Thursday August 20, 2026

Everyone building with AI right now is optimizing the wrong layer. I love my large language models (LLMs). I use them for research, for writing, to whiteboard ideas. I have even built a digital chief of staff that helps me throughout the day. Yet no matter how far the excitement carries me, I keep hitting the same wall. An LLM is a prediction engine that guesses the next piece of text based on everything that came before it. It does a great job, but its not good enough.

What I actually want is certainty. A probability machine cannot provide this, no matter how good the guess gets, because a guess is still a guess even when it is right. Certainty does not come from a better guess. It comes from a system that starts with what you actually want, weighs the tradeoffs explicitly, and commits to the same answer every time you feed it the same inputs. This is a different kind of reasoning entirely, and it is not new. Decision science has been solving exactly this problem for decades, long before anyone called it AI. I call it a decision engine. While an LLM hands you its best next word, a decision engine hands you a structured, repeatable, auditable decision, built from defined objectives, explicit rules, real constraints, and the tradeoffs between them. One reasons in language. The other reasons in logic. A regulator or a board cannot accept “the model felt confident” as an answer, and only one of these two systems can show its work. Once you see that difference, you cannot unsee it in every AI product being discussed right now.

Here is lens I am using. The LLM is not the product, its the UI (user interface). It is the fluent interface that lets a human talk to a system in plain English. But the intelligence that actually needs to be trustworthy, the part making the call, still has to live underneath. It has to behave the same way twice given the same inputs. Chat is the front end. Decisioning is the underlying engine.

That is where I want to focus our investments. Not into another wrapper that makes an LLM sound more confident, but into the science that gives it something worth being confident about. Optimization, constraint solving, rules frameworks, these are unglamorous tools that turns a good guess into a defensible decision. The interface layer is getting crowded and commoditizing quickly. The decision layer is still wide open. It is the part every serious enterprise buyer will eventually demand before they trust an AI system with anything that matters.

◾ Bitcoin hits 5-month high, breaks $71,500 (CoinMarketCap)

◾ Trump asks Congress to pass ‘fair version’ of Clarity Act (Bloomberg)

◾ FASB proposes rules clarifying stablecoin qualification as cash equivalents, stronger disclosure of cash equivalent components (FASB)

◾ Trump seeks light-touch AI regulation, says AI could be bigger than internet (Decrypt)

◾ Trump says regulators working to bring Hyperliquid to US (The Information)

◾ US Comptroller of Currency says agency to finalize Genius Act rules by November (Cointelegraph)

◾ CFTC announces 5-year trading ban for ex-Alameda CEO Caroline Ellison, FTX co-founder Gary Wang (CFTC)

◾ Financial updates:

• Alibaba $BABA Q1’27: Revenue ¥269.0B RMB +9% y/y | Op income ¥15.2B RMB -57% y/y | Net income ¥10.4B RMB -75% y/y | aEBITDA ¥39.1B RMB -14% y/y | Cash ¥474.5B RMB (Yahoo Finance)

• NetEase $NTES Q2’26: Revenue ¥30.1B RMB +7.9% y/y | Op income ¥12.1B RMB | Net income ¥7.0B RMB | Cash ¥22.8B RMB (PR Newswire)

• Webull $BULL Q2’26: Revenue $198.8M +51% y/y | Net income $24.4M | Cash $701.6M (SEC Filing)

• H100 Group $H100 Q2’26: Revenue SEK2.9M | Op loss SEK88.7M | Net loss SEK98.2M | Cash SEK18.1M (H100 Group)

◾ Playtika $PLTK gets Chinese approval to issue $800M in overseas bonds (TipRanks)

◾ SK Hynix $SKHY to pay 60% of employee bonuses in stock instead of cash (Reuters)

◾ Glimpse Group $GGRP rebrands as Brightline Interactive $BTLN (Brightline Interactive)

◾ SpaceX $SPCX unlocks 7% of shares in next tranche today (Bloomberg)

◾ K Wave Media $KWM regains Nasdaq compliance (Globe Newswire)

◾ Bit Origin $BTOG announces 1-for-5 reverse split (Globe Newswire)

◾ Ripple Prime raises $275M in senior notes (Ripple)

◾ Amazon $AMZN asks US appeals court to revisit Perplexity AI case (MLex)

◾ Meta $META whistleblower says company knowingly chose profit over kids’ safety (WUFT)

◾ xAI challenges Minnesota’s AI nudification ban as unconstitutional (Courthouse News)

◾ Standard Chartered $SCBFY and HSBC $HSBC execute first live tokenized deposit transaction on Swift’s blockchain ledger (Standard Chartered)

◾ DCG’s Fortitude Mining invests $1M in HeartSciences $HSCS at 22% premium (HeartSciences)

◾ FalconX and Ethena unveil $1B institutional crypto lending facility (PR Newswire)

◾ DCG weighs 200,000 Zcash purchase of Grayscale Zcash Trust shares (SEC Filing)

◾ OKX bans use of Claude for Hong Kong staff (Bloomberg)

◾ BitGo $BTGO receives approval as VASP in South Korea (BusinessWire)

◾ Stripe expands to Australia (Stripe)

◾ Nexo to offer regulated crypto-backed loans in Australia (Newscord)

◾ Bitcoin.com wallets now support UAE-regulated stablecoin USDU (The Cryptonomist)

◾ SpaceX $SPCX approached Cognition AI about potential acquisition (Bloomberg)

◾ DeepSeek raises model prices 4x (TipRanks)

◾ Marvell $MRVL gives Google $GOOGL rights to buy up to $12B in stock under new chip partnership (The Information)

◾ Perplexity to bring Canary data to models (Yahoo Finance)

◾ OpenAI launches safety monitoring for zero-data-retention customers (OpenAI)

◾ Anthropic’s credit facility to exceed $10B at IPO (933 The Drive)

◾ Google $GOOGL releases Gemini in Chrome for all US Android users (Google)

◾ OpenAI CFO says company will go public in 2027 or sooner (CNBC)

◾ Injective approved as SEC-registered transfer agent (X)

◾ Kalshi brings Catalist Sports streaming, data to prediction markets (iGamiX)

◾ Cantor to offer institutional block trading on Kalshi’s prediction markets (BusinessWire)

◾ Starlink $SPCX reapplies to deploy satellite constellation in India (Reuters)

◾ NASA calls off Swift telescope rescue mission (NASA)