Kansas Assistant Jeremy Case Interviews for Arizona State Head Coaching Job - roundtable.ioKansas Assistant Jeremy Case Interviews for Arizona State Head Coaching Job roundtable.io
The Quiet Normalization of Artificial IntelligenceOver the past few years, it has become almost routine to hear warnings about artificial intelligence. Headlines often focus on the risks. Job displacement, algorithmic manipulation, surveillance, loss of control over autonomous systems. In public debates, AI frequently appears as a looming threat, something powerful and unpredictable that could reshape society in dangerous ways.Yet at the very same time that these fears dominate the conversation, something quieter has been unfolding. Millions of people have already woven artificial intelligence into their daily routines. They rely on algorithms to help write emails, ask digital assistants for research guidance, receive recommendations about what to watch or read, and use AI-powered tools to accelerate their work. The criticism grows louder, but the habit grows stronger.The data reveals a striking contradiction. A global study involving tens of thousands of participants found that around 66 percent of people worldwide already use artifi
Ask HN: AI Agents vs. Gateways vs. HarnessesHi,I'd like to get everyone's take on the different components in the AI Agents ecoysystem. I find the current terminology quite confusing as it's not always obvious what I'll be actually getting when I examine the available options.A lot of things get called Agents, but that term seems hard to define as it often refers to overlapping functionality. To me, agents seem to be currently composed of the following components:## Harnesses- Adds UI and system instructions around an LLM and may also augment it with tools such as memory, tool calls, etc...- Examples are Claude Code, Code, Gemini CLI, pi.dev, ...## Gateways- These connect agents to your communication tool of choice, e.g. Whatsapp, Telegram, Slack, Discord, ...- Examples are OpenClaw and Nanoclaw## Sandboxes- Isolated environments where Agents can run with limited or auditable capabilities- Examples can be from physical ones like separate Mac Minis to docker-agent, agent-sandbox, localsandbox, ...This then takes us to *Agents*:##
Trump is trapped by the logic of war Boots on the ground might become inevitable - UnHerdTrump is trapped by the logic of war Boots on the ground might become inevitable UnHerd
Ask HN: How do you prompt AI to talk to you?Like many, I use Claude for therapy/personal related stuff. I have a project set up which has a personal history, as well as things I talk about with my therapist, schemas, things I believe in and value, stuff like that. In challenging moments where there are things I want to process, I do voice recordings and use this project to reflect/process feelings and thoughts.In the project instructions, I tell it to not be sycophantic, to always talk to me with a grounding in science and evidence-based theory, and to hold me accountable and gently challenge me when needed.I find this good, but sometimes I feel like its responses are a bit boilerplate and generic.Curious how other, more techy people set up projects / etc to do similar sorts of things.
Notes from the trough of sorrow: why we killed our own productIn December 2024, Roundtable was a survey fraud detection company. Our product Alias analyzed typing patterns in open-ended text fields to catch bots and AI-generated responses. It worked and market research firms used it to flag the 30%+ of survey responses that are now AI-generated.Twelve months later, we open-sourced it and built something else.---Matt and I spent five years at Princeton studying how humans and machines process information differently. We've published in Science, PNAS, and Nature Human Behaviour on computational models of cognition. The thing we kept coming back to: humans overshoot mouse targets, slow down on the Stroop task, type with irregular timing. Language production is a cognitive process, not a copy-paste operation. AI agents don't produce these signals. And scaling AI doesn't close the gap. It widens it.Alias captured a narrow slice of this: typing patterns in text fields on surveys. Proof of Human captures the full surface: keystrokes, mouse movement, scr
Show HN: Prompt optimizer that refactors messy prompts using a P-C-V pipelineI built a small web tool that tries to automatically refactor messy prompts into structured prompt specifications.The motivation was simple: most prompts people write are poorly structured, which often leads to unstable AI responses.So I implemented a Proposer–Critic–Verifier optimization pipeline.Proposer restructures the prompt. Critic evaluates clarity, structure, and task definition. Verifier checks consistency. Arbiter detects convergence.The system runs several iterations and produces a structured prompt specification.It usually takes around 30–40 seconds to run the full optimization loop.The tool also includes: • prompt quality analysis • explain mode • version history • prompt diff viewerDemo: https://how-to-grab-me.vercel.app/Would love to hear if people building AI tools find something like this useful.
Ask HN: How do you find early users willing to bet on a rough product?I built an Instagram-focused AI tool for a client that is trained on a brand’s product catalog and internal documentation. The interesting part is not generic generation, but that it actually understands the product and brand context well enough to produce content that is much less slop-prone than normal prompt-based workflows.I’m now trying to abstract and productize it beyond the initial client.My idea is to onboard 5 to 10 ecommerce teams through a small “founder edition” style program, work closely with them, and use that to shape the product. The product is useful already, but still rough around the edges, so this is not a polished self-serve SaaS yet.I’m curious about two things:Where do you find the kind of early users who are comfortable betting on a product at this stage?Has anyone here tried a founder edition or lifetime-style early access offer, and did it work well or create bad incentives later?I’d especially like to hear from people who have sold early B2B tools before th
Trilateral Roundtable on Artificial Intelligence - Le gouvernement luxembourgeoisTrilateral Roundtable on Artificial Intelligence Le gouvernement luxembourgeois
Ask HN: General API for accessing bank transactions?I currently track expenses monthly by downloading credit card/checking account statements as CSVs and importing into excel. It's a bit annoying as I have to 1. wait for statements to post so I can't track the current month and 2. statements don't line up exactly with calendar months.I'm looking for a general way to access transactions across all my banks, but the ones I've found all have some issues.- Plaid: seems oriented more for businesses and isn't very friendly for personal use. They do have a personal use version but it doesn't support certain banks (e.g. Chase)- Teller.io: unfortunately doesn't support one of my accounts (Bilt)- Fintable.io: syncs transactions to airtable/google sheets which maybe could work, but doesn't have a simple API.Am I missing any alternatives? Plaid supports all the institutions I want but getting setup as a non-business seems to require too many steps. Maybe any consumer-facing wrappers around Plaid's API?
Ask HN: Value and demand for space-manufactured products?I'm exploring a startup idea around in-space manufacturing (ISM) in low Earth orbit using automated, modular payloads (free-flyers or commercial stations like Starlab/Axiom). The core thesis is leveraging microgravity to produce materials that are meaningfully superior to Earth-made versions in small-but-high-value batches (no need for mass production in orbit).Three specific product swimlanes that are viable (all hybrid model: grow superior version in space → return to Earth → use/expand on Earth):1. Pharma/biotech crystals: Novel polymorphs or ultra-pure API crystals (e.g., for monoclonal antibodies, small molecules) with better solubility, stability, bioavailability, or controlled release. Think reformulating IV drugs into injectables or enabling new patents on blockbusters.2. Semiconductor seeds: Ultra-pure wide/ultra-wide bandgap crystal seeds (GaN, SiC, AlN, etc.) grown defect-free in microgravity. These propagate into larger terrestrial boules with dramatically lower dislocation
Show HN: League Donation – Comprehensive Fantasy Baseball Analytics DashboardMost fantasy tools hand you a list. One source, one opinion, presented as settled fact. This one shows you the argument behind the list: where your sources agree, where they diverge, and what the divergence means. Source disagreement is the feature. A player ranked 40th by consensus but going in the 70s in ADP is a market inefficiency. A player ranked 40th by consensus but 90th by Statcast projections is a question worth asking. The tool surfaces those gaps rather than collapsing them into a clean number.Rankings and xStats work without a league connection. The xStats section identifies hitters whose surface stats diverge from their expected stats, players whose hard contact has been finding gloves at an unsustainable rate, and gives you the buy signal before the correction happens.Connect your league and it builds around your actual scoring context. Z-scores calibrated to your draftable pool. VORP measured against real positional replacement levels scaled to your league size. Tier clu
Show HN: EV range calculator – range circles and charger map (700 models)I built EV Mapping to answer the question EV owners actually ask: "Can I get there and back without charging?"You pick your car, drop a start pin, set your current charge %, and hit Compute Reach. You get two circles — outer for one-way range, inner for round-trip — plus all charging stations within reach as pins. Click any pin to open navigation in Google Maps.Stack: vanilla JS + Leaflet, no build step, hosted on Vercel. Charger data merges OpenChargeMap and OpenStreetMap Overpass with deduplication. 700+ EV presets across 20+ markets (US, EU, India, China, AU, etc.) synced every 12h via a Python pipeline on GitHub Actions.The catalog pipeline ingests fueleconomy.gov, afdc.energy.gov, cardekho.com, and greenvehicleguide.gov.au, plus region-native seed files. It has anti-regression guardrails and a canary→stable promotion workflow so bad syncs don't silently ship.https://ev-mapping.vercel.app Source: https://github.com/novelmartis/ev-mappingFeedback welcome — especially on range accura
36yo: Career at home vs. Simple life abroad?I am 36 years old, single, and currently unemployed, living with my parents in my home country. I am at a point that might define the next decade of my life. I am struggling with a choice between two paths that offer completely different types of security.Option A: Relocating to Southern Europe (Portugal)The Income: A low-skill remote role (Content Analysis) with night shifts (PST hours), paying ~1100 EUR. I also have some passive income to supplement this.The Lifestyle: Living in a studio or small apartmentin in smallish Portuguese town. For around 800 EUR.The Perspective: The move isn't about a specific career goal or a passport; it’s about the higher life standards, safety, and the stable social environment of Western Europe.The Trade-off: I would be far from my aging parents. I would be working an unskilled job that doesn't build professional equity, potentially living in studio at 36, which might be isolating during night shifts.Option B: Staying in my Home Country (Ankara, Turkey
Show HN: pg_stream – incremental view maintenance for PostgreSQL in Rustpg_stream is a PostgreSQL 18 extension that keeps materialized views current incrementally — no external infrastructure, no separate streaming pipeline.You define a stream table (a SQL query + a freshness bound) and the extension derives a delta query that processes only changed rows on each refresh cycle: SELECT pgstream.create_stream_table( 'regional_totals', 'SELECT region, SUM(amount) AS total, COUNT(*) AS cnt FROM orders GROUP BY region', '1m', 'DIFFERENTIAL' ); One INSERT into a million-row table → pg_stream touches one row's worth of computation. Query it like any regular table.How it works: - Trigger-based CDC captures row changes into buffer tables inside the same transaction (no wal_level = logical required) - A background worker walks a topological DAG of stream tables and fires refreshes in dependency order - The DVM engine (differential view maintenance, DBSP framework) rewrites your SQL into a delta query at definition time — JOINs, GROUP BY, CTEs, window functions, LATER
I built a global city safety database, here's what the data revealsOut of curiosity, I pulled together a city database combining safety stats from various places: crime records, how safe people feel walking around during daylight versus after dark, health care ratings, plus stories from neighborhood updates. The results? They tell a story words often miss.The Rankings Since standard tables don't render well here, here is the raw data parsed out:Rank City Country Score ----------------------------------------------- 1 Sapporo Japan 93.2/100 2 Nagoya Japan 91.0/100 3 Abu Dhabi UAE 88.9/100 4 Suzhou China 88.7/100 5 Hue Vietnam 87.6/100 6 Kyoto Japan 85.8/100 7 Fukuoka Japan 85.3/100 8 Doha Qatar 84.5/100 9 Dubai UAE 83.9/100 10 Taipei Taiwan 83.5/100 11 Kotor Montenegro 82.7/100 12 Queenstown New Zealand 82.5/100 13 Nuwara Eliya Sri Lanka 82.4/100 14 Ajaccio France 82.3/100 15 Dubrovnik Croatia 82.2/100 What Stood Out Japan Japan stands apart from most places. Four cities rank in the top seven: Sapporo, Nagoya, Kyoto, and Fukuoka, all above 85. Nighttim
‘Around the Table’: Sen. Cory Booker talks water crisis, income inequality - ABC News - Breaking News, Latest News and Videos‘Around the Table’: Sen. Cory Booker talks water crisis, income inequality ABC News - Breaking News, Latest News and Videos
Show HN: OctoFlow v1.0.0 – GPU VM where the GPU runs autonomously, CPU is BIOSThree days ago I posted OctoFlow 0.83 here (GPU-native programming language, 2.2 MB binary). The feedback was great. Since then I've pushed v1.0.0 with the thing I've actually been building toward: a GPU Virtual Machine.The idea: the GPU is the computer, the CPU is the BIOS.You boot a VM, program a dispatch chain of kernel instances, submit once with vkQueueSubmit, and everything — layer execution, inter-layer communication, self-regulation, compression, database queries — happens on the GPU without CPU round-trips. The CPU just provides I/O. let vm = vm_boot() let prog = vm_program(vm, kernels, 4) vm_write_register(vm, 0, 0, input) vm_execute(prog) let result = vm_read_register(vm, 3, 30) 4 VM instances, one submit, no CPU involvement between stages.The memory model is 5 SSBOs: Registers (per-VM working memory), Metrics (regulator signals), Globals (shared mutable — KV cache, DB tables), Control (indirect dispatch params), Heap (immutable bulk data — quantized weights).What makes it i
Show HN: RMirror Cloud – Open-Source OCR and Notion Sync for ReMarkable TabletsI built an open-source cloud service for reMarkable tablets that does three things:Syncs notebooks from your reMarkable to the cloud via a macOS background agent Runs AI-powered OCR on handwritten pages (Claude API, not Tesseract) Pushes transcribed notes to Notion as searchable pages Stack: FastAPI backend, Next.js dashboard, Python agent. Self-hostable. The reMarkable's handwriting recognition (via Connect) is okay for typed-looking text but struggles with real handwriting. I wanted something better that also gave me web access to my notes and pushed them to Notion where I actually work.GitHub: https://github.com/gottino/rmirror-cloud Request beta access: https://rmirror.io/Looking for beta testers — especially people with messy handwriting.
Show HN: KGBaby – A WebRTC based audio baby monitor I built on pat leaveBaby monitors are a great boon. We use the audio-only Motorola AM21, which is excellent. But being on pat leave with a 2-month-old right now, I decided to build a browser-based alternative using WebRTC and some AI coding agents (Codex & Gemini).It is an open-source, zero-latency P2P monitor. Hardware reuse: You can repurpose an old phone or tablet to be the child unit instead of buying single-purpose hardware. Actually private: Unlike using a never-ending Google Meet or Zoom, your stream stays private via WebRTC (PeerJS for signaling). No cloud routing. The backgrounding hack: Mobile Safari aggressively kills background audio. I used a hidden 1x1 base64 looping video to keep the microphone active when the screen dims.Links: Live Demo: https://legodud3.github.io/kgbaby/ Repo: https://github.com/legodud3/kgbabyWelcome all feedback!