davewiner's subscription list, hackerNewsStars category. List created by feedlandDatabase v0.9.1.
Troy Hunt
• Troy Hunt
Building out an IoT environment is a little like the old Maslow's Hierarchy of Needs. All the stuff on the top is only any good if all the stuff on the bottom is good, starting with power. This week, I couldn't even get that right, but
In 2025, Reinforcement Learning from Verifiable Rewards (RLVR) emerged as the de facto new major stage to add to this mix. By training LLMs against automatically verifiable rewards across a number of environments (e.g. think math/code puzzles), the LLMs spontaneously develop strategies that look like "reasoning" to humans - they learn to break down problem solving into intermediate calculations and they learn a number of problem solving strategies for going back and forth to figure things out (see DeepSeek R1 paper for examples).
— Andrej Karpathy, 2025 LLM Year in Review
Tags: andrej-karpathy, llm, generative-ai, llm-reasoning, definitions, ai, llms, deepseek
Sam Rose explains how LLMs work with a visual essay
Sam Rose is one of my favorite authors of explorable interactive explanations - here's his previous collection.Sam joined ngrok in September as a developer educator. Here's his first big visual explainer for them, ostensibly about how prompt caching works but it quickly expands to cover tokenization, embeddings, and the basics of the transformer architecture.
The result is one of the clearest and most accessible introductions to LLM internals I've seen anywhere.
Tags: ai, explorables, generative-ai, llms, sam-rose, tokenization
Krebs on Security
• BrianKrebs
The Trump administration has pursued a staggering range of policy pivots this past year that threaten to weaken the nation’s ability and willingness to address a broad spectrum of technology challenges, from cybersecurity and privacy to countering disinformation, fraud and corruption. These shifts, along with the president’s efforts to restrict free speech and freedom of the press, have come at such a rapid clip that many readers probably aren’t even aware of them all.
Ik ben eigenlijk op vakantie, maar toch een korte blogpost. Het is (in)formatietijd, en er worden plannen gemaakt. Eerder dit jaar kwam het ministerie van Economische Zaken met het idee een Nationaal Agentschap Disruptieve Innovatie op te richten, zoals ook al bestaat in Duitsland, en het meest beroemd in Amerika. Ik schreef hier eerder al een artikeltje over met meer details: Klassieke manieren om wetenschap en innovatie te stimuleren zijn erg ongeschikt voor disruptieve innovatie, ideeën die misschien helemaal niet gaan werken, maar als ze dat wel doen de wereld kunnen veranderen.
The art of not (directly) coming back: tail call optimisation
Kyle ran some benchmarks to compare the different implementations:
- Rust (html5ever) total parse time: 303 ms
- Swift total parse time: 1313 ms
- JavaScript total parse time: 1035 ms
- Python total parse time: 4189 ms
Tags: html5, ai, generative-ai, llms, ai-assisted-programming, vibe-coding, swift
My new hobby project.
DYNOMIGHT
• dynomight
because truth is many
Pluralistic: Daily links from Cory Doctorow
• Cory Doctorow
Jeff Geerling
• Jeff Geerling
Apple gave me access to this Mac Studio cluster to test RDMA over Thunderbolt, a new feature in macOS 26.2. The easiest way to test it is with Exo 1.0, an open source private AI clustering tool. RDMA lets the Macs all act like they have one giant pool of RAM, which speeds up things like massive AI models.
Jeff Geerling
• Jeff Geerling
Apple gave me access to this Mac Studio cluster to test RDMA over Thunderbolt, a new feature in macOS 26.2. The easiest way to test it is with Exo 1.0, an open source private AI clustering tool. RDMA lets the Macs all act like they have one giant pool of RAM, which speeds up things like massive AI models.
The stack of Macs I tested, with 1.5 TB of unified memory, costs just shy of $40,000, and if you're wondering, no I cannot justify spending that much money for this. Apple loaned the Mac Studios for testing. I also have to thank DeskPi for sending over the 4-post mini rack containing the cluster.
Inlining doesn't have to be all-or-nothing
