Quick lookback on technical AI themes of 2026: (dataset: daily ainews title stories) Jan: - Ralph Loops - RLMs - Filesystems are all you need - X LLM recsys open sourced - X Grok Imagine #1# Video model - Anthropic blocked OpenCode / OpenClaw - MCP Apps - Moltbook - Apple picks Gemini for new Siri - Notable models: Kimi K2.5 - Fast inference: OAI x Cerebras partnership - Big raises: xAI @ $230B + SpaceX, OpenEvidence @ $12B Feb: - Codex App launch - Claude Code 1 year anniversary - Context Graphs - Fast Inference: Taalas 17k tok/s/user Llama 3 8B - Anthropic vs Deepseek/Moonshot/Minimax distillation accusation - Notable models: Opus/Sonnet 4.6, 5.3 Codex and Codex Spark, Qwen-Image 2.0, Seedance 2.0, GLM-5, Gemini 3.1 Pro and Deep Think, Nano Banana 2 - Big raises: OpenAI @ $840B, Anthropic @ $380B, ElevenLabs @ $11B, Cerebras @ $23B Mar: - T3 Code - Autoresearch - Claude Code source leak - Claude Cowork Dispatch - OpenAI buys Astral - Cloudflare-Next.js "Slop Fork" - Meta buys Dreamer - NVIDIA GTC: OpenClaw, Vera CPU, 1T sales backlog, "The Inference Inflection" - ARC-AGI 3 introduced, SOTA models are 1% - Notable models: GPT 5.4 unifies chat and codex, Gemini 3.1 Flash-Lite/Live, Nemotron 3 Super, Composer 2, MiniMax 2.7, Arcee’s Trinity-Large-Thinking - Big Raises: Yann LeCun AMI Labs $1B seed Apr: - Claude Design - OpenAI Symphony & Harness Engineering discourse - @LilianWeng writeup later - OpenAI and Microsoft break up - Tokenmaxxing fears - Codex Superapp - AIE Europe, AIEi Miami - Hermes Agent gains traction w/ Manim skill - Cyber: Anthropic Project Glasswing - Notable models: Claude Mythos "Capybara", Opus 4.7, Muse Spark, Gemma 4, Kimi K2.6, GPT-Image-2, GPT 5.5 "Spud", DeepSeek v4 Pro/Flash Base/Instruct on Huawei Ascend, Poolside Laguna XS.2, Grok 4.3 - Big Raises: not much but Anthropic $19B->$30B ARR in one month... May: - Cyber: OpenAI Daybreak, Tanstack Mini-Shai Hulud - FDE: OpenAI DeployCo, Ode by Anthropic - OpenAI solve Erdos planar unit distance problem - OpenAI deprecates finetuning - "How to land a job at a frontier lab" - Anthropic rents Colossus I from SpaceX for $5B/yr - Thinking Machines Interaction Models (still underrated!!!) - Notable models: GPT-Realtime-2/Translate/Whisper, Gemini 3.5 Flash, Omni, Antigravity 2.0, Opus 4.8, Cosmos 3, Minimax M3 - Big Raises: Fireworks @ $15B, Baseten @ $11B, Cognition @ $26B, Cerebras $60B IPO, Anthropic @ $965B Jun: - Claude Tag - OpenAI Jalapeno - "The Untrainable" - @saranormous - Loopcraft: The Art of Stacking Loops - Metaharnesses discourse? Flue v Eve - Zai Prof @jietang predicts Open Fable by Dec - "SpaceX is a $28B/yr neocloud" - @jaminball - Anthropic-Fable White House issue (resolved) - Notable models: GPT 5.6 Sol/Terra/Luna, Fable 5, Sonnet 5, MAI-Thinking-1 and MAI Family, Nemotron 3 Ultra, GLM 5.2, Reve 2/ Ideogram 4 - Big Raises: SpaceXai buys Cursor for $60B, Supabase @ $10B Jul: - AIE World's Fair 3 - OpenCode 2.0 - Cyber: OpenAI-HuggingFace incident - Chinese Model ban discourse - GPT 5.6 Luna 80% price cut - Notable models: Grok 4.5, Opus 5, Muse Image, Thinky Inkling, Kimi K3, BFL FLUX 3, DeepSeek-v4-Flash-0731 - Big Raises: Aug: - EU AI text watermarking - Microduck - Megakernels - Memory prices up 500% in 12 months - How to steal a Reasoning Trace - Grok @bot! - OpenAI shuts off Cursor - Fast Inference: AMD buys Taalas - @andrewng stars AI Engineering series - Hot Chips: Jalapeno, Cerebrase CS-5, Groq 3, Apple M6 - Demis steps down from CEO, & Dean et al leave - Notable models: Qwen 3.8 Max, Muse Spark 1.2, Muse Glimmer, Grok 4.6, Gemini 3.7 Flash, GLM 5.3, @Fal H3 Max Live - Big Raises: Stripe buys OpenRouter $7B, Poolside $12B reverse-execuhire to NVIDIA, NVIDIA buys HuggingFace for $13B Sep: - AIEi Paris - OpenAI Navier-Stokes singularity - collusion . wiki - Yegge shuts down Gas Town - AEF-1 standard for Third Party Evaluators - Meta Muse App launch. Also new glasses and Charm - OpenAI blocks Cursor - OpenAI Dots - Opus 5.5 explainer videos - Pi 1.0 & Pi Durable - Notable Models: Fable 5.1, Mythos 5.1, Opus 5.5, GPT-6 Astra, GPT Image 2.5, DeepSeek v4.1-Flash, @typesafeai Jev, MiMo-v2.6-Pro new Open SOTA, GPT 6.1 Sol & Ultrafast & Decisions API, Gemini 4 Argon - Big Raises: Cognition @ $48B, Mistral @ $24B, AMD buys World Labs for $8.2B Oct to date: - AIE NYC - @dhh stops using Rails - ArtCraft clones Adobe suite - Interpretability vs Polysemanticity - OpenAI Quasi-Riemann-Hypothesis & 722 Math - Notable Models: Reflection Beam, Claude Haiku 5.5, Muse Wonder & Wonder Full reach Opus 5.5 tier - Big Raises: TypeSafe $7.5B 3 weeks after launch, Modal @ $16B (this was done for research purposes for the keynote in Nov, just sharing notes in case helpfuil)

Annual Compound Growth Rate of Bitcoin's 200 week Moving Average: Lookback Annualized growth 1 year 24.1% 3 years 32.8% 4 years 29.3% 5 years 33.6% 7 years 46.8% 10 years 68.7%

Friday Update - Week of August 10th, 2026: - Rate limits: volume-based tiers now live in production, already curbing spam traffic. ( - TWAP: 5m crypto markets moved from a 30s to a 60s Chainlink TWAP lookback. Live on 5m markets deployed since Aug 14 - Passkeys: passkey for login live for Magic wallet users (email and social logins) - Perps: ADL fills flagged in trade history with both sides notified, new mark-price history endpoint, mobile web trading and margin control improvements, share cards to be released early next week. - Combos: odds visible while logged out, finished games hidden, geopolitics combo markets in preparation for next week - Data API: new /v1/approvals endpoint, faster native serving for positions and PnL, fixes for resolved markets showing open and a PnL price bug; price-history now on its new serving layer after last week's slowdown - Gamma: clearer 503s, graceful search degradation, keyset pagination on comments - Bridge API: pagination live, docs updated

⚠️ Breakdown: Papertrade BBO oracle manipulation — ~$1.28M taken 8 hours after launch On Oct 10, 2026, HyperEVM perp venue @papertrade_xyz went live, with $85.3M in pre-deposits and TVL reaching $118M within 24 hours. About 8 hours after launch it was hit by an oracle manipulation attack. Loss: ~$1.28M. This was a cross-market oracle and quote-manipulation incident. The suspected attacker first opened oversized synthetic longs and shorts on Papertrade, then used real capital to buy and sell ETH on Hyperliquid, briefly displacing Hyperliquid’s BBO mid. That displaced mid was then used for execution and PnL on Papertrade. The resulting PnL was paid by the Martingaler LP. 🔍 Root cause 1. Instant BBO from one market was the oracle Opens and closes both read the Hyperliquid best-bid/ask mid directly. The official docs explicitly state there is no max-deviation cap, no TWAP, no lookback, and no independent reference price. A non-zero, non-crossed BBO could therefore be accepted even if it had been manipulated. Papertrade’s own risk docs identified “Hyperliquid BBO manipulation” as a major protocol-level risk before launch: 2. 1,000x leverage turned a tiny move into a large payout On hundreds of millions in notional, pushing ETH ~10–14 bps was enough to generate hundreds of thousands of dollars in Papertrade PnL. Trading loss on the external venue was far smaller than the synthetic PnL paid by the LP. 3. The per-position cap was bypassed by splitting The per-position notional cap was ~$10M. The final $297M short was split into 30 positions of ~$9.9M each, just under the limit. Directional OI controls still left ~$1B of headroom, so existing controls did not meaningfully constrain aggregate exposure in this case. 4. LP capital did not match the exposure The Martingaler LP is the sole counterparty, and its pool is largely built from trader losses. Public data showed ~$3.14M in LP against hundreds of millions in notional, up to 1,000x. Available LP capital was severely mismatched with the potential payout. 🔗 Attack flow 1. Displace the upstream BBO The suspected manipulator executed concentrated buying and selling on Hyperliquid, moving the ETH perp’s best bid and ask. Papertrade used that mid as its execution price, so the distorted BBO mid fed directly into Papertrade’s pricing. This activity occurred on HyperCore’s order book and does not map to a single HyperEVM transaction hash. 2. Open the first high-leverage positions Once the upstream BBO had shifted, the suspected manipulator submitted signed orders to the Papertrade relayer. The relayer then called the Exchange contract to open the positions. The relayer appearing in the `from` field should not, by itself, be treated as the attacker. 3. Split positions to bypass the per-position cap The per-position notional cap was ~$10M. The suspected attacker split the exposure into ~30 positions of ~$9.9M each, for total exposure far above the single-position limit. Batch that appears to include ~30 shorts: 4. Trade the other way and close in profit After the Papertrade positions were in place, the suspected manipulator traded in the opposite direction on Hyperliquid so the BBO moved in favor of those positions. With no TWAP, no max deviation, and no lookback, the distorted mid was used directly to close the positions and calculate PnL. 5. Position termination path About 30 related positions were subsequently terminated. This transaction appears to correspond to the batch termination or suspected liquidation path: 📌 Addresses Suspected attack / manipulation Funds receiver / suspected coordinating address Final suspected liquidation caller Papertrade trading / settlement contract 🛡️ Security recommendations Simply blocking the suspected attacker address does not address the root cause. 1. Stop using an instant BBO as the only settlement price. 2. Use Hyperliquid’s multi-venue weighted oracle or mark price as a reference, plus TWAP, deviation limits, and freshness checks. Official oracle docs: 3. Bind total OI, per-address OI, and max payable PnL to LP capital that is actually available. 4. Trigger circuit breakers on rapid long/short reversals by the same address, large trades on external markets, and anomalously high win rates.

The most expensive thing in systematic trading is often not a losing strategy. It's the two years you spent not being in the market. Everyone underestimates this because time doesn't show up anywhere in your P&L. A blown account is visible and painful and you learn from it fast. Three years spent building a custom machine learning backtest engine, a feature pipeline and an ML framework before you've placed a single live trade costs you *far* more, and the damage is completely invisible. And the cruel part is that the building genuinely feels like progress. You're learning, you're solving real problems, you're getting better at something. But you're getting feedback about your code, not about the market. You only get it by being on the field. Meanwhile the thing you were going to build usually already exists in a simpler form for almost nothing. If you want trend, start with a basic 12 month lookback and a monthly rebalance. That's your first sleeve done in an afternoon, live, compounding, teaching you things. Then every afternoon after that goes into the second sleeve, and the third. Five boring strategies stacked together will beat the one brilliant one you're still building, because their drawdowns land in different places. So before you build anything, ask what you can do *right now*, with the tools you already have, that is likely to make money. Do that this week. Complexity comes later, if the simple version genuinely can't harness the edge.

CTAs are apparently massively short bonds: CTAs tripled their underweight/short positions in global bonds a record short per UBS ~$300 million P&L impact for CTAs per 1 bp move in the 10-year Treasury yield You can see similar positioning in the TLT fixed strike vols. This is set to stats-mode, with a 90-day lookback. TLT ATM vol is particularly high (green), with downside strikes at their highs. Conversely, it reads like the call side is being sold hard (red) - that could go bid if rates come down. We'd think a bond short unwind would in turn jack up equities.

TWO NUMBERS CAN LOOK AT THE SAME STOCK ON THE SAME DAY AND TELL YOU COMPLETELY OPPOSITE THINGS!!! One says the premium is average. The other says it's the highest it's been in a year. Both are correct. Here's which one to actually trust: They're called IV Rank and IV Percentile, and most people don't know there's a difference. Quick bit of context first. Implied volatility on its own tells you nothing. 40% IV is enormous on Coca-Cola and completely normal on a biotech. You need to know if it's high for that specific stock. IV Rank puts today's reading between the lowest and highest of the last year. Current minus the low, divided by the range. If IV ran 20 to 60 and it's 40 today, that's a rank of 50. IV Percentile is the share of days in the last year where IV closed lower than today. Now the example that shows why this matters. Say IV sat between 20 and 25 for basically the whole year. Then one day it spiked to 60 on some news. Today it's at 40. Rank says 50. Looks average, nothing special. Percentile says 98. IV has been below today's level on 98% of trading days. Rank got destroyed by one outlier because it only looks at two extremes. Percentile counts every single day, so it tells you what's actually normal for that name. How I use them. Percentile decides whether I sell. Above 50 you're getting paid more than usual. Above 70 is genuinely good. Under 30 and you're taking real risk for almost nothing. Rank is still handy for spotting a stock that just moved. A sudden jump means something happened recently. Two things worth knowing. Both default to a one year lookback. After a calm year the baseline resets underneath you and high IV starts reading as normal. And high IV is high for a reason. Earnings, litigation, a merger. A great percentile reading isn't a green light, it's a prompt to go find out why.