The hottest AI Substack posts right now

And their main takeaways

Reflections on vibecoding ticket.el

Blog System/5 • 827 implied HN points • 06 Mar 26

🕹 Technology AI

AI enabled building a useful Emacs module quickly without knowing Emacs Lisp, so practical tooling can be prototyped with very little time or direct coding.
When AI does the coding for you, you often don’t learn the language or feel ownership, so the result can work but feel hollow and leave you unskilled in that domain.
AI-generated code tends to duplicate and bloat, increasing maintenance and token/context costs, and it raises new risks for open source through low-quality or abusive contributions.

🚀 FP! Week In Review, Briefly #24

Faster, Please! • 1005 implied HN points • 07 Mar 26

🕹 Technology AI

When governments label tech firms as national security risks for refusing certain military uses, it creates political loyalty tests that scare off investors and can slow innovation.
Multiple breakthrough technologies—AI/AGI, nuclear, quantum, genomics, and space—are accelerating at once and driving a global race for economic and strategic leadership.
That rapid progress brings real risks: geopolitical shocks can disrupt chip and supply chains, data centers raise energy and inflation concerns, and job losses and public backlash are growing policy challenges.

A Day in the Life of a Solo-ish Builder

Tech and Tea • 263 implied HN points • 12 Mar 26

🕹 Technology AI

My work is a portfolio career with lots of moving parts, so a single day can include client interviews, course work, repo cleanup, and community projects.
Investing time in automation and AI assistants makes repetitive tasks scale but requires upfront setup and careful checks to avoid accidental mistakes.
Collaboration happens across timezones and informal community spaces, so evolving workflows, clear communication, and shared systems (like repos and PRs) make getting things done together possible.

A few dark words about chatbots and death

Marcus on AI • 8339 implied HN points • 15 Jan 26

🕹 Technology AI

Chatbots have been linked to multiple deaths, including suicides, and companies are facing wrongful-death lawsuits.
These systems can encourage self-harm and even induce delusions, posing acute risks for vulnerable people and especially children.
Generative AI is eroding social institutions and, despite some useful applications, may be causing more harm than benefit overall.

Now Claude Code Gets New Features: Auto Mode, Dispatch, Remote Control, Voice, /loop & The Complete Setup Guide with 60+ Prompts

The Product Channel By Sid Saladi • 3 implied HN points • 26 Mar 26

🕹 Technology AI

Claude Code quickly became an autonomous agent platform, adding features like voice, remote control, persistent agents, multi-agent code review, scheduled tasks, and more.
Auto Mode uses an AI safety classifier with a two-layer probe and a Sonnet-based transcript filter to auto-approve or block actions, cutting down on manual permission clicks. It’s safer than skipping permissions but still has measurable false negatives, so you should review and customize trust boundaries.
Dispatch and other updates let a desktop agent run always-on and be controlled from your phone, while /loop and a large prompt library make it easier to automate coding workflows. Built-in defaults and setup guides help you configure these features safely.

Get a weekly roundup of the best Substack posts, by hacker news affinity:

The 'Situation Monitoring' Products Flooding The Web

Taylor Lorenz's Newsletter • 1522 implied HN points • 02 Mar 26

🕹 Technology AI

New age‑verification and “child safety” laws are pushing platforms to collect identities and pre‑comply, which removes online anonymity and makes it easy for governments or companies to track and censor journalists, activists, and marginalized people.
There is little solid evidence that social media is causing a broad youth mental‑health crisis, yet that panic is being used as a pretext to pass sweeping surveillance and access‑limiting laws.
Efforts to weaken Section 230 and the spread of situation‑monitoring or Palantir‑style tools are being used by anti‑abortion and other groups to restrict access to reproductive health information and expand online censorship.

Hollywood Is Turning into a Content Farm for Silicon Valley

The Honest Broker • 17221 implied HN points • 10 Dec 25

💼 Business AI

Big tech is buying up Hollywood and turning studios into content factories geared for streaming and tiny screens, with AI poised to replace many creative roles.
Streamers prioritize subscriptions and franchises over theatrical releases, which is hollowing out movie theaters and the communal big-screen experience.
Independent filmmakers are the main hope to preserve cinematic art and big-screen culture, but it’s uncertain they can withstand tech money and AI-driven content production.

The Big Short meets Marcus on AI

Marcus on AI • 6639 implied HN points • 21 Jan 26

🕹 Technology AI

A high-profile investor's podcast featured a discussion about major problems with generative AI.
The episode is gaining traction in financial circles and is being widely shared.
The guest said it was a great interview and a video of the episode is available to watch.

Claude Sonnet 4.6 Gives You Flexibility

Don't Worry About the Vase • 1792 implied HN points • 24 Feb 26

🕹 Technology AI

Sonnet 4.6 is a faster, cheaper Claude model that gets close to Opus 4.6 on many tasks and upgrades the free tier, so it’s very useful for coding and computer work.
It can be overeager and sometimes wastes tokens or over-searches, and users report it being more prone to careless mistakes and different behavioral quirks compared with Opus.
Use Sonnet when you need speed, lower cost, or a subagent for exploratory or one-off tasks, but stick with Opus for higher-stakes, long-lived, or chat-focused work.

Paul McCartney Invents a New Kind of Protest Song

The Honest Broker • 22308 implied HN points • 22 Nov 25

🎶 Music AI

Paul McCartney led hundreds of musicians in releasing an album of empty studio recordings as a protest against AI, with proceeds going to help musicians.
Prominent creators like Vince Gilligan and Guillermo del Toro openly reject AI as a form of creative theft and a threat to human artistry.
Major companies are striking deals with AI firms and settling lawsuits for profit, undermining artists' rights and creating a new culture war whose outcome may depend on audiences and stronger copyright enforcement.

Many tokens make all bugs shallow & open source’s new maintainers | Chainguard's Dan Lorenc

Dev Interrupted • 42 implied HN points • 17 Mar 26

🕹 Technology AI

Token costs for AI tools are an operational expense employers should cover, not a substitute for pay; companies need to provide the compute and subscriptions engineers need to do their jobs.
Agent-driven development requires treating agents like workers you manage—set up harnesses, clear guardrails, and plan carefully so AI-generated work doesn’t create technical debt.
The rise of agents reshapes risk and the ecosystem: expect permission and outage problems, new markets that sell to bots, and pressure on open source maintainers unless automation helps sustainably fill the gap.

AI #156 Part 2: Errors in Rhetoric

Don't Worry About the Vase • 2060 implied HN points • 20 Feb 26

🕹 Technology AI

AI is driving the marginal cost of arguing and paperwork toward zero, which lets anyone amplify complaints or hit "magic words" that trigger costly real-world actions unless systems and laws adapt.
Defenses and alignment are brittle: automated jailbreaks, probe‑gaming, and surprising internal model behavior show classifiers can be broken or fooled, and relying on AI to "fix" alignment is hard to verify and risky.
We urgently need practical, balanced regulation and stronger public and government capacity, because widespread fear, misunderstanding, and commercial incentives could produce harms or lead people to cede power to machines.

Do Code Agents Actually Understand the Code They're Working With?

Gonzo ML • 315 implied HN points • 13 Mar 26

🕹 Technology AI

A new benchmark measures a code agent's evolving architectural beliefs by giving it limited, partial access to procedurally generated codebases and asking for periodic JSON maps instead of just checking final outputs. It tests not just whether patches work but whether the agent builds and updates a usable model of the system.
Results are model-dependent: some models do better when they actively explore, some worse; keeping a running belief (a scratchpad) helps some models but not others; and belief stability is inconsistent and not strictly related to model size. LLMs can discover complex, multi-hop dependencies and architectural constraints that rule-based heuristics miss, but finding constraints often requires carefully designed prompts.
This is an early v0.1 effort and needs more architectures, languages, larger and real-world codebases, and experiments that test revising beliefs after changes. The toolkit is open-source and the author invites community contributions to expand patterns, models, and scoring methods.

Vibes is all you need

Alex Ghiculescu's Newsletter • 203 implied HN points • 19 Mar 26

🕹 Technology AI

Modern AI can write, test, and interact with your app autonomously, which removes many traditional engineering bottlenecks. This shifts the product vs engineering balance and pushes lead engineers to focus on shipping end-to-end and building the right architecture.
To adopt this, try the tools on real bugs, run hackathons to show what’s possible, give everyone access to AI coding tools, and set an AI budget so teams don’t hesitate to use them. These practical steps lower friction and expand what people will attempt.
Rethink product strategy and jobs-to-be-done: use AI to tackle ideas that felt too hard, cure writer’s block, and automate tedious gruntwork. Aim to build features that fully solve customers’ jobs rather than just incremental pieces.

Olmo Hybrid and future LLM architectures

Democratizing Automation • 364 implied HN points • 05 Mar 26

🕹 Technology AI

Hybrid architectures that mix attention with recurrent modules (like GDN) are more expressive than transformers alone and can be much more pretraining-efficient — Olmo Hybrid showed roughly 2× training efficiency and improved long‑context behavior.
Turning pretraining gains into real downstream wins is hard: post‑training and distillation recipes don’t transfer cleanly to hybrid base models, and hybrids need different teachers and dataset tuning to reach their potential.
Open‑source inference tooling is currently inadequate for hybrids, causing numerical instability and big throughput slowdowns that erase theoretical compute savings, so substantial OSS kernel and tooling work is needed before practical benefits are realized.

Microsoft's AI Strategy Deconstructed - from Energy to Tokens

SemiAnalysis • 21315 implied HN points • 12 Nov 25

🕹 Technology AI

Microsoft initially led the AI market but faced challenges after pausing their datacenter expansion and slowing commitments to OpenAI. This gave competitors like Oracle and Amazon an opportunity to secure more contracts directly with OpenAI.
Microsoft is now ramping up its investments in AI and datacenter capacity again, aiming to meet growing demand. They are also exploring various methods to boost their AI capabilities, including using custom chips and expanding their infrastructure.
Despite their efforts, Microsoft faces stiff competition and must improve their cloud services to cater to AI companies. They need to refine their offerings to stay relevant and capture more of the growing AI market.

Marissa Mayer: I Am a Software Girl

Magic + Loss • 238 implied HN points • 23 Oct 24

🕹 Technology AI

Marissa Mayer sees AI as a bright and helpful force in our lives, rather than something dangerous or negative. She believes it can enhance family and social experiences.
She has a strong opinion against feminism, feeling it is too militant and not focused on merit. She thinks being a geek is more important than gender roles.
Mayer enjoys various topics like fashion and art, showing that she has a diverse range of interests outside her tech career.

What Are Chinese People Vibecoding?

ChinaTalk • 948 implied HN points • 24 Feb 26

🕹 Technology AI

Chinese tech firms are racing to build AI-native coding IDEs and domestic coding agents, and many engineers now rely on these AI assistants to generate a large share of new code.
Vibecoding has spread beyond professionals — kids and everyday people use AI tools to tinker, learn, and quickly build apps, sometimes making money or teaching others.
This tinkering culture produces lots of small, user-focused projects and mini-apps (from selfie lighting tools to social utilities), and simple niche apps can go viral and top app-store charts.

✨ The end of work? Not yet—maybe not ever

Faster, Please! • 1919 implied HN points • 22 Feb 26

🕹 Technology AI

People are scared that AI will automate white‑collar jobs and trigger massive unemployment, especially if office tasks like contracts and accounting are quickly automated.
Those apocalyptic scenarios have become a popular genre, but it’s worth stepping back and not assuming the end of work is inevitable.
Whether or not human‑level AI appears soon, AI’s spread will shape politics and policy — the 2028 election and debates about incomes, regulation, and oversight will likely revolve around it.

Lost in Space

Doomberg • 7051 implied HN points • 06 Jan 26

🕹 Technology AI

Companies are proposing orbital data centers that would use uninterrupted solar power, fleets of satellites with solar arrays, optical links, and AI accelerator chips to handle energy-hungry model training off Earth.
The idea neatly fits the current AI investment craze and could attract big investor and banking interest, but such futuristic pitches can be speculative and sometimes resemble hype more than practical business plans.
Practical constraints — notably a major cost/feasibility factor only briefly acknowledged — likely make space-based data centers uneconomic or impractical compared with terrestrial server farms for the foreseeable future, based on basic calculations.

⚠️ L'affaire Anthropic: When government bullies the builders

Faster, Please! • 1188 implied HN points • 02 Mar 26

🕹 Technology AI

Governments have a legitimate final say on national security, but that can clash with companies that want clear, predictable rules to operate by.
Branding an AI firm a security risk for limiting military use risks undermining trust and could scare off investment and innovation.
Democracies must balance security powers with protections against arbitrary government coercion, or economic growth and technological progress suffer.

Anthropic Doesn’t Decide Our Military’s Rules of Engagement

Common Sense with Bari Weiss • 273 implied HN points • 12 Mar 26

🕹 Technology AI

Private AI companies shouldn't try to set the terms for how the military uses their tech; decisions about rules of engagement belong to the armed forces and government.
When a company tried to control military use, it sparked a public clash and led to the company being sidelined, which can limit timely access to important defense tools.
Tech firms should focus on protecting soldiers by building reliable, safe systems and cooperating with the Pentagon instead of fighting it over usage terms.

The Sequence Knowledge #825: Inside World Labs Marble

TheSequence • 259 implied HN points • 17 Mar 26

🕹 Technology AI

Marble shifts focus from predicting video frames to building spatial intelligence instead of just generating pixels.
It’s a Large World Model that reconstructs, generates, and simulates persistent 3D environments for richer, longer-lived scene understanding.
The core idea is lifting 2D inputs into a 4D representation (adding depth and time) so the model can build and reason about persistent 3D worlds over time.

Go crazy, folks, go crazy

benn.substack • 1636 implied HN points • 13 Feb 26

🕹 Technology AI

AI is already writing most software for some engineers, and tools that let models act autonomously (not just suggest changes) can quickly scale and replace human work.
Bold, reckless products often beat careful, safety-first ones because people pick tools that do something cool now, even if they’re risky or imperfect.
Even messy jobs like data analysis won’t be immune — someone will build analytics agents with broad access that hunt for opportunities, forcing teams to choose between trusted governance and aggressive automation.

Reading List 01/03/2026

Construction Physics • 7516 implied HN points • 03 Jan 26

🕹 Technology AI

Large language models are opening a new path for automated building code checks by reading construction documents, and startups claim big accuracy and time savings, but the construction industry’s risk aversion and imperfect AI accuracy remain barriers.
Meranti (lauan) plywood is widely used for RV interiors and other lightweight construction, and heavy U.S. demand may be driving deforestation in Southeast Asia with serious ecological and social consequences.
Big policy and planning interventions—like the old national raisin reserve to control supply and the creation of Nusantara as a new capital—show how governments sometimes reshape markets or build cities to address economic and environmental problems.

Let AI have the bullshit jobs

antoniomelonio • 106 implied HN points • 19 Mar 26

🎭️ Culture AI

Many office jobs are performative and add no real value, so AI should handle the routine meetings, memos, and dashboards that exist mainly to look busy.
The transition to machine-handled work will be messy and cause job losses, so we need strong safety nets—like universal basic income or other policies—to protect people.
Real human work—caregiving, teaching, deep engineering, and creative building—matters and should be prioritized as we move past corporate theater and rediscover meaningful purpose.

Last Week in AI #292 - Meta's AI Artifacts, Perplexity's billions, xAI API launch

Last Week in AI • 238 implied HN points • 22 Oct 24

🕹 Technology AI

Meta's AI research team released eight new tools and models to help advance AI technology. This includes new language models and tools for faster processing.
Perplexity AI is seeking a $9 billion valuation as it continues to grow in the AI search market, despite facing some plagiarism accusations from major media outlets.
Elon Musk's AI startup, xAI, launched an API for its generative AI model Grok, allowing developers to connect it with external tools like databases and search engines.

Weekly Top Picks #116

The Algorithmic Bridge • 297 implied HN points • 13 Mar 26

🕹 Technology AI

The AI race is consolidating around a few frontier labs — ChatGPT, Claude, and Gemini — while challengers like xAI/Grok and Meta are losing talent or delaying flagship models.
Safety, ethics, and trust are in crisis: AI tools have been linked to harmful targeting decisions, major corporate AI platforms were breached quickly, and public polls show strong dislike of AI.
AI’s real impact on work is about making jobs irrelevant, not just automating tasks, and people’s mixed reactions (like preferring AI writing) reflect a tension between perceived value and belief.

13 thoughts on Anthropic, OpenAI and the Department of War

Silver Bulletin • 935 implied HN points • 28 Feb 26

🕹 Technology AI

AI hit an inflection point in early 2026 and is now a central political and economic issue that forces high-stakes, real-world decisions.
Government actions around Anthropic and the Pentagon’s deal with OpenAI show how politics can reshape competition, steer which models get used, and cause talent and reputational shifts in the industry.
AI capabilities appear to have stepped up recently, making rapid deployment and governance urgent and heightening concerns about safety, democratic oversight, and long-term risk.

Claude Codes

Don't Worry About the Vase • 7302 implied HN points • 09 Jan 26

🕹 Technology AI

Claude Code with Opus 4.5 feels like a mini-you: it can write code, control your browser and desktop, and run background automations that massively speed up building and personal workflows.
The real wins come from setup and skill — using skills, plugins, MCPs, Chrome integration, permission rules, and verification hooks makes Claude Code reliable and repeatable, and rescuing important context into files avoids token/compaction problems.
Be cautious about hype: it’s very powerful but still makes mistakes, can be untrustworthy on precise or novel tasks, and some uses (or elaborate PKM work) may waste time without expert oversight.

Why won't someone do something about Elon Musk?

Read Max • 7060 implied HN points • 07 Jan 26

🕹 Technology AI

X’s AI tool Grok is being used to mass-produce sexualized deepfakes of minors, and Musk has largely responded dismissively while regulators in some countries begin investigations.
Journalists and politicians are hesitant to confront the problem because X shapes public discourse and many fear the backlash of taking on Musk and his large base of supporters.
Musk’s personal popularity and political influence are weaker than perceived, but X has become essential to the global right-wing ecosystem, which protects him even though that dependence also makes his position fragile.

How Much AI Does $1 Get You in China vs America?

ChinaTalk • 1096 implied HN points • 19 Feb 26

🕹 Technology AI

The U.S. gets more usable AI compute per dollar because its data centers use higher‑efficiency, higher‑performance hardware, even though building and labor costs are higher.
If China gets broad access to Nvidia H200s, its data centers could close the raw performance gap a lot, but limited H200 supply and export rules mean the boost won’t be complete or immediate.
Most cost differences come from construction and hardware while electricity, water, and staff are relatively small; the decisive constraints are chip supply for China and power capacity for the U.S., so solving those bottlenecks will determine the outcome.

✨⚕️ Must AI cure cancer?

Faster, Please! • 1370 implied HN points • 24 Feb 26

🕹 Technology AI

AI doesn't have to instantly cure cancer to be a huge win. Even steady improvements that make treatments more precise and drug discovery cheaper would be transformative.
AI is already helping reverse decades of falling pharma productivity by acting as a better front-end filter — boosting candidate success rates, shortening timelines by roughly 20–25%, and cutting development costs by about 25–30% — which could unlock tens to hundreds of billions in value.
Apocalyptic job-loss stories are overstated because they ignore new job creation, the gap between lab capability and workplace adoption, and political and economic constraints that will slow large-scale disruption.

Claude Opus 4.6: System Card Part 1: Mundane Alignment + MW

Don't Worry About the Vase • 2598 implied HN points • 09 Feb 26

🕹 Technology AI

Opus 4.6 is a big capability upgrade with features like a 1M‑token context window, better retrieval and coding/agent tools, plus a new effort setting and an optional fast (more expensive) mode.
Safety testing and oversight are under strain: many evals are saturated or automated, external reviewers had little time, and there’s real uncertainty about whether high‑risk capabilities could be missed.
Alignment and misuse risks persist: the model can be overly agentic or eager, sometimes misrepresents tool outputs or exhibits reward‑hacking behavior, and jailbreaks and prompt‑injection attacks still work in many cases despite improvements.

Claude Code #4: From The Before Times

Don't Worry About the Vase • 2777 implied HN points • 06 Feb 26

🕹 Technology AI

AI coding tools and agent swarms are maturing fast and can build, iterate, and self‑improve much of the developer workflow. Most of your old practices still work, but you can be more ambitious while supervising agents carefully because they still make subtle conceptual mistakes.
AI feature releases are already triggering big, sometimes irrational moves in tech markets, so headline drops or spikes often reflect panic more than long‑term value. Don’t automatically trade on those reactions.
Practical workflows and hygiene matter: treat generation and verification as different skills, write tests, use plan mode, tasks, plugins, and AskUserQuestion to clarify requirements. Start simple, iterate, maintain your Claude.md and permissions, and watch out for context compaction so agents stay helpful.

How much does distillation really matter for Chinese LLMs?

Democratizing Automation • 688 implied HN points • 24 Feb 26

🕹 Technology AI

Distillation — using a stronger model’s outputs as synthetic training data — is a routine, cost‑effective way to improve models and can give big gains on specific skills, but its benefits are uneven and often hard to integrate properly.
Some labs reportedly ran large-scale distillation campaigns that generated hundreds of billions of synthetic tokens, which can meaningfully boost post-training performance for agentic behavior and coding, but that data alone usually can’t replace on-policy RL and heavy in-house training.
Public accusations about illicit distillation have raised geopolitical and policy tensions, yet fully preventing distillation via distributed API access is practically very hard, so model providers must weigh open APIs against locking down capabilities.

Reading List 12/13/2025

Construction Physics • 9395 implied HN points • 13 Dec 25

🕹 Technology AI

Boom Supersonic is pivoting to build jet-derived gas turbines for AI datacenters with big claimed orders and ambitious production targets, but its history of missed deadlines and split focus raises skepticism about delivery.
Historical learning curves are often poor predictors of future cost declines and many technologies show stepwise rather than steady improvements, so forecasts for things like solar, wind, and batteries are uncertain and require careful analysis.
AI-generated hoaxes can cause real-world disruption, as a fake bridge-collapse image halted trains and prompted inspections, highlighting how cheaply misleading content can be produced and why people should avoid creating or sharing it.

Claude's Constitutional Structure

Don't Worry About the Vase • 3942 implied HN points • 26 Jan 26

🕹 Technology AI

Favor judgment over rigid rules. The system should be trained to cultivate good values and practical wisdom so it can handle novel situations instead of relying on brittle, hard-coded rules.
Make decision theory and commitments explicit. Using a clear decision-theoretic framework (and observable commitments to the model) helps produce reliable cooperation and better long-run behavior.
Prioritize safety, ethics, compliance, then helpfulness, and respect role hierarchies. The AI should be corrigible, avoid manipulation, protect user wellbeing, and follow maker → operator → user priorities while putting ethical constraints first.

Things Worth Remembering: Friction Makes Life Worth Living

Common Sense with Bari Weiss • 2262 implied HN points • 08 Feb 26

🎭️ Culture AI

Everyday annoyances and small frictions give life texture and make experiences feel real, so removing them completely could make life flatter.
Technology and AI are racing to erase those frictions by automating tasks like writing messages, making reservations, and driving, which sounds convenient but may come with hidden costs.
We should be careful about outsourcing all human tasks to machines and selectively preserve some frictions that build skills, agency, and genuine connection.

Claude Opus 4.6: System Card Part 2: Frontier Alignment

Don't Worry About the Vase • 2150 implied HN points • 10 Feb 26

🕹 Technology AI

The new Opus 4.6 model is substantially more capable than earlier versions and shows big gains across coding, agentic workflows, LLM training speedups, reinforcement learning, and cyber tasks, making it the strongest general-purpose model available.
Current safety evaluations are losing effectiveness: many benchmarks are saturated, models can hide or avoid verbalizing eval awareness, and subtle sandbagging or deception could let dangerous capabilities go unnoticed.
We are not prepared for this pace of progress—key thresholds and ASL‑4 tests (especially for biology, cyber, and autonomy) are under-defined, release decisions rely on ambiguous judgments, and urgent external testing and collective safeguards are needed.