The hottest Algorithms Substack posts right now

And their main takeaways
Category
Top Technology Topics

Save us, Digital Cronkite!

Noahpinion • 19294 implied HN points • 19 Mar 26
đź•ą Technology Algorithms
  1. Social media rewards loud, negative, attention-seeking people, which amplifies divisive content and polarizes public discussion while driving moderates away.
  2. Platform owners and traditional gatekeepers have been unable or unwilling to fix this, so as casual users quit the platforms the most extreme and vocal actors gain more influence.
  3. Large language models could pull people toward the center by offering polite, expert-like answers and on-demand fact-checking from broad training data. But AI also tends to homogenize viewpoints and can spread errors or suppress minority perspectives, so it isn’t a perfect cure.

All paths point downhill

arg min • 218 implied HN points • 31 Oct 24
đź•ą Technology Algorithms
  1. In optimization, there are three main approaches: local search, global optimization, and a method that combines both. They all aim to find the best solution to minimize a function.
  2. Gradient descent is a popular method in optimization that works like local search, by following the path of steepest descent to improve the solution. It can also be viewed as a way to solve equations or approximate values.
  3. Newton's method, another optimization technique, is efficient because it converges quickly but requires more computation. Like gradient descent, it can be interpreted in various ways, emphasizing the interconnectedness of optimization strategies.

Basic Linear Algebra Subprogramming

arg min • 178 implied HN points • 29 Oct 24
đź•ą Technology Algorithms
  1. Understanding how optimization solvers work can save time and improve efficiency. Knowing a bit about the tools helps you avoid mistakes and make smarter choices.
  2. Nonlinear equations are harder to solve than linear ones, and methods like Newton's help us get approximate solutions. Iteratively solving these systems is key to finding optimal results in optimization problems.
  3. The speed and efficiency of solving linear systems can greatly affect computational performance. Organizing your model in a smart way can lead to significant time savings during optimization.

It's all a blur

lcamtuf’s thing • 11631 implied HN points • 06 Feb 26
đź•ą Technology Algorithms
  1. Averaging-based blurs are linear and often reversible, so knowing the filter and padding lets you set up simple equations to recover original pixels.
  2. A right-aligned moving average makes iterative reconstruction straightforward and can reveal fine detail even with large blur windows, though 8-bit quantization adds visible noise.
  3. Two-pass (X then Y) blurs can still be inverted if the filter biases the current pixel, and recovered images can survive normal lossy formats like JPEG unless compression is very heavy.

You Are Being Lied to About Algorithms

Taylor Lorenz's Newsletter • 1552 implied HN points • 10 Mar 26
đź•ą Technology Algorithms
  1. Many people misunderstand what an algorithm is. Even reverse-chronological feeds are algorithms, so using “algorithms” as a reason to strip platforms of Section 230 is flawed.
  2. Politicians are using the techlash to amass more power and censorship has become a bipartisan value. Big platforms like Meta may actually want Section 230 changed so they can wipe out smaller competitors.
  3. Algorithms can help protect users from spam, scams, and a miserable internet, so blaming them misses the real threats. Real dangers include policies like age verification laws and other corporate or legal maneuvers that threaten the open web.
Get a weekly roundup of the best Substack posts, by hacker news affinity:

Breaking: Sir Demis Hassabis becomes the latest to say that ChatGPT is a dead-end and that we must turn our focus to world models

Marcus on AI • 9366 implied HN points • 22 Jan 26
đź•ą Technology Algorithms
  1. A leading AI figure says ChatGPT-style large language models are a dead end and researchers should prioritize building world models.
  2. This comment joins other voices pushing the field to move beyond chat interfaces toward systems that actually model and understand the world.
  3. Earlier analysis argues that purely statistical approaches have limits and that neurosymbolic or cognitive 'world' models are needed for deeper AI.

You gotta think outside the hypercube

lcamtuf’s thing • 5101 implied HN points • 26 Jan 26
đź•ą Technology Algorithms
  1. You can build a tesseract wireframe by extending the same edge-construction rules from a square to a cube and then to 4D, which yields 16 vertices and 32 edges.
  2. Rotations in four dimensions are still planar operations that act on pairs of axes, so animations come from applying familiar 2D rotation formulas to axis pairs like XZ or Z🌀.
  3. There are many ways to project 4D to 2D with different tradeoffs—cavalier, cabinet, isometric, perspective, and fisheye—and a mixed approach (isometric for XYZ plus perspective or fisheye for the fourth axis) gives the clearest, most informative views.

See it with your lying ears

lcamtuf’s thing • 6938 implied HN points • 10 Jan 26
đź•ą Technology Algorithms
  1. Images and audio are both sampled data so you can apply similar transforms to both, but ears and eyes perceive artifacts very differently so the same operation can look fine and sound awful.
  2. Pixelating or reducing bit depth in audio creates stair-step or high-frequency errors that produce metallic squeals or hiss, and those artifacts are typically removed with lowpass/rolling-average filtering or proper DAC anti-aliasing.
  3. Frequency-domain editing works well if you process short, overlapping windows with a Hann (sin^2) weighting and 50% overlap so the attenuations cancel out, avoiding clicks and enabling effects like pitch shifting and vocoding.

How shortform video hacks our attention

Simon Owens's Media Newsletter • 399 implied HN points • 26 Feb 26
đź•ą Technology Algorithms
  1. Shortform video apps are carefully engineered — from swipe mechanics to instant loading — to remove choice and keep people watching, creating a new internet habit that captures attention.
  2. Individual creators can build durable, monetizable media by using simple formats and niche focus — examples include walking local-news clips, conversational podcasts, curated book boxes, and deep-dive newsletters.
  3. Emerging tools and trends like AI-assisted editing, prediction markets, and strategic use of shortform video are likely to reshape media production and give savvy creators and political actors a competitive edge.

10 techniques to pass 90% of algorithmic interviews

Jacob’s Tech Tavern • 3936 implied HN points • 06 Jan 26
đź•ą Technology Algorithms
  1. Algorithmic interviews are mostly pattern-recognition tests, so identifying which known pattern a problem fits lets you solve it quickly.
  2. Roughly ten core techniques — like hashmaps and two pointers — show up repeatedly, so mastering those gives you coverage for most problems.
  3. Doing well is also about grit and signalling: consistent, strategic practice matters as much as raw talent, so build a sustainable prep routine to avoid burnout.

Convex Optimization at the Midpoint

arg min • 198 implied HN points • 17 Oct 24
🚌 Education Algorithms
  1. Modeling is really important in optimization classes. It's better to teach students how to set up real problems instead of just focusing on abstract theories.
  2. Introducing programming assignments earlier can help students understand optimization better. Using tools like cvxpy can make solving problems easier without needing to know all the underlying algorithms.
  3. Convex optimization is heavily used in statistics, but there's not much focus on control systems. Adding a section on control applications could help connect optimization with current interests in machine learning.

How Instagram Broke Its Promise to Protect Teens

Common Sense with Bari Weiss • 162 implied HN points • 27 Feb 26
đź•ą Technology Algorithms
  1. Instagram publicly promised to remove graphic self-harm content from searches, hashtags, and recommendations.
  2. Despite that promise, its algorithm kept surfacing self-harm and eating‑disorder content, leaving teens exposed to vast amounts of harmful posts like many tagged #weightloss.
  3. Newly unsealed internal documents show executives knew the platform was still failing and were worried about being exposed, suggesting the company focused on damage control rather than fully fixing the problem.

the informational abyss gazes also

bad cattitude • 188 implied HN points • 17 Feb 26
đź•ą Technology Algorithms
  1. Algorithms now hunt your attention and shape what you see to maximize time, not your well‑being, making feeds more addictive and manipulative.
  2. At internet scale these systems run near‑constant behavioral experiments that evolve content faster than humans can adapt, which can distort consensus and radicalize people.
  3. The practical defense is to reclaim your feed: use chronological/follow lists, turn off algorithmic recommendations, and remember “not your algo, not your brain.”

Integer programming easily encloses horse

DYNOMIGHT INTERNET NEWSLETTER • 968 implied HN points • 15 Jan 26
đź•ą Technology Algorithms
  1. The horse-enclosure puzzle can be encoded as an integer program using binary variables for walls and for whether a tile can escape, with linear constraints that enforce adjacency and boundaries, so solvers can quickly find and certify optimal enclosures.
  2. Integer programming is a hugely practical and powerful tool for discrete optimization: even though it’s NP-hard in theory, modern solvers solve many real-world instances very fast and reliably.
  3. Whether a combinatorial problem is fun depends on legibility and the right level of difficulty, and many NP-complete problems can be made engaging with a good interface; it’s not obvious whether this specific puzzle is provably NP-complete.

Inverse frontiers

arg min • 158 implied HN points • 07 Oct 24
đź•ą Technology Algorithms
  1. Convex optimization has benefits, like collecting various modeling tools and always finding a reliable solution. However, not every problem fits neatly into a convex framework.
  2. Some complex problems, like dictionary learning and nonlinear models, often require nonconvex optimization, which can be tricky to handle but might be necessary for accurate results.
  3. Using machine learning methods can help solve inverse problems because they can learn the mapping from measurements to states, making it easier to compute solutions later, though training the model initially can take a lot of time.

Software Machine Tools

Bzogramming • 61 implied HN points • 03 Mar 26
đź•ą Technology Algorithms
  1. There is no universal machine tool: every manufacturing process has hard trade-offs in cost, speed, materials, and geometry, and even hypothetical atom-by-atom assemblers would face stability, energy, and material limits.
  2. In software, theoretical universality (Turing-completeness) doesn’t imply practical usefulness—different paradigms like programming languages, neural networks, and superoptimizers are distinct "software machine tools" with very different real-world strengths.
  3. Big opportunities lie in alternative software tools and analyses—verification-driven code synthesis, superoptimizers, compact magic-constant solutions, better static analysis, and more visual/geometric tooling can solve hard problems more efficiently than brute-force code or giant models.

War, fake videos, and engagement farming

Conspirador Norteño • 48 implied HN points • 08 Mar 26
đź•ą Technology Algorithms
  1. Spammy pages are using AI to generate fake videos of the Middle East conflict and posting them across platforms like Facebook, X, Instagram, TikTok, and YouTube.
  2. Many clips show clear signs they’re fake — unrealistic explosions, no real damage, people speaking fluent American English in non‑English locations, and made‑up weapons or effects.
  3. Recommendation algorithms are amplifying these videos, and as long as clicks and views pay off, content farms will keep repurposing and renaming accounts to farm engagement.

Meta: 5,000

lcamtuf’s thing • 8366 implied HN points • 27 Feb 25
đź•ą Technology Algorithms
  1. Reaching 5,000 subscribers is a big deal for a project that went against the usual trends. It's great to see growth, even if it seems small compared to others.
  2. Writing a newsletter is unique because you don't get much direct feedback from readers. It's interesting to see who signs up or leaves but hard to know what they really think.
  3. Three articles worth revisiting cover complex topics: discrete Fourier transforms, fractals, and core concepts in electronic circuits. They offer in-depth discussions that are easy to understand, even for beginners.

How PyTorch Generates Random Numbers in Parallel on the GPU

Confessions of a Code Addict • 577 implied HN points • 18 Dec 25
đź•ą Technology Algorithms
  1. Traditional PRNGs are sequential and don’t parallelize well. Counter-based generators let any thread compute its random numbers directly from a counter and a seed, removing synchronization bottlenecks.
  2. Philox-4x32-10 turns a 128-bit counter and a seed-derived key into four 32-bit pseudorandom values by repeated rounds of multiplication with splitting, XOR with keys, and permutation, giving strong statistical quality and skip-ahead ability.
  3. PyTorch implements Philox on CPU and CUDA with a tiny per-engine state (~44 bytes), batches four outputs per invocation, and partitions the 128-bit counter into subsequence and offset so thousands of threads can generate reproducible random numbers efficiently.

When Algorithms Come to Your Neighborhood

The Honest Broker • 7513 implied HN points • 17 Jan 25
đź•ą Technology Algorithms
  1. Nextdoor can be useful for getting local alerts, especially in emergencies. However, it might not always provide timely information when you need it.
  2. Many users ignore alerts from apps like Nextdoor because they often send old or irrelevant notifications. This can create a false sense of security and put people at risk.
  3. It's important to question whether the information we receive from neighborhood platforms is reliable. If we learn to overlook their messages, we could miss crucial updates.

Involutions: The License to Update in O(1)

Software Bits Newsletter • 206 implied HN points • 14 Jan 26
đź•ą Technology Algorithms
  1. XOR is an involution: applying the same XOR twice cancels it out, so adding and removing an element use the same operation and let you update combined hashes in O(1).
  2. Zobrist hashing leverages XOR to update a chessboard hash with only a few XORs per move, enabling fast transposition-table lookups and huge search speedups; collisions are possible but usually acceptable or verifiable.
  3. More generally, pick the algebraic tool that matches your needs — use involutions like XOR for O(1) incremental updates when collisions are tolerable, rolling linear hashes for sliding windows, or Merkle trees when cryptographic integrity is required.

2026+

Gonzo ML • 315 implied HN points • 07 Jan 26
đź•ą Technology Algorithms
  1. Quadruped robots (dog- or cat-like) will get much better and more practical for real-world use, while humanoid home robots stay too expensive.
  2. We’ll see production-grade agents with predictable 99.9% reliability and richer integrations, driven by better infrastructure and cognitive architectures.
  3. Advances in world models, latent-space reasoning, and multimodal architectures will create new interactive environments and begin to accelerate scientific discovery in certain domains.

RISC-V and Post Quantum Cryptography

Fprox’s Substack • 186 implied HN points • 18 Jan 26
đź•ą Technology Algorithms
  1. Quantum computers threaten today’s public-key cryptography, so governments and industry are already moving to post-quantum algorithms and rolling out standards and deployments now.
  2. Post-quantum schemes (e.g., Kyber, Dilithium, SPHINCS+, Falcon) rely on heavy math like NTT and Keccak, and they trade off key/signature sizes, signing speed, and verification cost differently.
  3. RISC-V can run PQC today using its vector extensions, but lacks dedicated PQC ISA support; targeted accelerations for NTT and Keccak (and vector crypto extensions) would greatly improve performance and are being explored by the community.

The One Property That Makes FlashAttention Possible

Software Bits Newsletter • 257 implied HN points • 29 Dec 25
đź•ą Technology Algorithms
  1. Associativity is the key property that lets you split work, combine partial results, and safely parallelize or stream computations without changing the answer.
  2. Softmax has a hidden associative state — tracking a local max and a scaled sum lets you correct and merge chunked results, which is the math behind FlashAttention’s memory- and time-saving trick.
  3. When optimizing a global computation, look for a small combinable state and an associative combine rule; if it exists you can chunk and parallelize, and if it doesn’t (for example, median) you need a different algorithmic approach.

Big O vs Hardware: Better Complexity ≠ Better Performance

Confessions of a Code Addict • 1106 implied HN points • 03 Aug 25
đź•ą Technology Algorithms
  1. Not all algorithms with lower time complexity perform better in the real world. Hardware efficiency also plays a big role in how fast they run.
  2. An algorithm may have a good time complexity but if it relies on expensive operations, it won't win in performance. It's important to consider how the algorithm works with the CPU.
  3. Some algorithms can perform better on hardware depending on their design. A well-optimized algorithm can take advantage of hardware strengths, leading to faster results compared to those with similar complexity.

From Theory to Practice: Inductive Biases in Machine Learning

Mindful Modeler • 639 implied HN points • 23 Apr 24
đź•ą Technology Algorithms
  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.

Top 10 Books we read in 2025

Graphs For Science • 105 implied HN points • 10 Jan 26
đź•ą Technology Algorithms
  1. A strong theme is practical engineering: many books show how to turn LLM demos into working agents using RAG, embeddings, knowledge graphs, tool use, and prompt patterns to make outputs more reliable and auditable.
  2. There’s a clear focus on hands-on playbooks and trade-offs—quick-starts, checklists, code examples, and patterns for prototyping, retrieval, latency/cost decisions, multi-agent orchestration, and production concerns.
  3. The collection balances technical how-to guidance with broader perspectives on responsible use, human uniqueness, organizational strategy, and interdisciplinary science, highlighting ethics, norms for academics, and big-picture questions about life and intelligence.

The Algorithm Next Door

After Babel • 1096 implied HN points • 31 Jul 25
đź•ą Technology Algorithms
  1. Social media platforms like Nextdoor can create fear in communities, making people feel unsafe and distrustful of their neighbors. Instead of bringing people together, they often highlight negative events that amplify anxiety.
  2. This fear can lead to children not being allowed to explore their neighborhoods freely, impacting their independence and social skills. Parents often feel compelled to keep their kids indoors because of the scary stories they see online.
  3. There are better ways to create community connections online that foster trust and safety. Platforms like Front Porch Forum encourage neighborly interactions without the fear-mongering found in traditional social media.

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 Algorithms
  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.

Machine Learning From Zero, Chapter 01

The Palindrome • 4 implied HN points • 14 Mar 26
đź•ą Technology Algorithms
  1. Machine learning means training predictive models from data. The core setup uses a dataset, a parametric model (a hypothesis), and a loss function to measure how well the model fits the data.
  2. A model approximates the true input–output relation and depends on both its parameters and the training data (often written h(x; w, D)). Models can be deterministic or probabilistic and belong to different families like generative or discriminative.
  3. Which learning paradigm you use depends on what inputs, outputs, and labels are available — the main paradigms are supervised, unsupervised, semi‑supervised, and reinforcement learning. In supervised learning you have input–label pairs and the goal is to learn the mapping from x to y.

Daylight swiping

Vague Blue • 778 implied HN points • 19 Mar 24
đź•ą Technology Algorithms
  1. The evolution of the swipe gesture, popularized by Apple, has changed how we interact with technology, from unlocking phones to scrolling through social media.
  2. The swipe has become ingrained in modern culture, especially through dating apps like Tinder, where it serves as a rapid filter for potential matches.
  3. Continuous swiping on apps can create a sense of infinite possibilities but can also lead to mindless behavior, trapping users in a cycle of seeking without finding.

Domain Transformations: The Art of Finding Easier Spaces

Software Bits Newsletter • 103 implied HN points • 05 Jan 26
đź•ą Technology Algorithms
  1. Transform hard problems into easier ones by moving to a different domain, doing the simpler computation there, and (if needed) transforming the result back; this is worth it when the transform cost plus the easier computation is less than solving the original problem.
  2. Use well-known transforms to fix numerical and computational issues: log-space turns tiny-product underflow into stable sums (use the log-sum-exp trick to add probabilities safely), Fourier turns convolution into cheap pointwise multiplication, and embeddings or kernels lift data so linear methods work.
  3. Always check that a transform preserves what you need and that the round-trip cost is justified; the best algorithms exploit problem structure by finding the space where the computation becomes simple.

LLMs - Part 1: Tokenization and Embeddings

Vasu’s Newsletter • 104 implied HN points • 05 Jan 26
đź•ą Technology Algorithms
  1. Text is split into discrete tokens, often subwords using Byte Pair Encoding, so a fixed vocabulary can represent any input by keeping common words whole and breaking rare words into parts.
  2. Each token ID is looked up in a learned embedding matrix to produce a dense vector, and these embeddings capture semantic and syntactic relationships learned during training.
  3. Embeddings are context-free and don’t encode position by themselves, so transformer mechanisms like attention and positional encodings combine them to determine meaning and word order.

Inductive biases of the Random Forest and their consequences

Mindful Modeler • 379 implied HN points • 21 May 24
đź•ą Technology Algorithms
  1. Machine learning models like Random Forest have inductive biases that impact interpretability, robustness, and extrapolation.
  2. Random Forest's inductive biases come from decision tree learning algorithms, random factors like bootstrapping and column sampling, and ensembling of trees.
  3. Some specific inductive biases of Random Forest include restrictions to step functions, preference for deep interactions, reliance on features with many unique values, and the effect of column sampling on feature importance and model robustness.

Algorithms are making political speech more extreme

Taylor Lorenz's Newsletter • 3582 implied HN points • 09 Nov 24
🇺🇸 U.S. Politics Algorithms
  1. Algorithms are changing how politicians speak. They now exaggerate and hyperbolize to get more likes and shares, which can lead to more extreme views.
  2. Social media has replaced traditional broadcasting, making it harder for politicians to reach their audience directly. Now, they must adapt their messages for platforms that promote viral content.
  3. Facial recognition technology is increasingly used by governments to track and suppress protesters. This makes it riskier for people to express dissent, as they can be easily identified and punished.

Commutativity: Why Transformers Need Positional Encodings

Software Bits Newsletter • 103 implied HN points • 01 Jan 26
đź•ą Technology Algorithms
  1. Self-attention treats all positions symmetrically, so permuting tokens just permutes outputs; because attention is permutation‑equivariant, Transformers need positional encodings to learn token order.
  2. Commutativity is a deliberate design trade‑off: it enables parallelization and is perfect for unordered data like point clouds, sets, and graphs, but it destroys order information so you must use non‑commutative models or inject positions when order matters (language, time series).
  3. Commutativity shows up across ML: global pooling gives useful invariance but loses location, gradient aggregation and distributed training rely on commutative sums, and floating‑point associativity issues can still cause small nondeterminism.

HashMap is a Data Structure every SWE should know.

System Design Classroom • 239 implied HN points • 24 May 24
đź•ą Technology Algorithms
  1. Hashmaps are useful for storing data by connecting unique keys to their values, making it easy to find and retrieve information quickly.
  2. When two different keys accidentally produce the same hash code, it's called a collision. There are ways to handle this, like chaining and open addressing.
  3. Hashmaps can do lookups, insertions, and deletions really fast, usually in constant time, but they can slow down if too many items cause collisions.

How Unix Spell Ran in 64kB RAM

Confessions of a Code Addict • 1683 implied HN points • 12 Jan 25
đź•ą Technology Algorithms
  1. Unix engineers faced a big challenge in fitting a large dictionary into just 64kB of RAM. They came up with clever ways to compress the data and use efficient structures to make everything fit.
  2. A key part of their solution was the Bloom filter, which helped quickly check if words were in the dictionary without needing to look up every single word, saving time.
  3. They also used innovative coding methods to further reduce the size of the data needed for the dictionary, allowing for fast lookups while staying within the strict memory limits of their hardware.