|
📖 Read In Depth
|
Is AI ruining our skills? Early results are in – and they're not good
Nature covers early empirical results on whether AI tool use degrades human skills — exactly the kind of question that matters for practitioners who care about craft. The framing isn't just productivity vs. atrophy; it gets at whether reliance reshapes cognitive habits in measurable ways. Relevant both as an ML engineer and as someone who builds things from scratch to understand them.
hn/Best Stories
|
GPT-5.5 hallucinates 3x more than MIT-licensed GLM-5.2
Claims GPT-5.5 hallucinates 3x more than the MIT-licensed GLM-5.2 — a striking benchmark result if it holds up. Worth reading to understand the methodology and whether the comparison is apples-to-apples, since hallucination rates are highly task-dependent and this could be either a genuine finding or a misleading benchmark.
hn/Best Stories
|
Zero-Touch OAuth for MCP
The MCP team introduces zero-touch OAuth for enterprise deployments — agents can acquire and manage credentials without user intervention. This is infrastructure-level work that matters for anyone building agentic systems; the auth layer is one of the real unsolved problems in production MCP deployments.
hn/Best Stories
|
There are no instances in ATProto
Dan Abramov (of React fame) explains a fundamental architectural distinction in ATProto — the protocol underlying Bluesky — where the concept of 'instances' doesn't apply the way it does in ActivityPub/Mastodon. This is a substantive systems design piece about how decentralized identity and data portability work at the protocol level, from someone who thinks carefully about abstractions.
hn/Best Stories
|
Project Valhalla, Explained: How a Decade of Work Arrives in JDK 28
Project Valhalla — value types for the JVM — is finally landing in JDK 28 after a decade of work. This piece explains what it actually changes: eliminating the object header overhead for small types, enabling flat array storage, and closing the performance gap between Java and C++ for numeric-heavy workloads. Relevant if you care about how language runtimes are designed.
hn/Best Stories
|
AI Engineer Claims to Have Cracked Linear A
An AI engineer claims to have cracked Linear A, the undeciphered Minoan script that has resisted interpretation for over a century. Extraordinary claim — worth reading to evaluate the methodology and evidence quality. If legitimate, this would be a major interdisciplinary result; if not, it's a useful case study in AI overclaiming on pattern-matching tasks.
hn/Best Stories
|
Where to Find the Colors Your Screen Can't Show You
A genuinely interesting piece on the gap between human color perception and display gamuts — covering how to actually experience colors outside sRGB in the physical world (specific pigments, lights, phenomena). Connects perceptual science to display hardware in an accessible but non-dumbed-down way.
hn/Best Stories
|
|
⚡ FYI
|
In the span of 3 days: Noam Shazeer (Transformer co-author) leaves Google for OpenAI, and John Jumper (Nobel laureate, AlphaFold lead) leaves Google DeepMind for Anthropic
In 72 hours: Noam Shazeer (Transformer co-author, returned to Google from Character.AI) has now left for OpenAI, and John Jumper (Nobel laureate, AlphaFold lead) is leaving Google DeepMind for Anthropic. This is a significant talent signal — the people who built Google's foundational AI work are actively choosing competitors, which aligns with reports that DeepMind is struggling to keep pace.
reddit/r/singularity
|
DeepMind is now reportedly struggling to compete with Anthropic and OpenAI while 3.5 Pro is not the step change they'd need to be competitive
Credible AI Twitter sources suggest Gemini 3.5 Pro won't be the step-change needed for DeepMind to close the gap with Anthropic and OpenAI — good for creative tasks but reportedly weak on agentic coding. Take with a grain of salt, but the Shazeer/Jumper departures lend it some credibility.
reddit/r/singularity
|
‘We created a monster’: companies rein in AI usage as costs strain budgets
Amazon, Walmart, Cisco, Uber, and Meta are introducing AI usage caps as costs at scale strain budgets. This is the other side of the AI hype cycle — the unit economics don't pencil out for many enterprise use cases, and companies are discovering the difference between demo value and deployment cost.
reddit/r/singularity
|
Hyundai buys Boston Dynamics
Hyundai has bought out SoftBank's remaining stake in Boston Dynamics for $325M, taking full control. This consolidates the robotics stack under a single industrial manufacturer — Hyundai can now integrate BD's locomotion capabilities directly into its manufacturing and logistics operations without a financial partner with misaligned incentives.
hn/Best Stories
|
Trump Says Apple to Buy Computer Chips from Intel
Trump says Apple will buy chips from Intel — neither company has confirmed, but the U.S. government already owns a 10% stake in Intel from a prior deal. If true, this would be a politically engineered supply chain shift with real implications for Intel's recovery and Apple's custom silicon strategy, which has been entirely TSMC-dependent.
nyt/Technology
|
Swiss parliament lifts ban on new nuclear power plants
Switzerland's parliament has lifted its ban on new nuclear power plant construction, reversing a policy direction set after Fukushima. This is part of a broader European reassessment of nuclear as AI data center energy demand makes decarbonization targets harder to hit with renewables alone.
hn/Best Stories
|
Google workspace threatening to block Firefox access
Google Workspace is apparently flagging Firefox as a non-supported browser and threatening to block access — a significant platform power move that raises questions about browser monoculture and Google's incentives to push Chrome adoption through enterprise software lock-in.
hn/Best Stories
|
Scientists just cured severed spinal cords (in fish) with a novel technique.
Researchers combined neural progenitor stem cells with magnetoelectric nanoparticles into 'NPCbots' that can be guided by magnetic fields to promote spinal cord regeneration — demonstrated in zebrafish and mice. Early-stage but mechanistically novel: using external magnetic fields to direct both cell placement and differentiation simultaneously.
reddit/r/singularity
|