|
📖 Read In Depth
|
Exploiting the most prominent AI agent benchmarks
Berkeley researchers show how the most prominent AI agent benchmarks can be systematically exploited — meaning leaderboard wins may not reflect real capability. For someone who builds from scratch to understand systems deeply, this cuts right at the question of whether the field's measurement infrastructure is trustworthy at all.
hn/Best Stories
|
Small models also found the vulnerabilities that Mythos found
After Mythos demonstrated that large frontier models can find real vulnerabilities, this piece examines whether smaller models can do the same — probing the jagged capability frontier in AI security. The highest-upvoted HN story in the batch, which signals real signal among technical readers.
hn/Best Stories
|
All elementary functions from a single binary operator
A theoretical CS paper deriving all elementary mathematical functions from a single binary operator — a kind of minimalist foundation for computation. The kind of thing someone who builds algorithms from scratch to understand them would find genuinely satisfying to work through.
hn/Best Stories
|
The peril of laziness lost
Bryan Cantrill on how engineering laziness — the instinct to avoid unnecessary work — is being eroded in the AI-assisted coding era. This is a thoughtful craft-level argument about what gets lost when shortcuts become default, from someone with serious systems engineering credibility.
hn/Best Stories
|
Apple's accidental moat: How the "AI Loser" may end up winning
The argument is that Apple's 'AI loser' label misses its actual moat: end-to-end hardware/software control, privacy as a differentiator, and on-device inference at scale. A genuine moats-and-incentives analysis rather than hype — directly relevant to how platform shifts play out.
hn/Best Stories
|
Pro Max 5x quota exhausted in 1.5 hours despite moderate usage
Claude Code's Pro Max tier quota is getting exhausted in 90 minutes for moderate usage — a window into Anthropic's real infrastructure economics and the tension between pricing, compute costs, and developer expectations. Pair with the cache TTL downgrade issue below for a fuller picture.
hn/Best Stories
|
447 TB/cm² at zero retention energy – atomic-scale memory on fluorographane
Researchers claim 447 TB/cm² storage density at zero retention energy using atomic-scale memory on fluorographane — orders of magnitude beyond current NAND flash. If it holds up, this is a physics-level breakthrough with long-term implications for compute and memory architecture.
hn/Best Stories
|
Bring Back Idiomatic Design (2023)
A 2023 essay resurfacing with 600+ HN points arguing that software design has drifted from platform-native idioms toward generic, framework-imposed patterns. Resonates with the craft-of-building perspective — the case for writing software that feels native to its environment.
hn/Best Stories
|
I run multiple $10K MRR companies on a $20/month tech stack
The highest-upvoted HN story in the batch: a solo developer running profitable businesses on a deliberately minimal stack. Relevant as a counterpoint to the build-vs-buy and platform-complexity questions — what does defensible simplicity actually look like in practice?
hn/Best Stories
|
The economics of software teams: Why most engineering orgs are flying blind
An analysis of why most engineering organizations lack meaningful economic models for their teams — no visibility into where value is created or destroyed. Bridges software craft and business incentives in a way that's useful for anyone thinking critically about how engineering orgs actually work.
hn/Best Stories
|
China’s Electrostate Is Poised to Win From War in the Middle East
China's vertically integrated energy-tech sector — batteries, grids, renewables — is positioned to benefit from the Middle East energy crisis in ways that compound its existing industrial lead. Connects geopolitics, hardware supply chains, and industrial policy in a way that goes beyond the obvious oil-price story.
nyt/Business
|
The future of everything is lies, I guess – Part 5: Annoyances
Kyle Kingsbury (of Jepsen fame) continues his series on AI-generated content and misinformation degrading the information environment — this installment focuses on everyday annoyances that compound into systemic trust erosion. Cantankerous, precise, and worth reading.
hn/Best Stories
|
|
âš¡ FYI
|
Anthropic downgraded cache TTL on March 6th
Anthropic quietly downgraded the prompt cache TTL on March 6th, significantly increasing costs for heavy Claude Code users. Combined with the quota exhaustion issue, this suggests Anthropic is still working out the unit economics of agentic coding tools at scale.
hn/Best Stories
|
4 Takeaways From Orban’s Defeat in Hungary’s Election
Viktor Orbán has been defeated in Hungary's election by Peter Magyar, ending over a decade of illiberal rule. The summary covers the four key takeaways: economics, corruption, JD Vance's failed endorsement visit, and what it means for EU cohesion and Ukraine aid.
nyt/Top Stories
|
European Airports Warn of Jet Fuel Shortages if Strait of Hormuz Remains Shut
European airports are warning of systemic jet fuel shortages within three weeks if the Strait of Hormuz blockade continues — a concrete, near-term consequence of the Iran conflict that goes beyond financial market volatility. Worth tracking as an infrastructure stress test.
nyt/Business
|