|
📖 Read In Depth
|
AI got the blame for the Iran school bombing. The truth is more worrying
A Guardian investigation into the Iran school bombing finds that AI targeting systems aren't the culprit — the real issue is how automated targeting pipelines compress human decision-making into a few clicks, creating accountability gaps that are structurally harder to fix than any single algorithm. This is the intersection of ML systems, military ethics, and institutional incentives done seriously.
hn/Best Stories
|
Go hard on agents, not on your filesystem
A Stanford research piece arguing that the right abstraction for agentic AI systems is to go hard on agent orchestration and treat the filesystem as secondary — essentially a rethinking of how agents should interact with compute resources. High-signal for anyone building or reasoning about agentic systems from first principles.
hn/Best Stories
|
Anatomy of the .claude/ folder
A deep dive into the .claude/ folder structure — how Claude Code's project-level configuration, memory, and tool permissions actually work under the hood. Practical and architectural, directly relevant if you're building or evaluating agentic coding workflows.
hn/Best Stories
|
Smaller Is Better in Silicon Valley’s ‘Tiny Team’ Moment
Tech executives in Silicon Valley are embracing 'tiny teams' — sometimes just one person plus AI — as AI handles more engineering, design, and ops tasks. This is a real structural shift in how products get built, and the piece engages seriously with what it means for hiring, moats, and organizational design.
nyt/Business
|
The Oil Shocks of the ’70s Changed the World. Will the Iran War Do the Same?
A serious historical comparison between the 1970s oil shocks and the current Iran war disruption — examining whether Hormuz closure, $200 oil forecasts, and inflation feedback loops rhyme with what transformed global energy markets 50 years ago. Useful macro framing for understanding the current economic environment.
nyt/Business
|
The Ezra Klein Show: How Fast Will A.I. Agents Rip Through the Economy?
Anthropic co-founder Jack Clark talks with Ezra Klein about the economics of AI agents and how fast they'll actually reshape labor and industries — not hype, but Clark's specific views on deployment timelines, bottlenecks, and what 'agentic era' actually means in practice. Worth engaging with given Clark's direct access to frontier systems.
nyt/Technology
|
We rewrote JSONata with AI in a day, saved $500k/year
A concrete case study: a team rewrote JSONata (a JSON transformation language) from JavaScript to Go in a day using AI, saving $500k/year by eliminating a fleet of Node.js pods in their Go pipeline. The real lesson is architectural — they'd been running a cross-language RPC boundary for years that AI-assisted rewriting finally eliminated.
hn/Best Stories
|
|
🎬 Check It Out
|
Review - Project Hail Mary (2026) Reaches for Greatness But Repeatedly Trips Over Its Own Silliness
A TrueFilm review of the Ryan Gosling adaptation of Project Hail Mary (2026) — the reviewer found it uneven but worth seeing, with moments of genuine hard-science ambition alongside missteps. If you read the Weir novel, this is probably worth evaluating; currently in theaters.
reddit/r/TrueFilm
|
|
⚡ FYI
|
Fortune reports Anthropic testing a new model that is a “step change” and “poses unprecedented cybersecurity risks”
Fortune reports Anthropic accidentally exposed details of an unreleased model in a public database, describing it as a 'step change' that 'poses unprecedented cybersecurity risks.' The security lapse itself is notable; the model capabilities claim is unverified but worth tracking given Anthropic's safety-focused positioning.
reddit/r/singularity
|
If you don't opt out by Apr 24 GitHub will train on your private repos
GitHub is defaulting users into training Copilot on private repos, with an opt-out deadline of April 24. If you have private repos on GitHub, go to github.com/settings/copilot/features now. The opt-in-by-default framing is becoming a recurring pattern worth noticing.
hn/Best Stories
|
Judge blocks Pentagon effort to 'punish' Anthropic with supply chain risk label
A federal judge blocked the Pentagon's attempt to label Anthropic a 'supply chain risk' — a designation that would have restricted government contracts and created serious institutional pressure. The case signals that the government's relationship with frontier AI labs is increasingly adversarial, not just regulatory.
hn/Best Stories
|
Meta and YouTube Found Negligent in Landmark Social Media Addiction Trial
A jury found Meta and YouTube negligent and liable for addictive design features that harmed a young user's mental health — a landmark verdict that could open the door to widespread product liability claims against social platforms. This is the legal theory that Congress couldn't pass; juries may do it instead.
nyt/Technology
|
From 0% to 36% on Day 1 of ARC-AGI-3
A team (Symbolica AI) claims 36% on ARC-AGI-3 on day one, up from 0% baseline — if verified, a notable jump on a benchmark designed to resist pattern-matching. The GitHub repo is public; worth a quick look to assess methodology before treating this as signal.
reddit/r/singularity
|
DeepMind’s New AI Just Changed Science Forever
DeepMind's Aletheia is described as capable of conducting novel publishable mathematical research — distinct from existing models that solve structured olympiad problems. The distinction between 'solving well-posed problems' and 'posing and solving open research questions' is exactly the right one to track here.
reddit/r/singularity
|