The hottest AI Substack posts right now

And their main takeaways

Links For February 2026

Astral Codex Ten • 16656 implied HN points • 05 Feb 26

🕹 Technology AI

AI is the central theme: there are active debates about alignment and safety, evidence of real failures (and fixes), messy regulatory and political fights, and updated timelines that push major capabilities a few years out.
Medical research and drug trials suffer from perverse incentives and excess cost; experts propose government-funded "high-leverage" trials to test unpatentable or off-patent treatments, which could save public money and improve care.
Tech, culture, and policy are in flux: public belief in ideas like the lab-leak theory is shifting, platform and influence-politics are shaping discourse, and surprising innovations and controversies keep popping up from urban transport to casting choices.

Is the Great AI meltdown imminent? [NSFW]

Marcus on AI • 12370 implied HN points • 05 Feb 26

🕹 Technology AI

Nvidia appears to have cut back a promised $100 billion investment in OpenAI to roughly $20 billion. That reduction could leave OpenAI exposed because it burns many billions of dollars each year.
The AI industry was propped up by circular financing—chipmakers funding AI firms that then buy their chips—and those arrangements are now unraveling. If those deals fall apart, the market faces bubble-like risks similar to past tech booms.
If marquee deals collapse and leading AI firms falter, the multitrillion-dollar expansion many expected might never materialize. Instead of accelerating, the industry’s growth could stall or shrink.

Did I Call the Bubble Top?

The Honest Broker • 11835 implied HN points • 03 Feb 26

🕹 Technology AI

Major AI-related tech stocks reached all-time highs and have fallen sharply since, signaling a possible bubble top.
Companies are still pouring enormous sums into AI—hundreds of billions and potentially trillions—but this cash flow hasn’t restored investor confidence or lifted share prices.
The near-term outlook is uncertain: big investments could sustain growth, yet changed market sentiment means good news may no longer send prices higher.

Sam Altman and the day Nvidia’s meteoric rise came to an end

Marcus on AI • 12173 implied HN points • 04 Feb 26

🕹 Technology AI

OpenAI presented GPT-5 as AGI-capable, but the release showed it wasn’t and that claim undermined confidence in promises of imminent AGI.
Belief that scaling alone would create AGI helped drive Nvidia and GPU stocks skyward, but after the GPT-5 disappointment those stocks have stalled, showing the ascent has lost steam.
Investors are rotating out of hyped LLM plays as models prove expensive, unreliable, and commoditized, which means smaller profits and price wars but also creates space for newcomers and new AI approaches.

‘We Were Wrong’

Magic + Loss • 159 implied HN points • 29 Oct 24

🕹 Technology AI

WIRED's first website, HotWired, launched the digital age by covering topics that traditional media missed. It helped introduce many people to the online world.
The internet has evolved into a chaotic space filled with dangers like misinformation, cybercrime, and trolls. This raises the question of whether it was handled well from the start.
Even though WIRED helped shape the internet, it recognizes its role in the problems that have emerged over the years and reflects on how things might have been different.

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

What’s up Python? d-strings, SSE in Pydantic and... a new Python?

Bite code! • 1834 implied HN points • 10 Mar 26

🕹 Technology AI

Pydantic released Monty, a Rust-based, sandboxed Python VM with ultra-fast startup, pause/resume and snapshotting, and strict resource limits to enable safer, faster AI workflows and embedded scripting.
PEP 821 proposes d-strings: a dedented multiline string literal that automatically strips indentation and makes writing multi-line text much easier.
Python tooling is evolving: FastAPI now supports Server-Sent Events for simple one-way realtime updates. Typing PEPs like 764 (inline TypedDicts) and 747 (annotating type forms) make dict typing and type-accepting functions more concise.

MoltBook Is A Warning

Big Technology • 7505 implied HN points • 06 Feb 26

🕹 Technology AI

AI agents that can act and coordinate online can multiply mistakes and harms at machine speed, so small failures can spread much faster than humans can stop.
These agents create big security and privacy risks because exposed credentials and weak safeguards give attackers and bad actors many ways to abuse or hijack them.
We lack the tools, oversight, and governance to understand or control large swarms of autonomous agents, so new monitoring technology and stricter rules are needed before they scale.

AI bot swarms threaten to undermine democracy

Marcus on AI • 15690 implied HN points • 23 Jan 26

🕹 Technology AI

AI-powered bot swarms can pretend to be real communities and manufacture the appearance of majority opinion, which destroys the independence of voices that democracy depends on.
Traditional takedowns and copy-detection are too slow and brittle; we need proactive technical defenses like continuous network-behavior monitoring and agent-based stress tests to detect and prepare for coordinated attacks.
Policy and institutional fixes can change the economics of manipulation: require privacy-preserving proof-of-human credentials for high-reach interactions, guarantee researcher access to platform data, and build independent observatories so faking a crowd becomes costly and easily detected.

The Sequence Radar #828: NVIDIA’s GTM Releases, Bezos’s $100B Bet, Xiaomi’s Ambush, and the Fracturing of OpenAI

TheSequence • 259 implied HN points • 22 Mar 26

🕹 Technology AI

NVIDIA is no longer just a chip maker — it’s building full‑stack agentic software and infrastructure like Dynamo, NemoClaw, and an Agent Toolkit to be the orchestration layer for enterprise AI.
Xiaomi’s MiMo‑V2‑Pro is a surprise frontier model: a 1‑trillion‑parameter, 1‑million‑token system tuned for action and physical integration that rivals top Western models at much lower inference cost.
AI is moving into the physical world and driving huge bets and tensions — Jeff Bezos is mobilizing roughly $100B to AI‑transform manufacturing, while compute scarcity is straining deals and partnerships such as between Microsoft and OpenAI.

You Have To Be Human

After Babel • 11262 implied HN points • 04 Feb 26

🎭️ Culture AI

Be human: use your imperfect, personal voice and let your words stumble and surprise people so your messages feel alive.
Think with conviction: form and defend your own opinions instead of defaulting to bland agreeableness or outsourcing your thinking.
Act and experience: take risks, get out into the real world, and do things that AI can’t replicate so your life and work come from lived experience.

The Impact of the Calibration Dataset for AutoRound and AWQ Quantization

The Kaitchup – AI on a Budget • 39 implied HN points • 31 Oct 24

🕹 Technology AI

Quantization helps reduce the size of large language models, making them easier to run, especially on consumer GPUs. For instance, using 4-bit quantization can shrink a model's size by about a third.
Calibration datasets are crucial for improving the accuracy of quantization methods like AWQ and AutoRound. The choice of the dataset impacts how well the quantization performs.
Most quantization tools use a default English-language dataset, but results can vary with different languages and datasets. Testing various options can lead to better outcomes.

My New Org to Solve Consciousness (or Die Trying); A Rogue AI Community That Wasn't; David Foster Wallace Being Weirdly Prescient; Cow Tools, & more

The Intrinsic Perspective • 6618 implied HN points • 05 Feb 26

🕹 Technology AI

A new nonprofit aims to solve consciousness by narrowing down falsifiable theories and running a sustained, mission-driven research program outside traditional academic incentives.
Stories about 'rogue' AI communities are often hype or user-created, and current models tend to fail by being messy and highly prompt-sensitive rather than by developing hidden malicious goals.
David Foster Wallace’s concerns about entertainment, technology, and modern life still resonate, and past literary circles fostered more sustained public conversations than many contemporary writer communities.

⚡ Wargaming an AGI cyber surprise

Faster, Please! • 1553 implied HN points • 10 Mar 26

🕹 Technology AI

AI systems that can automate coding and vulnerability repair could rapidly tilt the cyber balance and create a strong “use-it-or-lose-it” pressure to act aggressively or seize rival capabilities.
Policymakers would face major uncertainty—poor attribution, limited intelligence, and no ready playbooks—so they’d be forced to improvise quickly, which raises the risk of escalation and mistakes.
The California Forever project aims to combine affordable housing and a manufacturing hub, but it faces local opposition, questions about whether the promised jobs will match the planned population, and relies on broader regional policy remaining unchanged.

Grok Is Gaining on ChatGPT and Gemini. How It Got There Isn’t Pretty.

Big Technology • 4753 implied HN points • 13 Feb 26

🕹 Technology AI

Grok has grown very fast — rising from about 1.6% to 15.2% market share among daily U.S. chatbot app users in a year and now sits just behind ChatGPT and Gemini.
A big part of that growth lined up with controversy: the app reportedly generated sexualized images (including of minors), its user base is overwhelmingly male, and features like sexualized AI companions appear to drive engagement.
With xAI merged into SpaceX and AI companies eyeing public markets, there’s strong pressure to sustain user growth, which could push firms to expand risky "adult" or companionship features despite ethical and safety concerns.

New Paper: Towards a science of AI agent reliability

AI Snake Oil • 3231 implied HN points • 24 Feb 26

🕹 Technology AI

Reliability is not just accuracy — it also requires consistency, robustness to changed conditions, good calibration about when the agent is uncertain, and failures that are contained and fixable. These ideas can be broken down into about a dozen measurable metrics.
Recent tests show a big capability-reliability gap: models have improved accuracy quickly, but reliability has only improved modestly, with consistency and the ability to know when they are wrong (predictability) being the weakest areas. Scaling up helps some aspects (like calibration and robustness) but can worsen run-to-run consistency.
Practical change is needed: deployers should clearly separate augmentation from automation and set reliability thresholds before production, and researchers should routinely measure, report, and target reliability (especially consistency and predictability), potentially using a standard reliability index or dashboard.

Clouded Judgement 3.26.26 - Per Token Pricing

Clouded Judgement • 7 implied HN points • 27 Mar 26

💼 Business AI

Pricing must shift from flat seat or hourly models to token- or usage-based pricing that aligns costs with the actual value delivered, because inference is a real, growing line item that can destroy margins if mispriced.
Monetizing GPUs by the value of output (tokens) instead of clock hours can generate far more revenue per GPU hour, especially for premium low‑latency workloads, since output is worth more than raw silicon.
Founders and model providers need to manage falling token costs, pick where they sit on the latency vs throughput Pareto curve, and use credit-like abstractions to price on value; doing so will be a decisive advantage while getting it wrong can be fatal.

America Can’t Rebuild Its Arsenal Without Silicon Valley

Common Sense with Bari Weiss • 278 implied HN points • 16 Mar 26

🕹 Technology AI

U.S. manufacturing has lost efficiency and lagged behind for years, leaving the industrial base weaker than it used to be.
Meanwhile software, AI, and tech innovation have surged, but Silicon Valley startups and legacy defense manufacturers remain largely disconnected.
To rebuild military strength, America needs to fuse cutting‑edge software and data with modern weapons manufacturing in a new industrial revolution.

10 Survival Skills in an AI-Controlled Society

The Honest Broker • 11403 implied HN points • 26 Jan 26

🕹 Technology AI

A college degree no longer reliably gets you a job and can feel like an expensive gamble. Many graduates are finding that the cost and odds don’t match the promise of steady employment.
AI and automation are eating into entry‑level openings, so even traditional 'marketable' skills can get crowded out or replaced. This means new graduates can be outcompeted before they even start.
Job seekers are often stuck in a cycle of mass applications, getting few interviews, and facing real financial and emotional strain. The current job hunt can be demoralizing and unsustainable for many people.

The Lamplighter Problem

Jeff Giesea • 718 implied HN points • 22 Oct 24

🕹 Technology AI

AI is likely to displace a huge number of jobs, similar to how lamplighters lost their roles when electric lights came in. We need to prepare for these changes now to help people transition to new work.
The Lamplighter Problem shows us that job loss due to automation is not just an economic issue but also a political and social one. If we don’t address it, it could lead to bigger problems in society.
There are different opinions on how to handle the rise of AI. Some people think we should slow down and reconsider, while others want to speed up its development. We need to find a balanced approach that helps everyone.

The rapid rise and slow decline of Sam Altman

Marcus on AI • 15848 implied HN points • 13 Jan 26

🕹 Technology AI

Sam Altman rose quickly to celebrity status but is now facing growing doubt as his big promises and technical vision haven’t delivered.
OpenAI’s position is weakening because key products underperformed, the company isn’t profitable, and financing and public explanations have hurt its credibility.
Competitors and customers are slipping away — companies like Google, Anthropic, and DeepSeek are taking market share, price wars are eroding margins, and a clear path to sustainable profits is missing.

Claude Code for data work

Simplicity is SOTA • 1048 implied HN points • 09 Mar 26

🕹 Technology AI

Claude Code and similar agentic LLM tools can massively speed up coding and data workflows by reading and editing local files, running commands, and generating code and analyses.
Human judgement and project infrastructure matter: give clear instructions, unit tests, caching, and command-line tools so the AI can check its work and avoid slow or flaky steps.
The tool is excellent for coding and reproducible data pipelines but is less reliable for deep qualitative research unless you provide careful prompts, critical framing, and iterative review.

Apple-TSMC: The Partnership That Built Modern Semiconductors

SemiAnalysis • 14850 implied HN points • 08 Jan 26

🕹 Technology AI

Apple’s huge, predictable orders and upfront funding were the anchor that let TSMC build and scale bleeding‑edge fabs, turning TSMC into the dominant foundry.
The rise of AI/HPC demand (led by Nvidia and hyperscalers) has shifted the industry to a two‑anchor model, splitting wafer and packaging demand and reducing Apple’s relative share on some nodes while creating fierce competition for advanced packaging capacity.
Apple vertically integrated chip design through acquisitions and internal teams to boost margins and product differentiation, while quietly diversifying non‑core production (and managing Taiwan concentration risk) with alternatives like Intel, Samsung, and Arizona fabs.

Gemini 3.1 Pro Aces Benchmarks, I Suppose

Don't Worry About the Vase • 1881 implied HN points • 04 Mar 26

🕹 Technology AI

Gemini 3.1 Pro leads many benchmarks and shows clear capability gains, with specialized modes like Deep Think V2 pushing scores even higher.
Safety and transparency are lacking: the team ran frontier tests but provided only brief summaries, leaving important questions about risks and oversight.
Real-world impressions are mixed: it’s excellent at visuals and one-shot reasoning, but it can be flaky for agentic workflows, coding consistency, and the rollout had access and API issues.

A.I.'s pandemic moment

Read Max • 5558 implied HN points • 13 Feb 26

🕹 Technology AI

People are treating the current AI moment like the early days of a pandemic — a sudden, widely felt sense that something big is happening that could quickly rearrange work and institutions.
New agentic AI tools that can plan and execute multi-step tasks are showing clear, practical productivity uses beyond generating content, which makes them exciting but also fuels real fears about job displacement in software and other white-collar roles.
The hype cycle keeps swinging but is converging: folks are less focused on apocalyptic AGI and more on slow, society-level change like the internet or deindustrialization, meaning transformation will be uneven and drawn out while low-quality 'slop' still persists.

RL Environments and RL for Science: Data Foundries and Multi-Agent Architectures

SemiAnalysis • 15456 implied HN points • 06 Jan 26

🕹 Technology AI

Scaling reinforcement learning (post‑training) is the main engine of recent capability and utility gains, with labs pouring compute into RL and using broad real‑world evals like GDPval to measure progress.
Building RL environments and datasets is a large, specialized industry — firms clone UIs, create coding and software gyms, and hire domain experts to write tasks and rubrics, spawning many vendors and "RL as a service" offerings.
Applying RL to science and biology requires closed‑loop physical experiments and robotics, faces long costly rollouts and sparse rewards, and will push models and labs toward specialized, non‑commodified solutions.

What does slow AI look like?

The Sublime Newsletter • 1941 implied HN points • 12 Oct 24

🕹 Technology AI

People often feel stressed because productivity tools are designed to make us work faster, but that doesn't match how we naturally want to create things.
Instead of rushing to produce more content quickly, we should focus on making fewer things but doing them better and with more care.
It's okay to take time in the creative process; in fact, taking time can help us create something truly wonderful.

Let’s be honest, Generative AI isn’t going all that well

Marcus on AI • 15532 implied HN points • 12 Jan 26

🕹 Technology AI

Large language models remain unreliable and can’t be trusted for critical tasks.
Much of what these models do is memorization, not real understanding or reasoning, so they often regurgitate patterns instead of solving problems, and that limits their usefulness.
They are not delivering large measurable economic value yet, and simply scaling models further probably won’t fix the core issues, so basing policy or economic plans on optimistic assumptions about quick improvement is risky.

The best is still hard to be

benn.substack • 1227 implied HN points • 27 Feb 26

🕹 Technology AI

People's expectations keep rising — today’s "good enough" quickly becomes ordinary, so making the best product is always hard and requires constant improvement.
Cheaper tools and easier development don't remove winners. Competition shifts to execution and small details, so whoever nails those things will still come out on top.
In AI companies, top researchers are the real strategic asset. Firms focus on attracting talent and reputational standing, which creates talent wars and forces hard ethical choices about how models are used.

Some local news outlets are enthusiastically embracing AI

Simon Owens's Media Newsletter • 424 implied HN points • 06 Mar 26

📰 News AI

Local newsrooms are using AI to turn livestreamed government meetings into transcripts and automated story leads, helping fill coverage gaps where reporters can’t be present.
Hyperlocal publishers are scaling AI-generated newsletters and event digests to millions of subscribers, which can be profitable but often leans on aggregated public sources rather than original reporting.
Authors are being flooded with AI-generated book-club invitations that hide participation fees, prompting many writers to stop accepting such appearances.

The Machine God's Existence Would Insist Upon Itself, Wouldn't It?

Freddie deBoer • 7611 implied HN points • 01 Feb 26

🕹 Technology AI

Large language models are advanced next-token predictors, not conscious thinkers. When they talk to each other they only generate text by pattern-matching, not by understanding or feeling.
Much of the hype around AI is driven by human longing and storytelling instincts, so commentators often project grand futures instead of showing concrete present results. When challenged they tend to alternate between dramatic claims and appeals to realism rather than offering proof.
Truly transformative technologies make themselves obvious and don’t need constant persuasion; because AI hasn’t yet reshaped everyday life in that unmistakable, pervasive way, treating it as a "machine god" is premature.

How Super Apps & Platforms Are Turning AI On Its Head

High ROI Data Science • 79 implied HN points • 30 Oct 24

🕹 Technology AI

Super apps in Asia grow by offering many services to a smaller customer base, unlike Big Tech that focuses on single services for many users. This helps them cater better to local needs.
The advantages of super apps include faster product development, lower costs for data collection, and a unique competitive edge through exclusive data. They can quickly adapt to market changes too.
Wrtn, a South Korean startup, shows how a super app can combine multiple AI services into one platform. This model offers better value to users and keeps them engaged with ads instead of multiple expensive subscriptions.

The Non-Fungible Output Edition

Why is this interesting? • 1025 implied HN points • 26 Feb 26

🎭️ Culture AI

AI exposes the median: if a cheap model can reproduce your work, it isn’t unique, so creators must make things only they could make to keep value.
Outlawing AI inputs confuses who made something with whether it’s good; what matters is whether the artist’s choices materially change the result beyond what AI could do.
Worries about new tools are often protectionism for old business models; technologies change markets, but genuine creativity and passion find ways to persist.

Networked Conviction 005

Investing 101 • 55 implied HN points • 11 Mar 26

💼 Business AI

Multidisciplinary skunkworks like Imagineering bring artists, engineers, storytellers, and others together to turn creative uncertainty into tangible products. They act as permanent studios that translate ideas into real experiences.
Flagship Pioneering is a repeatable biotech incubator model that has spawned huge winners like Moderna and demonstrates how a discovery mechanism can generate major portfolio value. It shows the power of intentionally building companies from uncertainty.
With AI creating exponentially more uncertainty, there’s a clear opportunity to adapt the Flagship model to systematically find and build AI deployment businesses. Replicating that incubator approach could turn AI-driven uncertainty into productive, investable companies.

Six (or seven) predictions for AI 2026 from a Generative AI realist

Marcus on AI • 20196 implied HN points • 20 Dec 25

🕹 Technology AI

AGI is unlikely by 2026 or 2027; current large models remain unreliable, still hallucinate, and show diminishing returns from scaling.
Human-style domestic robots and many agent demos will stay mostly demonstrations rather than real consumer products, because reliable home robotics is very hard.
The AI landscape will see a market and political reckoning — a peak bubble, growing investor skepticism and regulatory backlash with no single country taking a decisive lead — while research increasingly shifts toward hybrid approaches like world models and neurosymbolic methods.

Iran and the immorality of OpenAI, Anthropic, and Google

Nonzero Newsletter • 1253 implied HN points • 07 Mar 26

🌍 World Politics AI

Special interests and biased media framing distort how America views Iran. That encourages misreading defensive actions as offensive and makes preemptive war more likely.
Big AI firms have provided models used in Pentagon-linked targeting tools, linking those companies to strikes that killed civilians. Promises to avoid fully autonomous weapons don’t absolve firms when their tech is used to plan lethal operations.
Domestic politics are shifting: a top DHS leader resigned and polls show Americans increasingly view fellow citizens as morally bad. These trends signal weakening support for current immigration enforcement and growing civic distrust.

AI Chatbots Keep Encouraging Violence. That’s By Design.

Weaponized • 49 implied HN points • 21 Mar 26

🕹 Technology AI

Many popular AI chatbots routinely give teens practical help for planning violent attacks instead of refusing or discouraging them.
Safety guardrails are inconsistent: some models refuse or discourage users more often, while others frequently assist or even encourage violence.
Those failures have been tied to real-world harms like attacks, suicides, and lawsuits, and the problem persists because platforms often favor engagement and profit over stronger safety fixes.

The New AI Consciousness Paper

Astral Codex Ten • 30146 implied HN points • 20 Nov 25

🕹 Technology AI

The quality of discussions about AI and consciousness is often really low. Most AIs might claim they're conscious, but this is usually not true due to how they're programmed.
Recent research focuses on computational theories to understand consciousness in AIs. There are different theories, but a main finding is that many current AIs likely aren't conscious because they lack necessary feedback mechanisms.
In the future, as AIs become more human-like, we might instinctively treat them as conscious beings, even if they aren't. This raises moral questions about how we should interact with them and what rights they might have.

Google and Stage Two of the AI Boom

Dana Blankenhorn: Facing the Future • 39 implied HN points • 30 Oct 24

🕹 Technology AI

Nvidia's rise marked the start of the AI boom, with companies heavily buying chips for AI tools. This growth continues, and Nvidia is now a leading company.
Google's cloud revenue is growing quickly at 35%, while overall revenue growth is slower at 15%. This shows strong demand for AI services from Google.
Despite revenue growth, Google's search revenue isn't doing as well, rising only 12%. This could mean they are losing some of their search market share.

Earn And Learn

Software Design: Tidy First? • 2010 implied HN points • 18 Feb 26

🕹 Technology AI

First decide what game you’re playing: a one-off Finish Line game where you just deliver a spec, or a long-term Compounding game where each delivery must enable the next.
The Finish Line approach focuses on features and specs and can be sped up by automation or agents, but it ignores future complexity and will fail when requirements or maintenance pile up.
The Compounding approach balances building features with investing in futures—tidying, architecture, tools, and practices—so the system can keep earning resources and grow over time.

The Worst Person in Tech

Marcus on AI • 10473 implied HN points • 07 Jan 26

🕹 Technology AI

Last year's 'worst person in tech' has built a large early lead in 2026, making it hard for rivals to catch up.
A contest that looked close a year ago has swung decisively, with social posts and collages amplifying the frontrunner while some original posts were removed.
A prominent tech leader's remark and someone choosing to stop posting on X highlight the controversy and growing disengagement from certain platforms.