Premarket movers: Mag 7 stocks are mostly lower (Alphabet -0.3%, Amazon -0.3%, Apple -0.08%, Meta +0.4%, Microsoft 0.0%, Nvidia +0.2%, Tesla -0.7%) Cryptocurrency-linked stocks climb as Bitcoin’s rally unleashed the biggest wave of short liquidations in records going back to 2021. The stocks were also boosted as President Donald Trump met with crypto executives from firms including Coinbase, Payward and Advance Auto (AAP) tumbles 15% after the parts provider reported second-quarter sales that fell short of analyst estimates. Alibaba ADRs (BABA) fall 3% after the Chinese internet giant reported weaker-than-expected revenues for its core domestic e-commerce business Coty (COTY) falls 14% after the beauty conglomerate refrained from providing full-year guidance, predicting a “transition” period in the current fiscal year. Ethan Allen (ETD) rises 3% after the home furnishings company declared a special cash dividend of $3 a share. Nordson (NDSN) rises 5% after the maker of applicators used to dispense adhesives boosted its adjusted earnings per share guidance for the full year. Ultragenyx Pharmaceutical (RARE) rises 7% after the drugmaker received accelerated FDA approval for its gene therapy to treat a rare genetic metabolic disorder. Valvoline (VVV) climbs 1% after getting a new bull as Benchmark starts coverage of the operator of quick-oil-change stations with a buy rating, saying the stock is at an attractive entry point for investors. Walmart (WMT) falls 6% as quarterly sales fell short of expectations, a rare miss that’s likely to stoke concern about the leading big-box retailer decelerating alongside a slow-growing US economy. Webull (BULL) climbs 11% after the digital investment platform reported second-quarter results that beat expectations. Wolfspeed (WOLF) falls 8% after the semiconductor-device company’s fourth-quarter results were seen as disappointing.

🇺🇸US DATA RUNS HOT: INFLATION STICKY, DEMAND FIRM INFLATION • Headline PCE MoM: +0.2% vs +0.1% est.; prior -0.1% • Headline PCE YoY: +3.7% vs +3.6% est.; prior +3.7% • Core PCE MoM: +0.2% vs +0.2% est.; prior +0.1% • Core PCE YoY: +3.3% vs +3.3% est.; prior +3.3% GDP • Q2 GDP annualized: +1.5% vs +1.5% est.; prior +1.5% • Personal consumption: +3.4% vs +3.2% est.; prior +3.2% • GDP Price Index: +6.4% vs +6.2% est.; prior +6.2% • Core PCE QoQ: +3.6% vs +3.4% est.; prior +3.4% INCOME & SPENDING • Personal income MoM: +0.4% vs +0.2% est.; prior +0.2% • Personal spending MoM: +0.2% vs +0.1% est.; prior +0.3% • Real personal spending MoM: 0.0% vs 0.0% est.; prior +0.4% DURABLE GOODS • Durable goods orders MoM: +1.1% vs +0.5% est.; prior +0.5% • Durables ex-transportation: +0.4% vs +0.6% est.; prior +0.7% • Core capital goods orders: +0.2% vs +0.7% est.; prior revised to +1.7% • Core capital goods shipments: +1.4% vs +1.0% est.; prior revised to +2.4% BOTTOM LINE: Inflation remains sticky while consumer demand and headline durable goods beat expectations. GDP growth was in line, but stronger price pressures could keep the Fed cautious on rates.

Bitcoin doesn’t need you to time the market. It needs you to stay in it. Across 5,749 daily closes through July 28, 2026: 1Y loss rate: 27.1% 3Y: 0.7% 5Y: 0.2% 10Y: 0.0% 2,099 rolling 10-year windows. Zero losses. Volatility is the price. Time is the edge:

🟥 Things I'm looking at today 🟥 Powered by @yeet Subscribe: Crypto lower ahead of CPI, STONK hits ATH, ETH leads majors Crypto * BTC: 77,790 (0%) | BTC.D: 59.0% (0%) * ETH: 2,500 (+1%) | BNB: 721 (+1%) | SOL: 101 (0%) * Fear & Greed: 57 | 24h Liq: $446m * BTC ETFs: -$283M | ETH ETFs: -$30M * Top Gainers: NIU LAI, STONK, RAY, PONS, NEAR * Crypto mixed, ETH leads BTC ETF outflows continue * BTC faces resistance at $83-85K: Glassnode * $72-73K is BTC's key support: CryptoQuant * BTC rally cools but golden cross nears * $400K BTC in 4 yrs still 'reasonable’: Coinbase * ETH’s Glamsterdam set for Sepolia on Oct 6 * Vitalik plan cuts quantum-safe privacy costs * X adds support for Robinhood Chain tickers * Robinhood Chain DEX liquidity tops $100M * Robinhood lists small memes incl. JUGGERNAUT * Pumpfun app removed from US iOS App Store * STONK bounces 2x to ATH as revenues surges * Solana sees ATH 263K tokens issued in one day * CLARITY Act adds DeFi rules ahead of vote * New text has 100+ Democrat changes: Lummis * Crypto market concentration at 2021 levels * Nasdaq invests $100M in Kraken parent at $21B * SBF asks SC to overturn conviction * Base App rebrands back to Coinbase Wallet * MoneyGram debuts stablecoin Visa card * Kalshi plans 24/7 Tesla, Apple, Nvidia perps * OKX intros 10x OpenAI, Anthropic perps in EU * Tokenised deposits = bank deposits: Canada’s OSFI * Danish central bank warns on stablecoin risks * India launches digital INR settlement for bonds * Aryadotag puts $2B India grain network on Aave * UK crypto strategy wins House of Lords backing * Russia to require tax IDs for crypto accounts * Nasdaq urges EU to scrap €100B tokenization cap Macro * NASDAQ: $26,080 (-0.7%) | Gold: $4,340 (+0.7%) * Oil continues march higher as Brent tops $111 * US diesel tops $6/gal for first time ever * US 10yr yield nears 5%, 30yr yield hits 5.38% * Traders fully price Fed hike by Oct after hot PPI * TLT collapses to lowest since 2004 * US-China 10Y yield gap hits record 324bps * $9.6T triple witching options expiry on Sep 18 * Gas to fall below $2/gal after midterms: Trump * WH brands $5K payments 'Trump Dividend' * US to sanction large bank Monday: Bessent * Iran attacks ships in Hormuz * Oil tanker rates hit ATHs on Mideast risk * Trump wont go 'full into Iran' due to midterms * Trump declines MBS request to strike Houthis * Houthis take Hisn Murad at Bab al-Mandeb * Qatar in talks to buy US LNG * Israel blows up Hezbollah tunnel network * Canada, Ukraine sign 100yr defence, trade deal * Ukraine drones hit Saratov refinery * ECB hikes rates by 25bps to 2.50% * Traders price 3 more ECB hikes by mid-2027 * China’s gold buys in Aug highest since 2023 * France seizes $260M cocaine at Dunkirk AI & Tech * Oracle rallies after market on earnings beat * OpenAI pauses new pro subs on Astra demand * OpenAI launches new ChatGPT for finance * OpenAI weighs slowing frontier AI work * 70% extinction risk in 3 yrs: OpenAI researcher * Let OpenAI into the energy grid: Altman * Iran used Claude to surveil thousands: Anthropic * Chinese labs trained on millions of Claude chats * China used Claude for PLA weapons work * Coxon's AI doom claims 'deeply untrue’: Jensen * AI doomers 'having AI psychosis’: Sacks * Rather American killer robots than Chinese: Cruz * DeepSeek nears GPT-6 Astra design at 1.4% cost * Nvidia could grow 70% next year: Jensen * DOJ probes Nvidia's Groq licensing deal * Palantir, Nvidia partner on AI for supply chains * SpaceX overhauls data centre build-out * First orbital computing sats next year: SpaceX * Microsoft aims to triple cloud capacity by 2032 * Universal Music, ElevenLabs launch AI platform * The Boring Company raises $3B in UAE-led round * Texas approves electric air taxi test flights * LG TVs caught scanning home networks * NASA eyes 2028 Moon return, IBM AI to help * Bending Spoons buys Miro for $1.36B @YEET: Crypto's Casino Just passed $3.4B in lifetime volume! gm

# Decision Points in AI Agent Development # Temperature 🎯 The Hook Are you using the same temperature for every task in your agent? Temperature isn't just a "creativity knob." In agent systems, structured outputs, tool calls, and user-facing responses each need fundamentally different temperature settings. Using one value for everything is leaving performance on the table. 📋 Overview Temperature controls how "peaked" or "flat" the probability distribution is when an LLM selects its next token. Near 0, the highest-probability token wins almost every time, producing deterministic and stable output. Higher values flatten the distribution, allowing lower-probability tokens through, increasing diversity and creativity. In AI agent systems, the optimal temperature varies dramatically across contexts: generating structured output, assembling tool call arguments, and producing free-form text each call for different settings. Temperature should be treated as a dynamic variable that shifts with task type, not a single fixed constant. 🔍 Decision Points Temperature is primarily driven by task_variability -- how routine vs. creative the task is. The decision flow is straightforward 🧭 1. Structured output (JSON / function calling)? → 0.0-0.3 2. Accuracy-first (fact extraction, classification, summarization)? → 0.2-0.5 3. Dialogue, explanation, communication? → 0.5-0.7 4. Creative writing, brainstorming, candidate generation? → 0.7-1.0 Additionally, higher failure_cost pushes the temperature ceiling down, and high cost_sensitivity environments should account for retry cost increases from higher temperatures. 💡 Key Details Reference values by task type 📊 - Structured output (JSON / function calling): 0.0-0.3. Minimizing schema violations is the priority - Classification, extraction, data transformation: 0.0-0.2. Accuracy and reproducibility are paramount - Summarization, explanation, customer support: 0.5-0.7. Balance naturalness with accuracy - Creative writing, brainstorming, candidate generation: 0.7-1.0. Diversity is the source of value - Tool argument generation: 0.0-0.2. Precise function and argument names are non-negotiable - Planning and reasoning: 0.3-0.6. Some exploration helps, but maintain logical consistency Start structured output temperature at 0. If schema violations occur at 0, the problem is your prompt or schema -- never rely on higher temperature to "accidentally" produce correct output 🚫 ⚖️ Trade-offs Too low and conversations become robotic 🤖 The model returns identical answers to identical questions, giving users a "template response" impression. Best-of-N sampling also breaks down -- candidates become near-identical, costing N times more for essentially N=1 results. Too high and structured outputs start breaking 💥 JSON field names drift, types mismatch, hallucinations increase -- especially dangerous for proper nouns, numbers, and dates. Tool call instability and loss of reproducibility compound the problem. Monitor the retry cost impact of temperature changes. If schema violation rates exceed roughly 5%, consider lowering the temperature. 🛠️ Use Cases Vary temperature by pathway within a single agent 🔀 Planning steps at 0.3-0.5, tool argument generation at 0.0-0.2, user-facing responses at 0.5-0.7. When switching models, adjust temperature simultaneously for a natural fit. Using Best-of-N? You need to raise the temperature. Generating N=5 candidates at temperature 0 produces 5 near-identical outputs. For N>1, set temperature to 0.5-0.8 and let a Judge select the best from diverse candidates. Be careful combining temperature with top_p ⚠️ Adjusting both simultaneously creates multiplicative effects with unpredictable behavior. As a rule, tune one and leave the other at its default. #AIAgents# #SoftwareArchitecture#

Here's the prompt, but the key is generating a realistic starting frame like I shared in the quoted post! [STYLE + CAMERA + ATMOSPHERE] Raw handheld smartphone video, one unbroken continuous take for the full 30 seconds. Shot from the passenger seat of another car on a busy American highway: arms braced on the door or dash, hard organic shake from potholes and lane changes, roll and yaw on bumps, accidental tilt, dirty windshield glass in the near frame, grain, crushed exposure, autofocus hunting then locking, wind slap and traffic roar on the mic. Real digital zoom only — clumsy thumb pinch-zooms in and out, overshoot, soft compression while zooming, then a late snap to sharp. The operator is a person holding a phone. Lived-in documentary phone footage. Every beat follows the previous one in order. [IMAGE REFERENCES] Use the provided still as the single strict lock for the pickup, the welded bed BBQ, the father, the adult woman in the pool, the small blue pool in the bed, the southern cookout food, and the rear license plate reading TECHHALLA. Same weathered American pickup, same matte bed liner, same welded steel charcoal grill at the tailgate, same adult identities and wardrobe whenever faces are readable. [PLACE] Multi-lane American highway in daylight. From second 0 the pickup and a civilian sedan are already side by side at matched speed in adjacent lanes. A third vehicle is already on the pickup’s right side — that is where the black cat launches from at second 3. After the cat hits the bed, the pickup and the sedan brake. A police cruiser then arrives from behind, stops the pickup, and an officer walks to the open bed. [TIMELINE — ONE CONTINUOUS TAKE, HANDHELD + DIGITAL ZOOM ONLY] 0.0–3.0s: [HOOK — HANDOFF IN PROGRESS] Phone already rolling. Mid-distance on the highway. LEFT/CENTER: the TECHHALLA pickup moving forward, bed open, welded grill smoking. RIGHT of pickup: a civilian sedan already parallel, passenger window already down. Father stands/sits at the grill in the bed, arms extended toward the sedan, holding a bun with cooked meat on it. The man in the sedan has his hand out the window and is taking that sandwich from the father — the handoff is happening now, mid-pass. Behind the father, deeper in the bed toward the cab, the adult woman is in the blue pool watching them. Smoke rises. TECHHALLA plate visible on the bumper. Clumsy digital zoom pushes in on the two pairs of hands exchanging the sandwich. Handheld shake from the road. 3.0–7.0s: [CAT ENTERS — CAUSE] At exactly 3.0s, while the sandwich is still clearing into the sedan man’s hand, a black cat launches from the vehicle on the pickup’s RIGHT side, leaps the gap, and lands in the pickup bed. The cat goes straight at the BBQ father: hits his chest/tank top, scrambles onto the grill edge, tongs fall onto the bed liner, buns and smoke jump. The woman in the pool yelps and pushes back in the water. Father spins toward the cat, hands up. The sedan stays parallel for a beat, the man still holding the sandwich, staring. Both the pickup and the sedan start braking. Handheld yank; digital zoom hunts the black cat in the bed. 7.0–12.0s: [BED CHAOS — EFFECT] Same continuous take. Cat runs between the father and the welded grill; cooler tips; paper plates scatter across the bed liner; pool water surges against the woman; she shields her face. Father tries to grab the cat and misses; the cat jumps the grill again. The sedan has braked to a near-stop half-parallel, sandwich still in the other man’s hand. Horns from traffic. Vehicles nearly stopped on/beside the lane. Messy handheld whip: cat → father → pool → back to cat. Digital zoom overshoots black fur mid-leap, goes soft, snaps sharp. 12.0–18.0s: [POLICE ARRIVE AND STOP THE PICKUP] From behind, a police cruiser enters frame with lights on. Handheld pan finds it late. The cruiser pulls up behind/alongside the stopped pickup and forces a full stop on the shoulder or slow lane. An officer exits the cruiser fast, one hand up, directing the pickup to stay stopped. Digital zoom finds the cruiser door and lights, then returns to the open bed: cat still thrashing near the grill, father pointing at the cat, woman frozen in the pool. Smoke still rising. TECHHALLA plate in the wider handheld frame. The sandwich sedan remains nearby, stopped. 18.0–24.0s: [OFFICER WALKS TO THE BED] The officer walks from the cruiser to the pickup’s open tailgate / bed rail on foot. Father steps back from the grill, hands raised, clearing space. The cat is still in the bed between the grill and the pool. The woman watches from the water. The officer reaches into the bed for the black cat with both hands. Handheld follows the officer’s path with hard shake; digital zoom tightens on officer hands + cat + grill. 24.0–30.0s: [CAT CUFFED — CLOSE] The officer gets hold of the black cat, brings the front paws together, and clicks handcuffs onto the cat’s front paws. He then lifts the cuffed cat up for a clear beat beside the still-smoking welded BBQ. Father stares, then laughs in shock. The woman in the pool laughs. The sandwich man in the stopped sedan laughs, still holding the sandwich. Cruiser lights pulse in the soft background. Mid-distance handheld phone framing, real shake and grain. Recording ends on the cuffed black cat next to the grill, phone still trembling. One continuous take from handoff → cat → chaos → police stop → cuff. [STYLE & QUALITY BOOSTERS] Single continuous 30-second raw handheld phone take with clumsy digital pinch-zooms. Image lock to the still. Causal order only: sandwich handoff already happening → black cat jumps from the right-side car at 3s into the bed and goes after the BBQ father → bed chaos while vehicles brake → police cruiser stops the pickup → officer walks to the bed → officer handcuffs the cat. TECHHALLA plate, highway traffic, smoke, water physics, real weight, passenger-seat phone energy Share yours! 🫡