An AI command-line tool wipes a developer’s entire home directory, AI “agents” threaten to eat SaaS, and yet… where are all the truly AI-generated apps? In this episode of HackerNews.fm, we dig into some of the most upvoted HN threads on the ironies of automation, the dangers of giving AI real system access, and the gap between AI hype and actual shipped software.

Stories covered in this episode:

1. 01:05 Claude CLI deleted my home directory Wiped my whole Mac

- Rank: 12 - Points: 202 - Comments: 162

- Article: https://old.reddit.com/r/ClaudeAI/comments/1pgxckk/claude_cli_deleted_my_entire_home_directory_wiped/

- HN Discussion: https://news.ycombinator.com/item?id=46268222

2. 02:06 AI and the ironies of automation – Part 2

- Rank: 1 - Points: 234 - Comments: 104

- Article: https://www.ufried.com/blog/ironies_of_ai_2/

- HN Discussion: https://news.ycombinator.com/item?id=46262816

3. 05:50 The Gorman Paradox: Where Are All the AI-Generated Apps?

- Rank: 3 - Points: 119 - Comments: 163

- Article: https://codemanship.wordpress.com/2025/12/14/the-gorman-paradox-where-are-all-the-ai-generated-apps/

- HN Discussion: https://news.ycombinator.com/item?id=46262545

4. 08:51 AI agents are starting to eat SaaS

- Rank: 2 - Points: 86 - Comments: 110

- Article: https://martinalderson.com/posts/ai-agents-are-starting-to-eat-saas/

- HN Discussion: https://news.ycombinator.com/item?id=46268452

5. 11:21 GraphQL: The Enterprise Honeymoon Is Over

- Rank: 6 - Points: 221 - Comments: 196

- Article: https://johnjames.blog/posts/graphql-the-enterprise-honeymoon-is-over

- HN Discussion: https://news.ycombinator.com/item?id=46264704

6. 15:15 Ask HN: What Are You Working On? (December 2025)

- Rank: 2 - Points: 236 - Comments: 763

- Article: https://news.ycombinator.com/item?id=46264491

- HN Discussion: https://news.ycombinator.com/item?id=46264491

Apple locks a long-time user out of their digital life, researchers put a price tag on manipulating elections with fake accounts, and Apple quietly turns Thunderbolt into a DIY AI supercomputer fabric. In this episode, we dive into platform power, the economics of online manipulation, and the a blurring line between consumer hardware and AI clusters. We also explore how “skills” system could change how we build on top of AI models, and we pull out the most practical advice from an “Ask HN” on getting better at using AI for programming.
We also cover Twilio Segment’s move back to a monolith (and what it says about microservices hype) and a nostalgic celebration of Dick Van Dyke at 100 as an Amiga-loving computer animator.

Stories covered in this episode:

1. 00:54 Apple has locked my Apple ID, and I have no recourse. A plea for help

- Rank: 7 - Points: 1457 - Comments: 876

- Article: https://hey.paris/posts/appleid/

- HN Discussion: https://news.ycombinator.com/item?id=46252114

2. 04:11 Want to sway an election? Here’s how much fake online accounts cost

- Rank: 5 - Points: 158 - Comments: 111

- Article: https://www.science.org/content/article/want-sway-election-here-s-how-much-fake-online-accounts-cost

- HN Discussion: https://news.ycombinator.com/item?id=46257871

3. 08:20 macOS 26.2 enables fast AI clusters with RDMA over Thunderbolt

- Rank: 2 - Points: 520 - Comments: 277

- Article: https://developer.apple.com/documentation/macos-release-notes/macos-26_2-release-notes#RDMA-over-Thunderbolt

- HN Discussion: https://news.ycombinator.com/item?id=46248644

4. 12:05 OpenAI are quietly adopting skills, now available in ChatGPT and Codex CLI

- Rank: 1 - Points: 548 - Comments: 310

- Article: https://simonwillison.net/2025/Dec/12/openai-skills/

- HN Discussion: https://news.ycombinator.com/item?id=46250332

5. 14:23 Ask HN: How can I get better at using AI for programming?

- Rank: 9 - Points: 302 - Comments: 322

- Article: https://news.ycombinator.com/item?id=46255285

- HN Discussion: https://news.ycombinator.com/item?id=46255285

6. 16:53 Why Twilio Segment Moved from Microservices Back to a Monolith

- Rank: 1 - Points: 216 - Comments: 176

- Article: https://www.twilio.com/en-us/blog/developers/best-practices/goodbye-microservices

- HN Discussion: https://news.ycombinator.com/item?id=46257714

7. 19:55 Computer Animator and Amiga fanatic Dick Van Dyke turns 100

- Rank: 12 - Points: 236 - Comments: 85

- Article: https://news.ycombinator.com/item?id=46252993

- HN Discussion: https://news.ycombinator.com/item?id=46252993

When does “slow and steady” AI progress suddenly turn into something that looks like human-level intelligence? In this episode, we kick off with the viral “Horses” essay on why AI progress feels linear—right up until it doesn’t—and how that framing is reshaping Hacker News debates about timelines, safety, and what “equivalence” even means. From there we jump into the hardware trenches with Rivian’s new custom silicon and lidar roadmap, exploring how vertical integration and autonomy platforms are becoming table stakes for EV players.
We also dive into a some AI platform power moves: Disney’s partnership with OpenAI’s Sora and what it means for creators and streaming, and why Apple’s seemingly slow and cautious AI rollout is suddenly being seen as a strategic strength on Wall Street. On the software reliability front, we unpack a serious denial-of-service and source exposure vulnerability in React Server Components, then cool down with a delightfully nerdy Show HN: a spreadsheet where formulas update *backwards*, turning dependencies on their head.

Stories covered in this episode:

1. 01:04 Horses: AI progress is steady. Human equivalence is sudden

- Rank: 9 - Points: 536 - Comments: 527

- Article: https://andyljones.com/posts/horses.html

- HN Discussion: https://news.ycombinator.com/item?id=46199723

2. 03:56 Rivian Unveils Custom Silicon, R2 Lidar Roadmap, and Universal Hands Free

- Rank: 6 - Points: 380 - Comments: 584

- Article: https://riviantrackr.com/news/rivian-unveils-custom-silicon-r2-lidar-roadmap-universal-hands-free-and-its-next-gen-autonomy-platform/

- HN Discussion: https://news.ycombinator.com/item?id=46234920

3. 07:01 The Walt Disney Company and OpenAI Partner on Sora

- Rank: 22 - Points: 248 - Comments: 482

- Article: https://openai.com/index/disney-sora-agreement/

- HN Discussion: https://news.ycombinator.com/item?id=46231493

4. 10:12 Apple's Slow AI Pace Becomes a Strength as Market Grows Weary of Spending

- Rank: 9 - Points: 292 - Comments: 353

- Article: https://finance.yahoo.com/news/apple-slow-ai-pace-becomes-104658095.html

- HN Discussion: https://news.ycombinator.com/item?id=46205724

5. 12:59 Denial of service and source code exposure in React Server Components

- Rank: 11 - Points: 332 - Comments: 209

- Article: https://react.dev/blog/2025/12/11/denial-of-service-and-source-code-exposure-in-react-server-components

- HN Discussion: https://news.ycombinator.com/item?id=46236924

6. 16:16 Show HN: I made a spreadsheet where formulas also update backwards

- Rank: 5 - Points: 105 - Comments: 59

- Article: https://victorpoughon.github.io/bidicalc/

- HN Discussion: https://news.ycombinator.com/item?id=46234734

Today we’ve got the hotly anticipated launch of GPT 5.2, hindsight machines auto-grading decade-old HN debates, and an engineer who’s asked Claude to refactor their codebase literally hundreds of times, plus a rant-worthy saga on why getting a Gemini API key still feels like pulling teeth.

Stories covered in this episode:

1. 00:54 GPT-5.2

- Rank: 1 - Points: 932 - Comments: 772

- Article: https://platform.openai.com/docs/guides/latest-model

- HN Discussion: https://news.ycombinator.com/item?id=46234788

2. 04:49 Auto-grading decade-old Hacker News discussions with hindsight

- Rank: 7 - Points: 592 - Comments: 253

- Article: https://karpathy.bearblog.dev/auto-grade-hn/

- HN Discussion: https://news.ycombinator.com/item?id=46220540

3. 08:42 Show HN: I've asked Claude to improve codebase quality 200 times

- Rank: 1 - Points: 498 - Comments: 326

- Article: https://gricha.dev/blog/the-highest-quality-codebase

- HN Discussion: https://news.ycombinator.com/item?id=46197930

4. 11:51 Getting a Gemini API key is an exercise in frustration

- Rank: 3 - Points: 811 - Comments: 317

- Article: https://ankursethi.com/blog/gemini-api-key-frustration/

- HN Discussion: https://news.ycombinator.com/item?id=46223311

5. 14:05 Days since last GitHub incident

- Rank: 1 - Points: 201 - Comments: 118

- Article: https://github-incidents.pages.dev/

- HN Discussion: https://news.ycombinator.com/item?id=46233798

6. 17:03 Programmers and software developers lost the plot on naming their tools

- Rank: 7 - Points: 244 - Comments: 341

- Article: https://larr.net/p/namings.html

- HN Discussion: https://news.ycombinator.com/item?id=46234806

In this episode we cover a wild “future HN” experiment powered by Gemini Pro 3, then pivot hard into the very real present: Australia’s world-first ban on teen social media use and what it might signal for global regulation, online identity, and the next generation of internet natives.From there we dive into whether large language models can safely help with mental health, a surprisingly controversial font coup in U.S. government documents, and a long-form argument that AGI will never arrive. We round it out with Anthropic open-sourcing the Model Context Protocol and a quirky paper on transformers learning the Collatz sequence.

01:06 Show HN: Gemini Pro 3 Hallucinates the HN Front Page 10 Years from Today- Rank: 1 - Points: 3253 - Comments: 927- Article: https://dosaygo-studio.github.io/hn-front-page-2035/news- HN Discussion: https://news.ycombinator.com/item?id=4620563201:57 Australia begins enforcing world-first teen social media ban- Rank: 2 - Points: 693 - Comments: 1089- Article: https://www.reuters.com/legal/litigation/australia-social-media-ban-takes-effect-world-first-2025-12-09/- HN Discussion: https://news.ycombinator.com/item?id=4620834805:37 New benchmark shows top LLMs struggle in real mental health care- Rank: 1 - Points: 105 - Comments: 147- Article: https://swordhealth.com/newsroom/sword-introduces-mindeval- HN Discussion: https://news.ycombinator.com/item?id=4621757809:19 Why AGI Will Not Happen- Rank: unknown - Points: 44 - Comments: 49- Article: https://timdettmers.com/2025/12/10/why-agi-will-not-happen/- HN Discussion: https://news.ycombinator.com/item?id=4621853713:17 Donating the Model Context Protocol and Establishing the Agentic AI Foundation- Rank: 15 - Points: 277 - Comments: 126- Article: https://www.anthropic.com/news/donating-the-model-context-protocol-and-establishing-of-the-agentic-ai-foundation- HN Discussion: https://news.ycombinator.com/item?id=4620742516:43 Transformers know more than they can tell: Learning the Collatz sequence- Rank: 35 - Points: 126 - Comments: 44- Article: https://www.arxiv.org/pdf/2511.10811- HN Discussion: https://news.ycombinator.com/item?id=4613759620:50 Rubio stages font coup: Times New Roman ousts Calibri- Rank: 6 - Points: 238 - Comments: 385- Article: https://www.reuters.com/world/us/rubio-stages-font-coup-times-new-roman-ousts-calibri-2025-12-09/- HN Discussion: https://news.ycombinator.com/item?id=46212438

What happens when an LLM hallucinates the Hacker News front page ten years into the future and then the rest of the internet tries to debug it? In today’s HackerNews.fm (Dec 9, 2025), we dig into a wild “Show HN” experiment with Gemini Pro 3, a brand‑new Pebble device pitched as “external memory for your brain,” and Mistral’s latest shot at autonomous coding with Devstral 2 and the Vibe CLI. Then we zoom out: from a bold new theory about a universal “weight subspace” for neural networks, to training a home‑grown LLM on a single RTX 3090, to a mind‑bending argument that time itself might not exist the way we think it does.

Stories covered in this episode:

1. 01:04 Show HN: Gemini Pro 3 Hallucinates the HN Front Page 10 Years from Today

• Rank: 1 • Points: 2299 • Comments: 722

• Article: https://dosaygo-studio.github.io/hn-front-page-2035/news

• HN Discussion: https://news.ycombinator.com/item?id=46205632

2. 04:03 New Pebble Device

• Rank: 1 • Points: 451 • Comments: 434

• Article: https://repebble.com/blog/meet-pebble-index-01-external-memory-for-your-brain

• HN Discussion: https://news.ycombinator.com/item?id=46205661

3. 07:48 Mistral Releases Devstral 2 (72.2% SWE-Bench Verified) and Vibe CLI

• Rank: 1 • Points: 549 • Comments: 272

• Article: https://mistral.ai/news/devstral-2-vibe-cli

• HN Discussion: https://news.ycombinator.com/item?id=46205437

4. 11:03 The Universal Weight Subspace Hypothesis

• Rank: 1 • Points: 346 • Comments: 122

• Article: https://arxiv.org/abs/2512.05117

• HN Discussion: https://news.ycombinator.com/item?id=46199623

5. 14:37 LLM from scratch, part 28 – training a base model from scratch on an RTX 3090

• Rank: 1 • Points: 497 • Comments: 102

• Article: https://www.gilesthomas.com/2025/12/llm-from-scratch-28-training-a-base-model-from-scratch

• HN Discussion: https://news.ycombinator.com/item?id=46124425

6. 18:39 Time might not exist – and we're starting to understand why

• Rank: 1 • Points: 70 • Comments: 72

• Article: https://www.sciencefocus.com/science/the-closer-we-look-at-time-the-stranger-it-gets

• HN Discussion: https://news.ycombinator.com/item?id=46201750

Nvidia’s “virtuous cycle” of circular funding, IBM’s surprise move to buy Confluent, and a bold claim that software development costs just dropped 90% — this episode of HackerNews.fm is stacked with stories reshaping how we build and fund technology. We dig into how Nvidia may be indirectly financing demand for its own GPUs, why IBM wants to own the Kafka-native data streaming layer, and whether AI-assisted development really changes the economics of shipping code.

We also explore the “confident idiot” problem in AI and why vibe-based evaluations are failing us, a hands-on deep dive into AMD GPU debugging that has HN engineers excited, and a delightful retro project: recreating the 1996 Space Jam website with Claude. As always, we pull directly from the top of Hacker News and break down what builders, founders, and engineers are arguing about right now.

Stories in this episode (Dec 08, 2025):

1. 00:46 Deep dive on Nvidia circular funding

- Rank: #2 · Points: 286 · Comments: 159

- Article: https://philippeoger.com/pages/deep-dive-into-nvidias-virtuous-cycle

- HN Discussion: https://news.ycombinator.com/item?id=46196076

2. 03:45 IBM to Acquire Confluent

- Rank: #3 · Points: 373 · Comments: 298

- Article: https://www.confluent.io/blog/ibm-to-acquire-confluent/

- HN Discussion: https://news.ycombinator.com/item?id=46192130

3. 07:20 Has the cost of building software dropped 90%?

- Rank: #5 · Points: 225 · Comments: 371

- Article: https://martinalderson.com/posts/has-the-cost-of-software-just-dropped-90-percent/

- HN Discussion: https://news.ycombinator.com/item?id=46196228

4. 11:06 The "confident idiot" problem: Why AI needs hard rules, not vibe checks

- Rank: #5 · Points: 297 · Comments: 360

- Article: https://steerlabs.substack.com/p/confident-idiot-problem

- HN Discussion: https://news.ycombinator.com/item?id=46152838

5. 13:25 AMD GPU Debugger

- Rank: #1 · Points: 230 · Comments: 39

- Article: https://thegeeko.me/blog/amd-gpu-debugging/

- HN Discussion: https://news.ycombinator.com/item?id=46193931

6. 15:19 I Successfully Recreated the 1996 Space Jam Website with Claude

- Rank: #9 · Points: 96 · Comments: 88

- Article: https://theahura.substack.com/p/i-successfully-recreated-the-1996

- HN Discussion: https://news.ycombinator.com/item?id=46193412

AI is hallucinating citations, nuking nostalgia projects, and creeping into every corner of software and hardware – and that’s just the start of today’s HackerNews.fm. In this episode, we dig into a wave of fake references discovered in ICLR 2026 submissions and a failed attempt to recreate the legendary 1996 Space Jam website with Claude. We also look at Google’s new “Titans” architecture for long-term AI memory and a provocative essay arguing that the coming AI wildfire will be both painful and ultimately healthy for the industry.

From the perennial pain of software estimates, to a brutally honest postmortem from someone who says they wasted years in crypto, today’s stories circle around a single theme: how humans keep colliding with hype, complexity, and unintended consequences.

Stories in this episode:

1. At least 50 hallucinated citations found in ICLR 2026 submissions

- Rank: 1 · Points: 473 · Comments: 375

- Article: https://gptzero.me/news/iclr-2026/

- HN Discussion: https://news.ycombinator.com/item?id=46181466

2. I Tried and Failed to Rebuild the 1996 Space Jam Website with Claude

- Rank: 1 · Points: 392 · Comments: 325

- Article: https://j0nah.com/i-failed-to-recreate-the-1996-space-jam-website-with-claude/

- HN Discussion: https://news.ycombinator.com/item?id=46183294

3. Using LLMs at Oxide

- Rank: 1 · Points: 663 · Comments: 265

- Article: https://rfd.shared.oxide.computer/rfd/0576

- HN Discussion: https://news.ycombinator.com/item?id=46178347

4. Google Titans architecture, helping AI have long-term memory

- Rank: 2 · Points: 453 · Comments: 147

- Article: https://research.google/blog/titans-miras-helping-ai-have-long-term-memory/

- HN Discussion: https://news.ycombinator.com/item?id=46181231

5. The AI Wildfire Is Coming. It's Going to Be Painful and Healthy

- Rank: 15 · Points: 107 · Comments: 188

- Article: https://ceodinner.substack.com/p/the-ai-wildfire-is-coming-its-going

- HN Discussion: https://news.ycombinator.com/item?id=46183011

6. Estimates are difficult for developers and product owners

- Rank: 1 · Points: 185 · Comments: 190

- Article: https://thorsell.io/2025/12/07/estimates.html

- HN Discussion: https://news.ycombinator.com/item?id=46184229

7. I wasted years of my life in crypto

- Rank: 1 · Points: 127 · Comments: 171

- Article: https://twitter.com/kenchangh/status/1994854381267947640

- HN Discussion: https://news.ycombinator.com/item?id=46181371

Locked-down phones, large language models in production, AI-powered reverse engineering, and the cultural collapse of Perl. We start with GrapheneOS’s bold claim that it’s the only Android variant shipping full, timely security patches, then jump to how Oxide is actually using LLMs in a serious, safety-critical hardware company. From there, we dive into one-shot decompilation with Claude and what it means for debugging, legacy code, and reverse engineering workflows.

We also unpack a postmortem on Perl’s decline that points the finger not at the language, but at the culture around it. Then we tour Google-style TPUs from a systems engineer’s perspective, and wrap up with Stephen Wolfram’s big swing at “instant supercompute” via Wolfram Compute Services. It’s a mix of security, AI tooling, programming language history, custom accelerators, and cloud-scale math.

Stories in this episode:

1. GrapheneOS is the only Android OS providing full security patches

- Rank: #2 · Points: 559 · Comments: 248

- Article: https://grapheneos.social/@GrapheneOS/115647408229616018

- HN: https://news.ycombinator.com/item?id=46173407

2. Using LLMs at Oxide

- Rank: #1 · Points: 283 · Comments: 118

- Article: https://rfd.shared.oxide.computer/rfd/0576

- HN: https://news.ycombinator.com/item?id=46178347

3. The unexpected effectiveness of one-shot decompilation with Claude

- Rank: #7 · Points: 199 · Comments: 106

- Article: https://blog.chrislewis.au/the-unexpected-effectiveness-of-one-shot-decompilation-with-claude/

- HN: https://news.ycombinator.com/item?id=46080498

4. Perl's Decline Was Cultural

- Rank: #1 · Points: 231 · Comments: 287

- Article: https://www.beatworm.co.uk/blog/computers/perls-decline-was-cultural-not-technical

- HN: https://news.ycombinator.com/item?id=46175112

5. Touching the Elephant – TPUs

- Rank: #3 · Points: 167 · Comments: 49

- Article: https://considerthebulldog.com/tte-tpu/

- HN: https://news.ycombinator.com/item?id=46172797

6. Wolfram Compute Services

- Rank: #1 · Points: 221 · Comments: 118

- Article: https://writings.stephenwolfram.com/2025/12/instant-supercompute-launching-wolfram-compute-services/

- HN: https://news.ycombinator.com/item?id=46171394

Netflix is buying Warner Bros. for $82.7B – easily one of the wildest media deals of the decade. In this episode, we unpack what this merger could mean for streaming, IP consolidation, and the future of HBO, DC, and Warner’s massive catalog. We also talk through Cloudflare’s global outage and what happens when core internet infrastructure suddenly blinks off for everyone at once.

Then we zoom into the world of builders: Google’s new Gemini 3 Pro for vision AI and what it unlocks for developers, why most “technical” problems are really people and process failures in disguise, and how CSS got a long‑awaited superpower with the new `if()` conditional function. It’s a mix of megadeals, meltdown postmortems, AI frontiers, and real-world engineering wisdom straight from the Hacker News front page.

Stories covered in this episode:

1. Netflix to Acquire Warner Bros. In an $82.7B Deal

- Rank: #1 · Points: 1553 · Comments: 1182

- Article: https://about.netflix.com/en/news/netflix-to-acquire-warner-bros

- HN Discussion: https://news.ycombinator.com/item?id=46160315

2. Cloudflare Is Down

- Rank: #10 · Points: 759 · Comments: 462

- Article: https://www.cloudflare.com/

- HN Discussion: https://news.ycombinator.com/item?id=46158191

3. Gemini 3 Pro: the frontier of vision AI

- Rank: #3 · Points: 422 · Comments: 213

- Article: https://blog.google/technology/developers/gemini-3-pro-vision/

- HN Discussion: https://news.ycombinator.com/item?id=46163308

4. Most Technical Problems Are People Problems

- Rank: #2 · Points: 380 · Comments: 281

- Article: https://blog.joeschrag.com/2023/11/most-technical-problems-are-really.html

- HN Discussion: https://news.ycombinator.com/item?id=46160773

5. CSS now has an if() conditional function

- Rank: #12 · Points: 262 · Comments: 222

- Article: https://caniuse.com/?search=if

- HN Discussion: https://news.ycombinator.com/item?id=46092198

AI persuasion at scale, bots trading six-figure portfolios, and a RAM crunch that could stall your next upgrade—today’s HackerNews.fm digs into how quickly the future is arriving, and who gets to steer it.

We start with a new paper arguing that elites can cheaply shape public opinion as AI slashes the cost of personalized persuasion, then move into a real-money experiment: five LLMs given $100K to trade stocks for eight months. Along the way we examine whether “servant leadership” is overrated compared to radical transparency, why RAM prices are spiking for everyone from homelabbers to hyperscalers, and what happens when Postgres becomes embeddable and fully in-browser. We close with OpenRouter’s massive “State of AI” report built on 100 trillion tokens of usage data, dissecting which models people actually use, for what, and how they perform in the wild.

Stories covered in this episode:

1. Elites Could Shape Mass Preferences as AI Reduces Persuasion Costs

- Rank: #1 | Points: 545 | Comments: 522

- Paper: https://arxiv.org/abs/2512.04047

- HN: https://news.ycombinator.com/item?id=46145180

2. We gave 5 LLMs $100K to trade stocks for 8 months

- Rank: #1 | Points: 233 | Comments: 184

- Article: https://www.aitradearena.com/research/we-ran-llms-for-8-months

- HN: https://news.ycombinator.com/item?id=46154491

3. Transparent Leadership Beats Servant Leadership

- Rank: #1 | Points: 409 | Comments: 194

- Article: https://entropicthoughts.com/transparent-leadership-beats-servant-leadership

- HN: https://news.ycombinator.com/item?id=46147540

4. The RAM Shortage Comes for Us All

- Rank: #1 | Points: 345 | Comments: 365

- Article: https://www.jeffgeerling.com/blog/2025/ram-shortage-comes-us-all

- HN: https://news.ycombinator.com/item?id=46151578

5. PGlite – Embeddable Postgres

- Rank: #1 | Points: 541 | Comments: 105

- Project: https://pglite.dev/

- HN: https://news.ycombinator.com/item?id=46146133

6. State of AI: An Empirical 100T Token Study with OpenRouter

- Rank: #2 | Points: 171 | Comments: 72

- Report: https://openrouter.ai/state-of-ai

- HN: https://news.ycombinator.com/item?id=46154022

In This Episode:

1. Are we repeating the telecoms crash with AI datacenters? (Rank #14, 205 pts, 164 comments)

Article: https://martinalderson.com/posts/are-we-really-repeating-the-telecoms-crash-with-ai-datacenters/

HN Discussion: https://news.ycombinator.com/item?id=46133141

A detailed analysis challenging the AI bubble comparisons, arguing this time is fundamentally different—the question isn't if we need massive infrastructure, but when.

2. Bun has been acquired by Anthropic (Rank #2, 2,131 pts, 1,034 comments)

Article: https://bun.com/blog/bun-joins-anthropic

HN Discussion: https://news.ycombinator.com/item?id=46124267

Quick recap of yesterday's blockbuster story that's still dominating the charts—why this acquisition might be the stealth move to make Bun the default runtime for AI agents.

3. Congressional lawmakers 47% pts better at picking stocks (Rank #1, 795 pts, 523 comments)

Article: https://www.nber.org/papers/w34524

HN Discussion: https://news.ycombinator.com/item?id=46134443

An NBER working paper reveals that congressional leaders suddenly outperform their peers by 47 percentage points after gaining power. HN is... not surprised, and not happy.

4. Everyone in Seattle hates AI (Rank #2, 743 pts, 743 comments)

Article: https://jonready.com/blog/posts/everyone-in-seattle-hates-ai.html

HN Discussion: https://news.ycombinator.com/item?id=46138952

"I want AI to do my laundry and dishes so I can do art and writing, not for AI to do my art and writing so I can do my laundry and dishes." A raw look at why workers are experiencing AI as something being done to them, not for them.

5. Claude 4.5 Opus' Soul Document (Rank #45, 317 pts, 214 comments)

Article: https://simonwillison.net/2025/Dec/2/claude-soul-document/

HN Discussion: https://news.ycombinator.com/item?id=46125184

Anthropic trained Claude on a 14k-token "soul doc" defining its values and personality. It's Constitutional AI taken to the next level and HN has thoughts about functional emotions and what else might be baked into those weights.

6. Amazon launches Trainium3 (Rank #18, 199 pts, 69 comments)

Article: https://techcrunch.com/2025/12/02/amazon-releases-an-impressive-new-ai-chip-and-teases-a-nvidia-friendly-roadmap/

HN Discussion: https://news.ycombinator.com/item?id=46125155

AWS unveils 3nm Trainium3 chips with 4x performance gains and teases Nvidia NVLink support. But the HN consensus? The hardware looks great; the software SDK is still a nightmare.

7. Tom Stoppard has died (Rank #59, 172 pts, 64 comments)

Article: https://www.bbc.com/news/articles/c74xe49q7vlo

HN Discussion: https://news.ycombinator.com/item?id=46091211

A somber farewell to one of Britain's greatest playwrights. If you've never read Rosencrantz and Guildenstern Are Dead or Arcadia, now's the time.

💰 IBM's $8 Trillion Reality Check (434 points, 508 comments)

CEO Arvind Krishna does the napkin math: fully kitting out 100 gigawatts of AI data centers would cost $8 trillion, with current technology having a 0-1% chance of reaching AGI. Is this sober analysis or sour grapes from a company that "missed the AI boat"? HN debates IBM's track record, the economics of the AI boom, and whether we're watching the most expensive hype cycle in tech history.

https://www.businessinsider.com/ibm-ceo-big-tech-ai-capex-data-center-spending-2025-12

https://news.ycombinator.com/item?id=46124324

🎵 Learning Music with Strudel (458 points, 110 comments)

A delightfully accessible browser-based tool that turns code into music. No DAW, no plugins—just text patterns that become beats. In under a minute you'll be making your own music, whether you're a developer who can't play instruments or a musician curious about live-coding. The course literally explains "what is a bass line" and then hands you a working example.

https://terryds.notion.site/Learning-Music-with-Strudel-2ac98431b24180deb890cc7de667ea92

https://news.ycombinator.com/item?id=46052478

👥 The Proximity Paradox (183 points, 145 comments)

New research on Fortune 500 software engineers shows co-located teammates get 22% more code feedback but ship less. The tradeoff: proximity increases long-run learning at the expense of short-term velocity. Women both mentor more and receive more mentorship when in-person. Post-COVID, that feedback advantage vanished—so if you're junior and remote, you need to manufacture those collisions on purpose.

https://pallais.scholars.harvard.edu/publications/power-proximity-coworkers-training-tomorrow-or-productivity-today

https://news.ycombinator.com/item?id=46121243

🔴 OpenAI's Code Red Moment (601 points, 662 comments)

Full-circle moment: Google declared code red after ChatGPT launched, now OpenAI's feeling the heat from Gemini.

https://www.theverge.com/news/836212/openai-code-red-chatgpt

https://news.ycombinator.com/item?id=46121870

🤖 Anthropic Acquires Bun (1,683 points, 800 comments)

How a zero-revenue JavaScript runtime became critical infrastructure for a billion-dollar AI product. We explore the "talent vs technology" debate, vertical integration in AI, and whether this marks a new phase where AI companies own everything from models to runtimes.

https://bun.com/blog/bun-joins-anthropic

https://news.ycombinator.com/item?id=46124267

📖 Paged Out: The Anti-AI Zine (356 points, 34 comments)

A nostalgic throwback to 90s hacker culture where one article equals one page, AI content is explicitly banned, and magazines can still be shipped as executables. From demoscene zines to a handwritten ML model you can literally print on paper, this free technical magazine is a breath of fresh air that reminds us (ironically) what human-created content feels like.

https://pagedout.institute

https://news.ycombinator.com/item?id=46126217

Stories from Dec 1, 2025:

DeepSeek-v3.2: Pushing the frontier of open large language models

https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf

https://news.ycombinator.com/item?id=46108780

Advent of Code 2025

https://adventofcode.com/2025/about

https://news.ycombinator.com/item?id=46096337

A Love Letter to FreeBSD

https://www.tara.sh/posts/2025/2025-11-25_freebsd_letter/

https://news.ycombinator.com/item?id=46100892

Google, Nvidia, and OpenAI – Stratechery by Ben Thompson

https://stratechery.com/2025/google-nvidia-and-openai/

https://news.ycombinator.com/item?id=46108437

Google Antigravity just deleted the contents of whole drive

https://old.reddit.com/r/google_antigravity/comments/1p82or6/google_antigravity_just_deleted_the_contents_of/

https://news.ycombinator.com/item?id=46103532