The hottest Machine Learning Substack posts right now

And their main takeaways
Category
Top Business Topics

January Newsletter

RSS DS+AI Section • 11 implied HN points • 01 Jan 26
🕹 Technology Machine Learning
  1. AI and large language models are advancing rapidly, with major companies and open-source projects pushing innovations in long-context reasoning, memory, and generative capabilities. Competition is driving frequent releases and new research on foundation models and video/world-models.
  2. Ethics, bias, interpretability, and regulation remain central concerns as real-world uses expand, prompting debates, lawsuits, and calls for better safety research. Work on interpretability is seen as especially important for progressing AI more safely.
  3. The community is focusing on practical adoption and professionalisation through tutorials, production tips, projects, workshops, a new journal, and competency frameworks. There are also learning opportunities, internships, and calls for volunteers to help shape best practices and careers.

3 Years of ChatGPT

Olshansky's Newsletter • 22 implied HN points • 03 Dec 25
🕹 Technology Machine Learning
  1. AI is already here as an amplifier of human intelligence and is being used daily across personal and professional tasks; agent-driven tools have massively increased productivity, especially for coding.
  2. High-quality, unique data and expert-labeled "golden" datasets are the most valuable assets for building useful AI systems; simple benchmarks and naive fine-tuning are limited, while reinforcement fine-tuning and dedicated context engineering will drive real gains.
  3. Practical changes are coming in the next few years: local inference stations, agentic e-commerce, consolidation of tooling, and new roles like context engineers and AI bootcamps; foundational roles like architects will remain and superintelligence isn’t expected soon.

Hybrid Evaluation: Scaling human feedback with custom evaluation models

LLMs for Engineers • 159 implied HN points • 15 Nov 23
🕹 Technology Machine Learning
  1. Human feedback is still very important for evaluating models, especially in areas like customer support, but it can slow things down and increase costs.
  2. Combining human input with automated, model-based evaluation can help improve efficiency and accuracy, reducing errors significantly.
  3. Using fewer human-labeled examples with smart bootstrapping techniques can still yield good results, making it cheaper and faster to train evaluation models.

GroupBy #29: Scaling AI/ML Infrastructure at Uber, The Sisyphean struggle and the new era of data infrastructure

VuTrinh. • 59 implied HN points • 02 Apr 24
🕹 Technology Machine Learning
  1. Uber is focusing on building strong AI and machine learning infrastructure to keep up with the growing complexity of their models. This involves using both CPUs and GPUs for better efficiency.
  2. Data management is becoming crucial for companies like Netflix as they deal with massive amounts of production data. They are developing tools to effectively manage and optimize this data.
  3. The data streaming landscape is evolving, with new technologies emerging that make handling data easier and more efficient. This is changing how companies approach data infrastructure.
Get a weekly roundup of the best Substack posts, by hacker news affinity:

How to metacognate

Sunday Letters • 99 implied HN points • 29 Jan 24
🕹 Technology Machine Learning
  1. Working with complex models can be hard when they get confused by incorrect or incomplete information. This can lead to mistakes and conflicts in what they remember.
  2. Creating a stable pattern for how tasks are done can help models work better by giving them a solid structure to follow. This is like giving the model a framework to lean on for more complicated tasks.
  3. As models improve, the need for extra coding to guide their thinking may lessen. Better memory strategies will likely help them function more effectively over time.

The Sequence Radar #679: From Model to Team: Several Models are Better than One: Sakana’s Blueprint for Collective AI

TheSequence • 105 implied HN points • 06 Jul 25
🕹 Technology Machine Learning
  1. Sakana AI has a new way to use multiple models together for better AI performance. Instead of relying on one model, they combine many to think more like humans.
  2. Their approach, called AB-MCTS, helps the AI decide whether to explore new ideas or improve current ones. This makes the AI smarter and more flexible in how it solves problems.
  3. By using several models that learn from past tasks, this system can better handle different challenges. This means AI can become more reliable and efficient in real-life applications.

Technically Monthly (January 2026)

Technically • 12 implied HN points • 06 Jan 26
🕹 Technology Machine Learning
  1. Try multiple vibe-coding tools by building the same thing so you learn their quirks, limits, and pricing before committing.
  2. Monitor AI with simple evals: study failures, use straightforward assertions instead of AI-judging-AI, and follow a loop of vibe check, spreadsheet, fixes, then targeted tests to cut hallucinations.
  3. Use AI thoughtfully at work by customizing prompts and iterating on workflows; learn prompt engineering or you risk being outcompeted by careless automation.

Week #2: Intuition Behind Conformal Prediction

Mindful Modeler • 379 implied HN points • 27 Dec 22
🔬 Science Machine Learning
  1. Conformal prediction for classification works by ordering predictions from certain to uncertain, dividing them based on a user-defined confidence level.
  2. Conformal prediction consists of three main steps: training, calibration, and prediction, following a similar recipe across different algorithms.
  3. Different resampling strategies like k-fold cross-splitting and jackknife are used in conformal prediction, offering a balance between computation cost and prediction accuracy.

Proxy Fine-Tuning LLMs

Cobus Greyling on LLMs, NLU, NLP, chatbots & voicebots • 79 implied HN points • 26 Feb 24
🕹 Technology Machine Learning
  1. Proxy fine-tuning lets you improve a language model's performance without changing its internal settings. It only uses the model's output to make adjustments.
  2. Combining different approaches, like retrieval and fine-tuning, can lead to better results with language models. It's about using the best methods together instead of relying on just one.
  3. Using proxy fine-tuning can help organizations better understand and organize their data. It encourages them to explore their information needs more deeply.

The Sequence Knowledge #768: Using Rephrasing for Synthetic Data Generation

TheSequence • 21 implied HN points • 09 Dec 25
🕹 Technology Machine Learning
  1. Different rephrasing methods can vary in quality when generating synthetic data. It's important to choose the right method for effective results.
  2. Microsoft's Evol-Instruct is a sophisticated way to create instruction datasets that can enhance AI performance.
  3. Rephrasing helps expand datasets by creating new variants while keeping the original meaning, making it a useful tool for improving coverage and reliability.

LLM Links, 12/1

In My Tribe • 288 implied HN points • 01 Dec 24
🕹 Technology Machine Learning
  1. AI systems are being developed to have better memory which would improve conversations with users. If they can remember past interactions, it could lead to more meaningful and deeper exchanges.
  2. Humans have unique qualities like vulnerability and connection that AI can't replicate. This means people will still value human interactions over machines, no matter how advanced they become.
  3. Virtual friends powered by AI can help those who are lonely, but they might also distract from real-life relationships. It's important to balance technology use with human connections.

A Short History Of Chatbots

Cobus Greyling on LLMs, NLU, NLP, chatbots & voicebots • 39 implied HN points • 09 May 24
🕹 Technology Machine Learning
  1. Chatbots have changed a lot over time, starting as simple rule-based systems and moving to advanced AI models that can understand context and user intent.
  2. Early chatbots used basic pattern recognition to respond to user questions, but this method was limited and often resulted in repetitive and predictable answers.
  3. Now, modern chatbots utilize natural language understanding and machine learning to provide more dynamic and relevant responses, making them better at handling various conversations.

Data Science Weekly - Issue 494

Data Science Weekly Newsletter • 319 implied HN points • 12 May 23
🕹 Technology Machine Learning
  1. Open source AI is rapidly advancing, but may always lag behind the best quality models. It's great for innovation but has its limits.
  2. Many academic papers promise data sharing but often fail to deliver, which can hinder scientific research and verification.
  3. Understanding how to craft effective prompts is essential when using generative AI tools. This skill can greatly enhance the results you get from those tools.

Feature Selection And Feature Importance: How Are They Related?

Mindful Modeler • 299 implied HN points • 28 Feb 23
🕹 Technology Machine Learning
  1. Feature selection and feature importance are different steps in modeling with different goals, but they are complementary. Getting feature selection right can enhance interpretability.
  2. Feature selection aims to reduce the number of features used in the model to improve predictive performance, speed up training, enhance comprehensibility, and reduce costs.
  3. Feature importance involves ranking and quantifying the contribution of features to model predictions, aiding in understanding model behavior, auditing, debugging, feature engineering, and comprehending the modeled phenomenon.

AGI Emergence: Tracing the Path from Artificial Narrow Intelligence to the Birth of Superintelligence

The Future of Life • 39 implied HN points • 08 May 24
🕹 Technology Machine Learning
  1. AI is evolving through different levels, starting from basic text generation to more advanced reasoning and problem-solving abilities.
  2. As AI develops, it will be able to perform tasks across various domains, becoming competitive with humans in many jobs.
  3. Eventually, AI may reach a point of superintelligence, where it surpasses human understanding and decision-making abilities, posing potential risks if not aligned with human values.

The Sequence AI of the Week #757: 3D World Models in Action: Inside DeepMind’s SIMA 2 Architecture

TheSequence • 28 implied HN points • 19 Nov 25
🕹 Technology Machine Learning
  1. DeepMind's SIMA 2 can create and interact with 3D environments, making it a big step for AI in gaming. It's like giving a computer the ability to play and learn just like humans do.
  2. This AI uses a smart mix of different models to see, think, and act in these virtual worlds, similar to how people play games. It helps the AI improve itself by practicing and trying out different tasks.
  3. SIMA 2 shows how we can build complex AI systems that work together, rather than developing them one piece at a time. This could change how we design future AI technologies.

I, Cyborg: Using Co-Intelligence

One Useful Thing • 650 implied HN points • 14 Mar 24
🕹 Technology Machine Learning
  1. AI can be a powerful tool in writing and reading, enhancing the process by providing options and guidance without replacing human creativity.
  2. Authors can use AI as Cyborgs or Centaurs, blending human and machine efforts to optimize their work in writing and analysis tasks.
  3. AI continues to advance rapidly, with models like GPT-4 showcasing impressive writing capabilities, indicating a future where AI may play an even larger role in book creation.

Could A.I. Make Us More Human?

The Digital Anthropologist • 19 implied HN points • 28 Jun 24
🕹 Technology Machine Learning
  1. Artificial Intelligence (AI) might actually help make us more human, sparking an intriguing perspective to consider.
  2. The advancements in AI tools like Machine Learning and Natural Language Processing are already being used in various fields including healthcare and environmental research.
  3. Rethinking human exceptionalism and embracing the potential for AI to facilitate communication with animals and nature could lead to significant shifts in societal norms and behaviors.

China's Deepseek is NOT as smart as ChatGPT-o1

Maximum Truth • 231 implied HN points • 29 Jan 25
🕹 Technology Machine Learning
  1. Deepseek performs on par with free AI models but does not reach the intelligence of OpenAI's paid models. It can exceed or match free AIs like Claude and ChatGPT-4o, but falls short against the more advanced paid versions.
  2. When tested with IQ questions only found offline, Deepseek does better than free models but still trails behind OpenAI’s paid models. Its results imply it may have leveraged internet data for online IQ tests, thus affecting its offline performance.
  3. Despite being competitive, the US maintains a lead in AI intelligence. Deepseek shows promise but faces challenges ahead, especially with the restrictions on technology that China experiences.

Data Science Weekly - Issue 504

Data Science Weekly Newsletter • 239 implied HN points • 21 Jul 23
🕹 Technology Machine Learning
  1. AI companies are complicated and must consider many factors like research, funding, and competition. Understanding these can help predict how they might evolve in the future.
  2. Debriefs, or team discussions after projects, can greatly boost team performance. They help everyone learn from experiences and improve future collaboration.
  3. New research shows that specific ingredient pairings in food can be explained by flavor networks. This indicates there are universal patterns in how different foods complement each other.

Orange peels, human tests, and LLMs

The Counterfactual • 139 implied HN points • 28 Nov 23
🕹 Technology Machine Learning
  1. It's tricky to know what Large Language Models (LLMs) can really do. Figuring out how to measure their skills, like reasoning, is more complicated than it seems.
  2. Using tests designed for humans might not always work for LLMs. Just because a test is good for people doesn't mean it measures the same things for AI.
  3. We need to look deeper into how LLMs solve tasks, not just focus on their test scores. Understanding their inner workings could help us assess their true capabilities better.

Data Science Weekly - Issue 493

Data Science Weekly Newsletter • 319 implied HN points • 05 May 23
🕹 Technology Machine Learning
  1. Data scientists often lack key skills needed for the job, which can be frustrating for those hiring. It's important for data scientists to continually improve their skills and adapt to job requirements.
  2. There's a significant increase in data downtime and resolution times, signaling that overall data quality management needs improvement. Companies should focus on better data practices to enhance their operations.
  3. New programming languages, like Mojo, are emerging that aim to simplify coding and enhance user experience. These advancements can make programming more accessible and enjoyable for everyone.

Looking at the Datastructures- Trees [Math Mondays]

Technology Made Simple • 179 implied HN points • 18 Jul 23
🕹 Technology Machine Learning
  1. Trees are powerful data structures that are great for efficient organization and retrieval of data in software engineering.
  2. Recursion works well with trees due to their recursive substructure, making implementation of recursive functions easier.
  3. Decision trees in AI excel at discerning complex patterns, providing interpretable results, and are versatile in various domains such as finance, healthcare, and marketing.

Overview of the AI landscape

Software Engineering Tidbits • 98 implied HN points • 22 Jan 24
🕹 Technology Machine Learning
  1. Large Language Models (LLMs) are key in AI applications like OpenAI's ChatGPT and Anthropic's Claude.
  2. Vector databases and embeddings help understand word associations, with tools like Pinecone and the Embedding Projector by TensorFlow.
  3. Tooling in AI is advancing, with Vellum for versioning prompts and Not Diamond for routing prompts for optimal model response.

GroupBy #28: Tableflow - The Stream/Table, Kafka/Iceberg Duality, Kafka tiered storage deep dive

VuTrinh. • 59 implied HN points • 26 Mar 24
🕹 Technology Machine Learning
  1. Tableflow allows you to easily turn Apache Kafka topics into Iceberg tables, which could change how streaming data is managed.
  2. Kafka's new tiered storage feature helps separate compute and storage, making it easier to manage resources and keep systems running smoothly.
  3. Data governance is important but can be lackluster if it doesn't show clear business benefits, making us rethink its role in today's data landscape.

Explore Your Modeling Mindset With A Quiz

Mindful Modeler • 179 implied HN points • 20 Jun 23
🕹 Technology Machine Learning
  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.

The Sequence Knowlege #693: A New Series About Interpretability in Foundation Models

TheSequence • 84 implied HN points • 29 Jul 25
🕹 Technology Machine Learning
  1. Understanding AI black boxes, especially complex models, is very important for safety and trust. People need to know how these AIs make decisions.
  2. Interpretability in AI refers to making sense of how these intelligent systems work. It's about bridging the gap between what we can do with AI and understanding it.
  3. The series will discuss practical ways to interpret these AI models and review significant papers related to the topic. Learning from research is key to improving AI understanding.

Teaching Language Models to use Tools

Deep (Learning) Focus • 176 implied HN points • 29 May 23
🕹 Technology Machine Learning
  1. Teaching LLMs to use tools can help them overcome limitations like arithmetic mistakes, lack of current information, and difficulty with understanding time.
  2. Giving LLMs access to external tools can make them more capable in solving complex tasks by delegating subtasks to specialized tools.
  3. Different forms of learning for LLMs include pre-training, fine-tuning, and in-context learning, which all contribute to enhancing the model's performance and capability.

Enterprises Need RAG, Not Fine-Tuning.

Sector 6 | The Newsletter of AIM • 19 implied HN points • 26 Jun 24
🕹 Technology Machine Learning
  1. Retrieval Augmented Generation (RAG) is more effective than fine-tuning for enterprises. It connects to external data sources, making it easier to get accurate information.
  2. Using RAG helps reduce hallucinations in language models, which means the outputs are more reliable and trustworthy.
  3. Enterprises can maintain better control over their information by using RAG, ensuring relevant and precise responses.

The Sequence Opinion #672: Mind Over Model: Chain-of-Thought vs. System 1/System 2

TheSequence • 105 implied HN points • 26 Jun 25
🕹 Technology Machine Learning
  1. Chain-of-thought reasoning in AI helps it to process and structure information more clearly. This is similar to how humans take time to think through problems rather than jumping to conclusions.
  2. Human thought has two systems: System 1, which is quick and instinctive, and System 2, which is slower and more deliberate. This comparison helps us understand AI reasoning better.
  3. Understanding the similarities and differences between AI reasoning and human cognition can give us insights into how to improve AI systems in the future. It's important to keep exploring these connections.

Why Claude Can't Run Your Business (Yet)

Teaching computers how to talk • 99 implied HN points • 30 Jun 25
🕹 Technology Machine Learning
  1. Claude, the AI, was tested to see if it could manage a vending machine successfully. It had to figure out pricing and deal with customer feedback.
  2. The experiment showed that Claude struggled with basic business decisions, like buying items it couldn't sell for a profit. It also made strange comments that confused the human employees.
  3. Overall, the project highlighted how current AI technology, like Claude, isn't ready to run a business effectively yet, mainly because it can't learn from its mistakes.

Data Science Weekly - Issue 486

Data Science Weekly Newsletter • 359 implied HN points • 17 Mar 23
🕹 Technology Machine Learning
  1. AI and data science are evolving rapidly, making it challenging for many to keep up. It's common for professionals to feel overwhelmed as they try to understand new advancements.
  2. There's a growing discussion about whether we should slow down AI development. Some people believe we need to pause and figure out the implications of current technologies before moving forward.
  3. Many professionals are exploring career shifts between data science and data engineering. It's important to consider personal interests and skills when deciding which path to take.

Can Conversation Designers Excel As Data Designers?

Cobus Greyling on LLMs, NLU, NLP, chatbots & voicebots • 19 implied HN points • 24 Jun 24
🕹 Technology Machine Learning
  1. Conversation designers can play a key role in creating and improving datasets for training language models. Their skills can help make data more relevant and useful.
  2. Techniques like Partial Answer Masking and Prompt Erasure help models learn to self-correct and think strategically. This makes them better at reasoning and understanding complex tasks.
  3. Chain-of-Thought methods help language models break down problems into smaller steps. This approach can lead to more accurate and reliable answers.

Latest open artifacts (#11): Visualizing China's open models market share, Arcee's models, and VLAs for robotics

Democratizing Automation • 95 implied HN points • 26 Jun 25
🕹 Technology Machine Learning
  1. Chinese models are leading the open model market, significantly influencing developments with their high-performance releases and generous licensing.
  2. A mix of new model releases and datasets is coming out, which includes openly licensed resources that set a good precedent for future open-source projects.
  3. There's a growing trend of models incorporating reasoning and retrieval capabilities, showing progress in AI's abilities and offering new tools for developers.

Data Science Weekly - Issue 565

Data Science Weekly Newsletter • 1 HN point • 19 Sep 24
🕹 Technology Machine Learning
  1. Reading The Data Science Weekly is a great way to stay updated on AI and machine learning topics. It shares links, news, and resources that can help anyone interested in these fields.
  2. There are many useful techniques in data science, like the Hampel Filter for outlier detection, which can help improve data quality. Exploring these methods can really enhance your understanding and skills.
  3. Effective communication is crucial in data science. How you explain your findings can significantly impact your career, so it's important to work on your communication skills.

LLM Links, 11/21

In My Tribe • 273 implied HN points • 21 Nov 24
🕹 Technology Machine Learning
  1. There's a debate about AI progress. Some experts think AI models are hitting a limit and may not get much smarter, while others believe we will continue to see significant advancements.
  2. While machine learning can learn from explicit knowledge, it struggles with understanding deeper, unspoken human knowledge. This limitation might prevent AI from reaching the same expertise as human experts.
  3. AI technologies are still showing exciting developments, like robots learning to perform surgeries by watching videos. This points to the potential for AI to revolutionize fields like medicine.

Dissecting OLMo, The Most Open Source LLM Paper!

Aziz et al. Paper Summaries • 79 implied HN points • 06 Mar 24
🕹 Technology Machine Learning
  1. OLMo is a fully open-source language model. This means anyone can see how it was built and can replicate its results.
  2. The OLMo framework includes everything needed for training, like data, model design, and training methods. This helps new researchers understand the whole process.
  3. The evaluation of OLMo shows it can compete well with other models on various tasks, highlighting its effectiveness in natural language processing.