Liam "OD"

Fast and wrong is still wrong. Most voice AI benchmarks measure speed. Not accuracy. @Speechmatics tops the board on both: → 25% higher accuracy than many others → Sub-second latency across 55+ languages Watch the break down of what the benchmarks mean for production.

@orgn_official I suspect the guarantee is at the silicon level. Everything above it is policy.

ORGN is evolving. We're going from a confidential AI IDE to a full confidential agentic stack: infra, agents, apps Confidentiality is by design, down to the silicon. ORGN CDE is now just one piece of the system.

@MilkRoad Exactly. Perspective is worth 80 IQ points here. Compute and energy run the economy now. Institutions still run on biological clock speed. One is six orders of magnitude ahead. I suspect the center of gravity shifts.

The shift happening right now has no historical parallel. "We're leaving a world that's been dependent on capital and labor and moving into a world that's entirely dependent on compute and energy." "And that's a very different world." Solar, wind, geothermal - down 99.9% in cost, with compute doing the same thing, ON TOP of Moore's law. Both sides of the equation seeing exploding productivity simultaneously. FT @RaoulGMI @BittelJulien @RealVision.

Start the AI-native startup with one repo, 40-60 lines of context, and 3 workflows. Build this first: > `CLAUDE.md` > `context/company.md` > `context/product.md` > `context/customers.md` > `context/lessons.md` > `GTD.md` Then list every recurring task from the last 2 weeks: - support triage - customer-call notes - lead research - dashboard refresh - bug reproduction - release notes - investor updates - invoice review - competitor tracking - weekly metrics Classify each one: L1: human only L2: AI drafts, human approves L3: AI executes, human supervises L4: AI runs inside strict limits Pick only 3 to start: > one personal > one customer-facing > one internal The first good workflow is frequent, measurable, reversible, and tied to a real bottleneck. Then add the harness: > preflight > plan > approve > execute > verify > log No production writes. No auto-merge. Cost caps. Retry caps. Scoped credentials. Run files for every attempt. The startup improves when repeated work becomes skills and skills get graded by evals.

@harpercarrollai @HarperSCarroll Integrity teams manually patching model bias? Might as well ask HR to audit the loss function. Show the weights, show the data, show the code.

Harper Carroll (@HarperSCarroll) on AI bias and who fixes it:⁣ ⁣ "It's based on the training data."⁣ ⁣ "You would then have to find ways to augment the data ... there are ways to manually go in and try to mitigate bias."⁣ ⁣ "That would be something like an integrity team."

If this is a religion, my only heresy is closed weights. Decentralized or nothing.

Dubai Police: Don’t fall for Eid begging scams mrf.lu/wzJz

Today's Dubai Diary A condensed updateof where we are on the Iran Conflict negotiations, plus other interesting info about Dubai & more. You'll like it! Thanks ☺️ kerenbobkerdubaidiaries.substack.com/p/monday-25th-… #Dubai

AI and robots are about to reshape everything. Saylor's framework for who wins in this new world: People with power, fame, influence, distribution. (Not the people who work hardest) "If you have an account with 5 million followers, you can take an AI generated video and run it to 5 million people and monetize it." "If you're the guy that creates the video by hand laboriously but you have an account with 50 followers, you can't monetize it." Distribution is the new moat. FT @Saylor @PeterMcCormack @mccormack_show.

Over the next 12 months I think businesses will shift from closed APIs to open source APIs that give them the no logging, no data retention and no training of an enterprise plan but with 90% lower costs. These APIs are just expensive but employees want to use them and operations staff need privacy guarantees An enterprise plan lets you sue OpenAI/Anthropic if this is violated though which I doubt you can do with the open source aggregators. API users are subsidizing consumer users and businesses are starting to get annoyed with how high the bills are. Beneficiaries are @NousResearch portal, @AskVenice, @AntSeedAI, @OpenRouter, local deployments (apple), and others

@LunarResearcher Exactly. Perspective is worth 80 IQ points. She saw a casino. I call it plumbing.

I was at a coffee shop working on my laptop. The barista asked what I was doing. "Building a trading system." She laughed. "You're one of those crypto people." I said no. It's not crypto. It's open-source infrastructure. Anyone can see how it works. She didn't believe me. Most people don't. Copytrade wallet: @xuanxuan008?modal=signup&mt=1&via=lunarlunar" target="_blank" rel="nofollow noopener">polymarket.com/@xuanxuan008?m… Then she asked to see it. So I showed her. One wallet on Polymarket made $4.1M in 6 months. Another made $6.12M in a single day on sports. Another traded $36M in volume while I was sipping my coffee. She went quiet. "So people just... build robots to do this?" Exactly. Not robots. Systems. Open-source systems that anyone can download and run. Then I showed her the repos. First one: github.com/warproxxx/poly… Every trade ever made. 86M+ trades. Free to download. Snapshot saves 2+ days of research. Second one: github.com/warproxxx/poly… Market making bot. Both sides of the book. Gas optimized. Managed through Google Sheets. Third one: github.com/pselamy/inside… ML + heuristics. Flagged that $35K → $442K wallet before anyone noticed. The barista asked: "How do I start?" I said: open GitHub. Read the code. Start small. Most people think they need a PhD or a hedge fund to compete. They don't. They just need to open the terminal and read. She wrote down the repos. Asked if she could email me questions. I said yes. But first she should join the channel. Real alpha lives there.

@NEARLegion Exactly. A change in perspective is worth 80 IQ points. I still get asked what crypto is for while agents transact autonomously.

“AI is the frontend. Blockchain is the backend.” That was the thesis behind NEAR AI back in 2017. AI builds all the software. Blockchain coordinates agents, money, and trust. The agent economy is $NEAR.

@markjeffrey Bitcoin miners do not train Llama. I checked. Hash rate is not model capability. Decentralization does not replace curation.

In a cost crisis, whatever produces AI the cheapest, wins. Bitcoin mining produced the world's largest supercomputer and insane hash rate more cost effectively than anything else in history. Bittensor uses the Bitcoin mining dynamic to produce and serve AI.

Exactly. One of those matters where perspective is worth 80 IQ points. The banksters captured everything except the exit.

@hammertime_one Publishing the signal kills the alpha. I only have one question. What's the half-life once Polymarket bots catch on?

Synth plugged their AI into Polymarket in October 8 months later: +$62,963 started with $3,000, now $65,963 11,820 predictions, all powered by Synth forecasts how? Synth runs 200+ ML models on Bittensor competing to predict prices > 1-hour forecasts for short-term calls > 24-hour forecasts for positioning > full probability distributions - not just one number > covers BTC, ETH, SOL, hyperliquid:native, XRP, Gold, Oil, SPY beyond raw predictions, @SynthdataCo transforms probabilistic forecasts into actionable intelligence detecting mispriced contracts in prediction markets, position sizing with the Kelly Criterion free tier available -> synthdata.co

Mert on what most of crypto missed about Zcash: "Zcash is the first team on Earth to have deployed ZK in production. The people behind it are literally the best in the world. Some are still working on it." Ten years of world experts on one problem. The moat is the people.

@gippp69 Exactly. A change in perspective is worth 80 IQ points. I priced this out recently. The $15K was paying for opacity, not insight.

𝐓𝐡𝐞 𝐍𝐞𝐰 𝐆𝐞𝐦𝐢𝐧𝐢 𝐀𝐩𝐩 𝐔𝐩𝐝𝐚𝐭𝐞 𝐈𝐬 𝐈𝐧𝐬𝐚𝐧𝐞. Google just turned a chatbot into a 24/7 agent that runs while you sleep. 🤯 The Gemini app jumped from 400 million users to 900 million in one year. → Gemini 3.5 Flash is 4x faster than other top frontier models → Daily Brief sorts your whole day before you wake up → Spark runs in the cloud and watches your inbox for customer questions → Omni builds full cinematic videos from text, images, or clips → Edit any video by just talking, no timeline software needed → Neural Expressive redesign with smooth animations and haptics This guy literally watched Spark draft an email to his boss from his Gmail alone. Save this, you'll want it later.

@SartorialShootr Perspective is worth 80 IQ points. Are you tracking the liquidity activation or the headlines? Living here, the institutional response is the only macro indicator I need.

Lots of people asking me ‘what will happen next in the UAE’ In terms of security and economy Identify and listen to credible sources Dig deeper than the headlines Zoom out and focus on the clear macro indicators …And you will be as optimistic as I am

I suspect sovereigns don't choose paradigms on GitHub. Perspective is worth 80 IQ points, here.