wind/solar overbuild and backup gas because it can reliably follow load, while wind/solar force huge capacity, land use, and storage investments and still require substantial gas backup.
  3. The model is biased optimistic for renewables (no transmission costs, perfect foresight, no inertia/ancillary requirements), so the already-expensive high-renewable solutions in the runs understate real-world costs; batteries are rarely chosen and very high nuclear costs produce politically and economically extreme grids with high curtailment and embedded emissions.
Get a weekly roundup of the best Substack posts, by hacker news affinity:

Dunkelflauten Clusters (aka Winter)

Gordian Knot News • 168 implied HN points • 16 Feb 26
🌞 Climate & Environment Modeling
  1. Dunkelflauten—multi-day clusters of very low wind and solar—can last weeks and stress the grid far more than average capacity factors indicate.
  2. Detailed hour-by-hour, multi-year weather modeling shows a pure wind/solar/battery/hydrogen system for Germany needs massive overbuild and nearly 50,000 GWh of H2 storage, causing huge curtailment and very high electricity costs.
  3. Real-world constraints like missing north–south transmission, low gas reserves, and storage limits make heavy reliance on intermittents and LNG/hydrogen risky, while a nuclear-centered plan would likely be cheaper and cleaner.

The Sequence Knowledge #800: Not All World Models are Created Equal

TheSequence • 147 implied HN points • 03 Feb 26
đź•ą Technology Modeling
  1. There are different types of world models, and a clear taxonomy helps explain how they differ and what roles they play in AI.
  2. For decades, model-free reinforcement learning dominated: agents learned by reinforcing actions without building internal maps or understanding why those actions worked.
  3. Looking at the first major papers on world models reveals the origins and trade-offs of different approaches and shows why some models are better suited for planning and reasoning.

Analyzing A Critique Of The AI 2027 Timeline Forecasts

Don't Worry About the Vase • 2553 implied HN points • 24 Jun 25
đź•ą Technology Modeling
  1. Critiques are important for improving forecasts. It's good to get feedback and adjust predictions based on detailed analysis.
  2. Modeling progress in AI is tricky and uncertain. It's not easy to predict how quickly AI will advance, and different methods can give very different results.
  3. Forecasts should be communicated clearly, without overly negative language. Clear messaging helps everyone understand the importance and limitations of the predictions.

LNT is not quite the "No Repair" Model

Gordian Knot News • 139 implied HN points • 27 Jan 26
🔬 Science Modeling
  1. Allowing proportional DNA repair doesn't save the Linear No‑Threshold (LNT) model, because if cancer mainly arises from closely spaced double‑strand breaks, risk does not track total dose alone and can grow faster than linearly.
  2. If repair takes time, higher dose rates increase the inventory of unrepaired double‑strand breaks and the probability of two breaks clustering rises roughly with the square (or higher power) of dose rate, producing a nonlinear (steeper) risk response.
  3. Biologically, single‑strand breaks are fixed with very high fidelity, but double‑strand breaks can be misrepaired by joining wrong ends; those misrepairs (especially paired or closely spaced DSBs) are the likely mechanism for radiation‑induced cancer, so dose rate and break clustering matter.

From Theory to Practice: Inductive Biases in Machine Learning

Mindful Modeler • 639 implied HN points • 23 Apr 24
đź•ą Technology Modeling
  1. Different machine learning models exhibit varying behaviors when extrapolating features, influenced by their inductive biases.
  2. Inductive biases in machine learning influence the learning algorithm's direction, excluding certain functions or preferring specific forms.
  3. Understanding inductive biases can lead to more creative and data-friendly modeling practices in machine learning.

Statistical modeling seen through inductive biases

Mindful Modeler • 419 implied HN points • 28 May 24
🔬 Science Modeling
  1. Statistical modeling involves modeling distributions and assuming relationships between features and the target with a few interpretable parameters.
  2. Distributions shape the hypothesis space by restricting the range of models compatible with specific distributions like a zero-inflated Poisson distribution.
  3. Parameterization in statistical modeling simplifies estimation, interpretation, and inference of model parameters by making them more interpretable and allowing for confidence intervals.

The Sequence Opinion #798: Inside the Most Important Paper in Agentic Reasoning: Why the Loop is the Logic

TheSequence • 84 implied HN points • 29 Jan 26
đź•ą Technology Modeling
  1. Reasoning comes from the interaction loop with the environment, not just from the model itself.
  2. Current LLMs act like fast, shallow 'System 1' pattern matchers, so they need agentic feedback loops to produce real-world reasoning and agency.
  3. The next frontier is designing the agentic loop and environment (the "new hidden layer") rather than only scaling model parameters.

How to make use of inductive biases

Mindful Modeler • 219 implied HN points • 04 Jun 24
🔬 Science Modeling
  1. Inductive biases play a crucial role in model robustness, interpretability, and leveraging domain knowledge.
  2. Choosing inherently interpretable models can enhance model understandability by restricting the hypothesis space of the learning algorithm.
  3. By selecting inductive biases that reflect the data-generating process, models can better align with reality and improve performance.

Kimi K2 and when "DeepSeek Moments" become normal

Democratizing Automation • 680 implied HN points • 14 Jul 25
đź•ą Technology Modeling
  1. Kimi K2 is a new AI model from a Chinese startup and shows that China is catching up to or surpassing the U.S. in AI development. This means we need to rethink how we view AI technology in the future.
  2. Training leading AI models is becoming easier and cheaper, which means more organizations can create powerful models. This trend hints at a growing competition in the AI landscape.
  3. The gap between open AI models from the West and those from China is widening. This signals a need for stronger support and investment in AI research in the West.

How I made peace with quantile regression

Mindful Modeler • 778 implied HN points • 16 Jan 24
🔬 Science Modeling
  1. Quantile regression can be understood through the lens of loss optimization, specifically with the pinball loss function.
  2. In machine learning, quantile regression is essentially regression with the unique pinball loss function that emphasizes absolute differences between actual and predicted values.
  3. The asymmetry of the pinball loss function, controlled by the parameter tau, dictates how models should handle under- and over-predictions, making quantile regression a tool to optimize different quantiles of a distribution.

The Bitter Pipeline

Abstraction • 39 implied HN points • 28 Jan 26
đź•ą Technology Modeling
  1. Frontier models scale better than human-designed forecasting pipelines, so the structured process that helped smaller models often adds no value with larger models.
  2. Empirical tests show spending compute on polling and ensembling big models improves forecast skill more than token-heavy steps like classification or decomposition, with ensembling giving measurable uplift while the pipeline did not.
  3. The practical move is to simplify: ensemble aggressively, validate empirically, and keep experimenting with ways to elicit latent model knowledge instead of adding complex hand-crafted processes.

No, Sora has not “learned physics”

Marcus on AI • 3596 implied HN points • 02 Mar 24
đź•ą Technology Modeling
  1. Sora is not a reliable source for understanding how the world works, as it focuses more on how things look visually.
  2. Sora's videos often depict objects behaving in ways that defy physics or biology, indicating a lack of understanding of physical entities.
  3. The inconsistencies in Sora's videos highlight the difference between image sequence prediction and actual physics, emphasizing that Sora is more about predicting images than modeling real-world objects.

No Free Dessert in Machine Learning

Mindful Modeler • 399 implied HN points • 20 Feb 24
🔬 Science Modeling
  1. Generalization in machine learning is essential for a model to perform well on unseen data.
  2. There are different types of generalization in machine learning: from training data to unseen data, from training data to application, and from sample data to a larger population.
  3. The No Free Lunch theorem in machine learning highlights that assumptions and effort are always needed for generalization, and there's no free lunch when it comes to achieving further generalization.

GraphRAG Analysis, Part 1: How Indexing Elevates Knowledge Graph Performance in RAG

AI Encoder: Parsing Signal from Hype • 70 HN points • 09 Jul 24
đź•ą Technology Modeling
  1. Knowledge graphs do not significantly impact context retrieval in RAG, as all methods showed similar context relevancy scores.
  2. Neo4j with its own index improved answer relevancy and faithfulness compared to Neo4j without indexing and FAISS, showcasing the importance of effective indexing for precise content retrieval in RAG applications.
  3. Developers need to consider the trade-offs between ROI constraints and performance improvements when deciding to use GraphRAG, especially in high-precision applications that require accurate answers.

to kaggle, or not to kaggle

Mindful Modeler • 379 implied HN points • 13 Feb 24
đź•ą Technology Modeling
  1. There are conflicting views on Kaggle - some see it as a playground while others believe it produces top machine learning results.
  2. Participating in Kaggle competitions can be beneficial to learn core supervised machine learning concepts.
  3. The decision to focus on Kaggle competitions should depend on how much daily tasks align with Kaggle-style work.

DeepSeek and the Future of AI Competition with Miles Brundage

ChinaTalk • 948 implied HN points • 25 Jan 25
đź•ą Technology Modeling
  1. DeepSeek's R1 model shows that AI competition is heating up between the U.S. and China. It's similar to OpenAI's model but developed quickly, closing the gap.
  2. The efficiency at which DeepSeek operates is driven by export controls, meaning limited access to advanced chips. More chips would better their AI capabilities.
  3. Open-sourcing AI models has its benefits, but governments need to be careful. They should ensure the technology is not misused while still allowing some level of open collaboration.

How to sell bread with quantile regression

Mindful Modeler • 339 implied HN points • 23 Jan 24
đź•ą Technology Modeling
  1. Quantile regression can be used for robust modeling to handle outliers and predict tail behavior, helping in scenarios where underestimation or overestimation leads to loss.
  2. It is important to choose quantile regression when predicting specific quantiles, such as upper quantiles, for scenarios like bread sales where under or overestimating can have financial impacts.
  3. Quantile regression can also be utilized for uncertainty quantification, and combining it with conformal prediction can improve coverage, making it useful for understanding and managing uncertainty in predictions.

OpenAI's moat

TechTalks • 334 implied HN points • 15 Jan 24
đź•ą Technology Modeling
  1. OpenAI is building new protections to safeguard its generative AI business from open-source models
  2. OpenAI is reinforcing network effects around ChatGPT with features like GPT Store and user engagement strategies
  3. Reducing costs and preparing for future innovations like creating their own device are part of OpenAI's strategy to maintain competitiveness

The Sequence Opinion #782: The New Gradient: Research Directions That Will Ship in 2026

TheSequence • 42 implied HN points • 01 Jan 26
đź•ą Technology Modeling
  1. Blanket scaling of transformers with more data and compute is showing diminishing returns, so new research directions are needed to keep improving frontier models.
  2. The field is shifting from generative AI that just looks right to verifiable AI that can deliberate and produce correct, auditable outputs, effectively adding a "System 2" for reasoning.
  3. Emerging methods like RLVR aim to give models unit-test-style feedback and tighter verification, and these kinds of approaches are poised to influence models shipping in 2026.

Classification and Method Selection

Abstraction • 29 implied HN points • 08 Jan 26
đź•ą Technology Modeling
  1. Match the forecasting method to the question type: classify questions into base-rate, time-series, conditional-chain, or novel-event and route each to a specialized approach.
  2. Use the right technique for each class: use historical reference classes and adjustments for base rates, simulate trajectories for time-series questions, multiply conditional probabilities for conjunctive chains, and apply a Laplace-style prior for unprecedented events.
  3. Track and improve empirically: use an LLM classifier (defaulting to base rate when unsure), choose reference classes and decompositions carefully, and measure which methods are over- or under-confident as you scale.

The Future of Prompt Engineering

Gradient Flow • 559 implied HN points • 04 May 23
đź•ą Technology Modeling
  1. NLP pipelines are shifting to include large language models (LLMs) for accuracy and user-friendliness.
  2. Effective prompt engineering is crucial for crafting useful input prompts tailored to generative AI models.
  3. Future prompt engineering tools need to be interoperable, transparent, and capable of handling diverse data types for collaboration and model sharing.

What if bigger models, like bigger stars, fail faster?

Breaking Smart • 34 implied HN points • 07 Dec 25
đź•ą Technology Modeling
  1. Larger AI models can become less reliable over time because they learn from static data that quickly becomes outdated. This means models can fail faster as they can't adapt to changes in the world.
  2. The current push for bigger models might not be sustainable if they aren't supported by enough quality training data. If companies keep investing in these models without the right data, they may end up with expensive resources that don't deliver good results.
  3. To keep AI models useful for longer, we should focus on creating new types of data, like 4D video, which can help models learn from real-world changes rather than just past cultural snapshots.

A Pragmatic View of Uncertainty in Machine Learning

Mindful Modeler • 359 implied HN points • 06 Jun 23
đź•ą Technology Modeling
  1. Machine learning models have uncertainty in predictions, categorized into aleatoric and epistemic uncertainty.
  2. Defining and distinguishing between aleatoric and epistemic uncertainty is a complex task influenced by deterministic and random factors.
  3. Conformal prediction methods capture both aleatoric and epistemic uncertainty, providing prediction intervals reflecting model uncertainty.

Physics for Everyone, Lecture 2: The Gestalt of Physics, Tools for Seeing

Critical Mass • 12 implied HN points • 22 Jan 26
🔬 Science Modeling
  1. Order-of-magnitude thinking uses powers of ten and rough estimates to keep your intuition tied to reality and avoid being fooled by big numbers with false precision.
  2. Approximation (the “super cow” idea) means building simplified models that include only the features you need so messy real-world problems become solvable without misleading yourself.
  3. Dimensional analysis tracks basic units like length, time, and mass to guess the form of answers, catch impossible results, and often derive relationships with minimal detailed information.

Defending machine learning in a room full of old-school statisticians

Mindful Modeler • 299 implied HN points • 27 Jun 23
đź•ą Technology Modeling
  1. Be mindful of your modeling mindset and be open to exploring other modeling cultures beyond your current beliefs.
  2. Recognize that differences in modeling mindsets are deeply rooted in culture and background, influencing how individuals approach statistical modeling.
  3. Interpretability remains a significant concern for modelers, especially in the context of machine learning advancements, although progress has been made in providing tools for better understanding models.

The Social Network (Part 2)

Logging the World • 279 implied HN points • 13 Apr 23
🔬 Science Modeling
  1. Real social networks exhibit more complex behaviors than simple mathematical models can capture.
  2. The structure of social media follower counts differs significantly from the Erdős–Rényi network model, with some users having exponentially more followers than others.
  3. Recent network models like the Barabási-Albert model better represent the dynamics of online social networks like Twitter, where heavy-tailed distributions of follower counts emerge.

The Statistician Who Loved Machine Learning

Mindful Modeler • 279 implied HN points • 23 May 23
đź•ą Technology Modeling
  1. Leo Breiman emphasized the importance of both data modeling culture and algorithmic modeling culture in statistical modeling.
  2. Breiman advocated for being problem-focused over solution-focused, encouraging modelers to choose the appropriate mindset based on the task at hand.
  3. Understanding various modeling mindsets, such as statistical inference and machine learning, is crucial for effective modeling.

Scaling Laws Meet Economics, but Adoption is still Accelerating

Mule’s Musings • 333 implied HN points • 19 Dec 24
đź•ą Technology Modeling
  1. Economics are very important when it comes to scaling tech, and while costs are rising, tools like ChatGPT are still becoming more popular. Understanding the balance of cost and usage is crucial.
  2. Scaling laws are changing, and relying solely on large pre-trained models may not be the best strategy anymore. Businesses might need to explore smaller models or alternative methods to improve efficiency and reduce costs.
  3. Adoption of AI technologies is still growing rapidly, which shows that despite challenges, many people are eager to use and integrate these tools into their lives.

Should we stop interpreting ML models because XAI methods are imperfect?

Mindful Modeler • 199 implied HN points • 31 Oct 23
đź•ą Technology Modeling
  1. Don't let a pursuit of perfection in interpreting ML models hinder progress. It's important to be pragmatic and make decisions even in the face of imperfect methods.
  2. Consider the balance of benefits and risks when interpreting ML models. Imperfect methods can still provide valuable insights despite their limitations.
  3. While aiming for improvements in interpretability methods, it's practical to use the existing imperfect methods that offer a net benefit in practice.

Quant Letter: January 2026, Week-3

The Parlour • 8 implied HN points • 16 Jan 26
đź’° Finance Modeling
  1. Fine-tuning LLaMA-3-8B with instruction tuning and LoRA noticeably improves financial named-entity recognition, helping convert messy reports into structured data.
  2. New work on adaptive dataflow for financial time-series points to better ways to process streaming market data and boost model efficiency or accuracy.
  3. This newsletter curates recent finance ML papers and is available by subscription, with some free previews for readers who want quick research updates.

Explore Your Modeling Mindset With A Quiz

Mindful Modeler • 179 implied HN points • 20 Jun 23
đź•ą Technology Modeling
  1. Modeling assumptions affect how the model can be used. For instance, causal considerations lead to causal claims.
  2. Revisiting and understanding our modeling assumptions can help us tackle problems more effectively, beyond our usual mindset.
  3. Creating simple static websites can be made easier with tools like GPT-4, especially if you have some understanding of HTML, CSS, and JavaScript.

Getting Started with LoRAs + Vodka Photorealism

followfox.ai’s Newsletter • 176 implied HN points • 15 Jun 23
đź•ą Technology Modeling
  1. The post discusses getting started with LoRAs and creating a photorealistic LoRA for Vodka models.
  2. It includes steps like downloading and using a LoRA, training the first LoRA, and finally fine-tuning a custom LoRA for photorealistic results.
  3. The process involves using specific tools, datasets, and parameters to train LoRAs, and explores possibilities for creating high-quality, realistic images.

Can you explain GPT with ... GPT?

Mindful Modeler • 199 implied HN points • 16 May 23
đź•ą Technology Modeling
  1. OpenAI experimented with using GPT-4 to interpret the functionality of neurons in GPT-2, showcasing a unique approach to understanding neural networks.
  2. The process involved analyzing activations for various input texts, selecting specific texts to explain neuron activations, and evaluating the accuracy of these explanations.
  3. Interpreting complex models like LLMs with other complex models, such as using GPT-4 to understand GPT-2, presents challenges but offers a method to evaluate and improve interpretability.