The hottest Machine Learning Substack posts right now

And their main takeaways
Category
Top Business Topics

Data Science Weekly - Issue 508

Data Science Weekly Newsletter • 379 implied HN points • 18 Aug 23
🕹 Technology Machine Learning
  1. Writing clear and effective research papers is essential, and there are tips specifically for NLP papers that can help improve your writing skills.
  2. The job market for data-related roles has changed over the years, and analyzing hiring trends can provide insights into what skills and positions are in demand.
  3. Understanding AI hardware is important because it forms the backbone of many AI models, and knowing how it works can help in making better tech decisions.

Five things I believe about actually-existing AI today.

The Future, Now and Then • 162 implied HN points • 16 Jul 25
🕹 Technology Machine Learning
  1. Generative AI is really about doing what's good enough for certain tasks. It's useful when perfection isn't needed, like for basic reports or planning a simple trip.
  2. The way generative AI is used often depends on the interests of investors, not users. Those making decisions may prioritize profit over quality, affecting how useful AI can be in fields like journalism and medicine.
  3. We need to be careful with how we talk about AI, as calling it 'intelligent' can lead to misunderstandings and conspiracy theories. This can have real-world consequences if people start believing silly claims.

A short history of SHAP

Mindful Modeler • 359 implied HN points • 30 May 23
🕹 Technology Machine Learning
  1. Shapley values originated in game theory in 1953 and contributed to fair resource distribution methods.
  2. In 2010, Shapley values were introduced to explain machine learning predictions, but didn't gain traction until the SHAP method in 2017.
  3. SHAP gained popularity for its new estimator for Shapley values, unification of existing methods, and efficient computation, leading to widespread adoption in machine learning interpretation.

Phi-3 Is A Small Language Model Which Can Run On Your Phone

Cobus Greyling on LLMs, NLU, NLP, chatbots & voicebots • 39 implied HN points • 19 Jun 24
🕹 Technology Machine Learning
  1. Phi-3 is a small language model that can run directly on your phone, making it accessible for local use instead of needing cloud connections. This means you can use it anywhere without relying on internet speed.
  2. Small language models like Phi-3 are good for specific tasks and regulated industries where data privacy is important. They can provide quick and accurate responses while keeping your data secure.
  3. Training for Phi-3 involves using high-quality data to improve its understanding of language and reasoning skills, allowing it to perform well on par with larger models, despite its smaller size.

Data Science Weekly - Issue 506

Data Science Weekly Newsletter • 399 implied HN points • 04 Aug 23
🕹 Technology Machine Learning
  1. Integrating large language models into systems can be done using seven key patterns that balance performance and cost.
  2. Ethics in AI isn't just about explainability and fairness; we need a deeper understanding to prevent overall harm from AI systems.
  3. New approaches in robotics focus on current challenges and opportunities while advancing understanding of AI's role in planning tasks.
Get a weekly roundup of the best Substack posts, by hacker news affinity:

Reflexes, Desires, Actions, Emergence, & the Mirage of Machine Minds: What MAMLMs Still Ain't

Brad DeLong's Grasping Reality • 176 implied HN points • 29 Jun 25
🕹 Technology Machine Learning
  1. Understanding complexity and emergence is crucial for grasping advanced artificial intelligence concepts. It's not just about scaling up technology but comprehending how simple rules can create complex behaviors.
  2. Human intelligence is a result of both evolution and shared knowledge as a species. We are already a network of minds working together, which influences how we create and interact with machines.
  3. The future of AI should focus on enhancing human capabilities rather than mimicking intelligence. We need to consider if we're creating true understanding or just sophisticated imitation.

The Large Language Model Landscape — Version 5

Cobus Greyling on LLMs, NLU, NLP, chatbots & voicebots • 79 implied HN points • 25 Apr 24
🕹 Technology Machine Learning
  1. Large Language Models (LLMs) are evolving with more functionality, combining various tasks into fewer models. This helps in making them more efficient for users.
  2. There are different zones in the LLM landscape, each focusing on specific uses, tools, and applications, ranging from available models to user interfaces.
  3. Tech advancements like prompt engineering and data-centric tools are making it easier to harness the power of LLMs, opening up new opportunities for businesses.

So... what is the lottery ticket hypothesis [Math Mondays]

Technology Made Simple • 159 implied HN points • 05 Feb 24
🕹 Technology Machine Learning
  1. The Lottery Ticket Hypothesis proposes that within deep neural networks, there are subnetworks capable of achieving high performance with fewer parameters, leading to smaller and faster models.
  2. Successful application of the Lottery Ticket Hypothesis relies on iterative magnitude pruning strategies, with potential benefits like faster learning and higher accuracy.
  3. The hypothesis works due to factors like favorable gradients, implicit regularization, and data alignment, but challenges like scalability and interpretability remain towards practical implementation.

The Case for Uninterpretable Machine Learning

Mindful Modeler • 319 implied HN points • 03 Oct 23
🕹 Technology Machine Learning
  1. Machine learning excels because it's not interpretable, not in spite of it.
  2. Embracing complexity in models like neural networks can effectively capture the intricacies of real-world tasks that lack simple rules or semantics.
  3. Interpretable models can outperform complex ones with smaller datasets and ease of debugging, but being open to complex models can lead to better performance.

The Latent State Window and the Limits of Reasoning in LLMs

jonstokes.com • 164 implied HN points • 05 Jul 25
🕹 Technology Machine Learning
  1. LLMs have limits when it comes to reasoning. If a problem is too complex or involves too many moving parts, the model can struggle to find a solution.
  2. The size of a language model's 'latent state window' matters. This window limits how much information the model can hold while trying to reason, separating it from just the number of tokens it can handle.
  3. To get good results from LLMs, it's best to keep tasks simple and broken down into manageable pieces. If you give the model too much to juggle at once, it won't perform well.

Google announces AI system for diagnostic medical reasoning and conversation

MLOps Newsletter • 176 implied HN points • 20 Jan 24
🕹 Technology Machine Learning
  1. Google announced an AI system for medical diagnosis and conversation called AMIE.
  2. AMIE's architecture includes multi-turn dialogue management, hierarchical reasoning model, and modular design.
  3. The AI system AMIE showed promising performance in simulated diagnostic conversations, outperforming PCPs and matching specialist physicians.

China’s GenAI Content Security Standard: An Explainer

ChinaTalk • 429 implied HN points • 07 Jan 25
🕹 Technology Machine Learning
  1. China has set rules for generative AI to ensure the content it produces is safe and follows government guidelines. This means companies need to be careful about what their AI apps say and share.
  2. Developers of AI must check their data and the output carefully to avoid politically sensitive issues, as avoiding censorship is a key focus of these rules. They have to submit thorough documentation showing they comply with these standards.
  3. While these standards are not legally binding, companies often follow them closely because government inspections are strict. These regulations mainly aim at controlling politically sensitive content.

Exploratory Data Visualization With GPT-4: Prompting for Python Folium Maps

Data at Depth • 79 implied HN points • 23 Apr 24
🕹 Technology Machine Learning
  1. GPT-4 can create choropleth and heatmaps from datasets if you know the right questions to ask
  2. Integrating GPT-4 into data visualization workflows can be beneficial for exploration and learning new libraries such as Python folium
  3. GPT-4 can be used to enhance code generation for data visualization projects by providing responses and solutions to specific coding challenges

LLMs - Part 2: Order Matters - Positional Encoding

Vasu’s Newsletter • 13 implied HN points • 11 Jan 26
🕹 Technology Machine Learning
  1. Large language models process tokens in parallel and need positional encoding to know word order; without it, reordered sentences look the same to the model.
  2. Positional encodings (like sinusoidal functions or methods such as RoPE and ALiBi) give each position a unique vector that’s combined with token embeddings, so the same word at different positions produces different vectors and relative distances can be inferred.
  3. Positional encoding only makes order visible — it doesn’t compute relationships or context; deciding which words matter to each other is handled next by self-attention.

LangGraph From LangChain Explained In Simple Terms

Cobus Greyling on LLMs, NLU, NLP, chatbots & voicebots • 39 implied HN points • 17 Jun 24
🕹 Technology Machine Learning
  1. LangGraph helps create clearer conversations by using graphs to map out how dialog flows between different points, making it easier to manage conversations in AI systems.
  2. Prompt chaining connects smaller tasks in a sequence, allowing AI models to handle complex jobs step by step, but can feel rigid like traditional chatbots.
  3. Autonomous Agents bring a higher level of flexibility in how actions are taken, but they can also lead to concerns about having enough control over their decision-making process.

How I Taught Myself to Critique AI Research [Storytime Saturdays]

Technology Made Simple • 119 implied HN points • 10 Mar 24
🕹 Technology Machine Learning
  1. Writing allows you to store knowledge for future reference, spot cognitive blindspots, and engage with topics more deeply for better understanding.
  2. Challenges in self-learning writing include lack of contextual understanding, a defined learning path, and a peer network for feedback.
  3. Addressing challenges in self-learning involves finding strategies to gain clarity, identifying knowledge gaps, and seeking feedback from peers.

The ArXiv Problem

Gonzo ML • 189 implied HN points • 19 Jun 25
🕹 Technology Machine Learning
  1. Many people struggle to keep up with the overwhelming number of research papers being published, which leads to frustration and unread lists.
  2. ArXivIQ is a tool designed to help curate and summarize papers in a quicker way, providing 15-minute reads instead of lengthy sessions.
  3. The author emphasizes transparency in using AI to assist with research, acknowledging that it's unrealistic for anyone to read every important paper.

How to Use AI to Do Stuff: An Opinionated Guide

One Useful Thing • 1801 implied HN points • 15 Jul 23
🕹 Technology Machine Learning
  1. Increasingly powerful AI systems are being released rapidly without proper user documentation.
  2. The major Large Language Models in use currently are GPT-3.5, GPT-4, Bard, Pi, and Claude 2.
  3. AI can assist with writing, generating images, coming up with ideas, making videos, and working with documents and data, but users must be cautious of biases and ethical concerns.

The Sequence Radar #688: The Transparent Transformer: Monitoring AI Reasoning Before It Goes Rogue

TheSequence • 154 implied HN points • 20 Jul 25
🕹 Technology Machine Learning
  1. AI researchers are exploring a way to monitor advanced AI reasoning to catch any dangerous behavior early. This method looks at how AI models 'think' through problems using something called chains of thought.
  2. This monitoring method is helpful but can be fragile. As AI models get better, they might stop using natural language reasoning, making it harder to understand their thought processes.
  3. There is a big push for more research to keep this monitoring effective. By establishing clear benchmarks, we can better evaluate and improve how we observe AI reasoning.

Data Science Weekly - Issue 516

Data Science Weekly Newsletter • 299 implied HN points • 13 Oct 23
🕹 Technology Machine Learning
  1. The newsletter is deciding whether to publish twice a week, but will stick to one issue for now to review feedback from readers.
  2. There's a focus on providing useful resources for data science, including articles and job opportunities in the field.
  3. New tools and methods in AI and data engineering are highlighted, addressing challenges like data integration and AI model training.

AI Unwrapped #2 | Feb 10 – Feb 24, 2026

The Product Channel By Sid Saladi • 3 implied HN points • 27 Feb 26
🕹 Technology Machine Learning
  1. Google’s Gemini 3.1 Pro reclaimed the lead with a major reasoning jump and top benchmark scores while keeping the same API pricing, making it far stronger for logic, coding, and multimodal tasks.
  2. AI capabilities are expanding fast — models now solve PhD-level science problems, generate music from images, find long-hidden security bugs, and power new agent platforms and browser/assistant integrations.
  3. If you build products, test these new models on your hardest multi-step problems and add AI-powered checks like security reviews, because the recent reasoning gains can materially change outcomes.

Data Science Weekly - Issue 511

Data Science Weekly Newsletter • 319 implied HN points • 07 Sep 23
🕹 Technology Machine Learning
  1. AI startups can receive significant support through programs like AI Grant, offering up to $250,000 for development.
  2. Recent studies have shown that large language models can learn from just one example, which challenges previous beliefs about their efficiency.
  3. Using advanced tools like the Semantic Layer and LLMs can greatly improve data accuracy and speed for businesses, making analytics much easier.

Data Science Weekly - Issue 515

Data Science Weekly Newsletter • 299 implied HN points • 06 Oct 23
🕹 Technology Machine Learning
  1. There's a lot happening in data science right now. The team is considering adding a second newsletter each week to cover more exciting content.
  2. High-performing data scientists have specific traits that set them apart from others. Companies are researching these traits to help improve their teams.
  3. Art institutions can greatly benefit from data and analytics. Collaborating with leaders can help them use data to improve their operations and strategies.

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

Mindful Modeler • 299 implied HN points • 27 Jun 23
🕹 Technology Machine Learning
  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.

WebVoyager AI Agent

Cobus Greyling on LLMs, NLU, NLP, chatbots & voicebots • 19 implied HN points • 23 Jul 24
🕹 Technology Machine Learning
  1. AI agents can make their own choices and decide how to reach a goal. They don’t just follow a set plan; they create their own steps as needed.
  2. These agents can try different actions and learn from the results until they find the right answer. They go through a thinking process to solve problems.
  3. While AI agents have some tools to use, they also have limits. If they can't find an answer after trying a few times, they might ask a human for help.

A Great Interruption Begins

From The Future • 294 implied HN points • 17 Jul 23
🕹 Technology Machine Learning
  1. A new era called The Great Interruption seems to be approaching, freeing us from the constant distractions and reclaiming our attention
  2. Information and attention have always been intertwined, impacting humanity's growth and evolution
  3. Interest Grids could be the future of organizing our attention, helping us navigate a world inundated with information and distractions

ChatGPT takes the Big Five Inventory

Vectors of Mind • 294 implied HN points • 27 Mar 23
🕹 Technology Machine Learning
  1. A language model like ChatGPT can take personality tests like the Big Five Inventory.
  2. ChatGPT's personality leans towards being conscientious and non-neurotic.
  3. It's fascinating how language models like ChatGPT can generate responses to personality test questions based on their programming and training.

SAI #22: Decomposing the Data System.

SwirlAI Newsletter • 294 implied HN points • 18 Mar 23
🕹 Technology Machine Learning
  1. Learning to decompose a data system is crucial for better reasoning and understanding of large infrastructure
  2. Decomposing a data system allows for scalability, identification of bottlenecks, and total event processing latency optimization
  3. The different layers in a data system include data ingestion, transformation, and serving layers, each with specific functions and technologies

Use Interpretability to Justify Your Model

Mindful Modeler • 199 implied HN points • 19 Dec 23
🕹 Technology Machine Learning
  1. Performance of a machine learning model is not always enough to justify its use; interpretability is crucial for justification.
  2. Interpretability plays a key role in justifying a model by making people trust the model and its predictions.
  3. Different interpretation approaches may be needed for justifying models to different audiences and contexts, understanding the roles of creators, operators, executors, decision-subjects, and examiners.

Data Science Weekly - Issue 512

Data Science Weekly Newsletter • 299 implied HN points • 14 Sep 23
🕹 Technology Machine Learning
  1. Nvidia has been a leader in AI technology, but its dominance might not last. Changes in the market and technology could shift the competitive landscape soon.
  2. For those who know R and want to learn Python, there are resources available to help make the transition easier. These resources provide advice and tips catered to R users.
  3. Reinforcement Learning with Human Feedback (RLHF) is an important part of training large language models. It's essential for improving how these models understand and respond to human preferences.

The Sequence Opinion #774: Everything You Need to Know About Audio AI Frontier Models

TheSequence • 28 implied HN points • 18 Dec 25
🕹 Technology Machine Learning
  1. Audio is a major next frontier in AI, with models now able to hear, understand, and generate speech, music, and environmental sounds at near-human levels.
  2. Audio is fundamentally different from text and images because it's a continuous, high-frequency time-series that requires modeling very long sequences and both short-term details (like phonemes or notes) and long-term structure (like phrases or whole melodies).
  3. Development is happening across open-source and commercial players, and a central debate is whether to build general multimodal systems that include audio or to focus on specialized audio models tuned for sound-specific challenges.

The Sequence Knowledge #752: Understanding the Different Types of Synthetic Data Generation Techniques

TheSequence • 49 implied HN points • 11 Nov 25
🕹 Technology Machine Learning
  1. Synthetic data generation involves methods to create data that can be used for training models. It's important that this data is true to real-life scenarios and diverse enough to cover different tasks.
  2. A good synthetic data process combines real examples with transformations to improve coverage and quality. This way, it can create stronger data by getting better labels and avoiding duplicates.
  3. The effectiveness of synthetic data also depends on being able to guide and control the specific types of data it generates. This helps make sure the data fits the intended purpose and remains high quality.

Sora What

Don't Worry About the Vase • 1075 implied HN points • 22 Feb 24
🕹 Technology Machine Learning
  1. OpenAI's new video generation model Sora is technically impressive, achieved through massive compute and attention to detail.
  2. The practical applications of Sora for creating watchable content seem limited for now, especially in terms of generating specific results as opposed to general outputs.
  3. The future of AI-generated video content may revolutionize industries like advertising and media, but the gap between generating open-ended content and specific results is a significant challenge to overcome.

Inside AI

John Ball inside AI • 39 implied HN points • 12 Jun 24
🕹 Technology Machine Learning
  1. AGI might not come from current machine learning methods. Instead, understanding how human brains work could be the key to achieving it.
  2. The theory behind brain functions can help solve AI challenges. Learning from how brains process information could lead us to better AI solutions.
  3. Language is crucial for interacting with AI. Building a trustworthy AI community focused on language can improve how we communicate and use technology.

Securing AI: Addressing the Emerging Threat of Prompt Injection

Gradient Flow • 219 implied HN points • 30 Nov 23
🕹 Technology Machine Learning
  1. Prompt injection is a critical threat to AI systems, manipulating model outputs for harmful outcomes.
  2. Mitigating prompt injection risks requires a multi-layered defense approach involving prevention, detection, and response strategies.
  3. Collaboration between security, data science, and engineering teams is essential to secure AI systems against evolving threats like prompt injection.

The Sequence AI of the Week #773: The Week Google Turned Gemini Into an Agent Runtime

TheSequence • 28 implied HN points • 17 Dec 25
🕹 Technology Machine Learning
  1. Google moved from just releasing models to shipping an agent runtime that coordinates and runs agents, making Gemini a platform for agent workflows.
  2. The Interactions API (Beta) and the Gemini Deep Research Agent (Preview) were released together, signaling a deliberate architectural pivot and providing both the runtime and a managed agent that uses it.
  3. Real agent systems are stateful, tool-heavy, and long-running, so most engineering effort goes into planners, tool routing, memory, retries, auditing, and UIs — the LLM call itself is the smallest piece.

What A Horse Can Tell Us About Machine Learning

Mindful Modeler • 279 implied HN points • 10 Oct 23
🕹 Technology Machine Learning
  1. Animals like horses and machines can appear clever by relying on cues and shortcuts, rather than true understanding.
  2. When designing or evaluating machine learning models, watch out for 'Clever Hans Predictors' that rely on spurious correlations.
  3. To spot potential Clever Hans Predictors, look for unexpectedly good model performance, apply causal thinking, examine data closely, and use interpretation methods to investigate model behavior.

8 Insights to Make Sense of OpenAI o3

The Algorithmic Bridge • 424 implied HN points • 23 Dec 24
🕹 Technology Machine Learning
  1. OpenAI's new model, o3, has demonstrated impressive abilities in math, coding, and science, surpassing even specialists. This is a rare and significant leap in AI capability.
  2. There are many questions about the implications of o3, including its impact on jobs and AI accessibility. Understanding these questions is crucial for navigating the future of AI.
  3. The landscape of AI is shifting, with some competitors likely to catch up, while many will struggle. It's important to stay informed to see where things are headed.