Take two: Mac Studio m4 Max recipe Qwen3.8-Flash-Next update Average 83tok/s • MTP off→MTP3 weighted decode: 36.64→68.31 tok/s (1.86×) • MTP3→MTP6 5,999-token decode: 70.44→83.06 tok/s (+17.9%) • TrueScore: 91.8→91.9 •256k context Deployment:

$ANET (September 8, 2026-Monthly Chart) The monthly chart shows four volatility holes. Each time price cleared the upper boundary of a hole, a new rally followed: First Volatility Hole: Cleared $11.8 → next leg up Second Volatility Hole: Cleared $42 → next leg up Third Volatility Hole: Cleared $162 → next leg up Will $ANET close a monthly candle above the upper boundary of the current volatility hole and start another expansion wave—or will it stay inside the box and digest the move from $162?

SHOWCASE: 19 Stocks: 8 Bets v 15 Bets ... WHY PORTFOLIO CONSTRUCTION MATTERS this is Altimeter Capital's latest 13F. Brad is one of the best tech investors of all time so we will use him as an example... every quarter people copy this filing line by line (this is for educational purposes to show LIQN BASKETS + SKIM + X-RAY) VOL is the price of admission in concentrated TMT investing, and ALTIMETER is world class at it Before the video: 13Fs have real problems for analysis 1: stale. This is the book on June 30th, filed August 14th 2: incomplete. Long positions only. No shorts, no cash, nothing private 3: newly public names like $CBRS and $SPCX have too little trading history to risk model. They're 19% of this filing, so the stats below cover the other 17 names PORT 1: AS FILED Vol 30.5% Beta 2.02 Max drawdown (1y) -19.0% THIS port in a 2022 like environment = -50% drawdown Money vs risk (share of the measured book): $NVDA 23.5% of $ → 23.5% of risk $CRWV 7.7% → 16.7% $ARM 7.3% → 12.8% $META 9.6% → 4.8% **3 names carry 53% of the risk PORT 2: INVERSE VOL *Same 19 stocks, only the sizing changed Effective bets 8.3 → 15.2 $NVDA risk share 23.5% → 4.7% No name above 10% of risk Max drawdown -19.0% → -16.3% 2022 replay -48.8% → -41.0% Beta 2.02 → 1.92. Same market bet CONCLUSION: Same 19 stocks. Not one pick changed. All we did was resize them, and the portfolio went from 8 real bets to 15, with half the concentration at the top. In the backcast it didn't cost return, and it kept the same market exposure. Picking the stocks is what everyone talks about AND IS IMPORTANT. How you size them is how you sleep at night. *educational only, not investment advice. This is not Altimeter's actual performance: it's built from the public filing plus a hypothetical re-weighting, replayed at today's weights which has inherent leakage

This is a crazy, weird Qwen3.8-27B “uncensored” release. It isn't because it's uncensored, but because the author measured what the guardrail surgery cost to get it uncensored and then preserved MTP. Jonathan Coletti took Qwen3.8-27B and used Heretic model surgery to reduce its refusal behavior. Measured on ~100 held-out test prompts ... 🛑 Base Qwen3.8-27B → w/98 refusals 🔓 Modified model → w/only 12 refusals Meanwhile, the model's capability barely changed ... 🧠 MMLU 83.4 → 83.3 🎯 HellaSwag 82.8 → 82.9 📊 Mean across 4 tests → ~0.5 point drop Abliteration initially dropped Qwen's MTP head and the author then grafted all 15 MTP tensors back from the original Qwen3.8-27B and then verified that the MTP block actually survived GGUF quantization. So the modified model you still get ... 🧠 27B dense Qwen 👁️ multimodal/vision 📚 262K context architecture 🔮 native MTP speculative decoding 🦙 llama.cpp GGUF And mradermacher already has iMatrix quantizations 📦 IQ3_S → 12.7GB 📦 IQ3_M → 12.9GB 📦 Q3_K_M → 13.6GB 👀 📦 IQ4_XS → 15.4GB 📦 Q4_K_M → 16.9GB That makes ~13GB versions especially interesting for 16GB GPU testing. 🔥 ⚠️ File size ≠ VRAM requirement, so don't assume the 15.4GB IQ4 automatically fits comfortably in 16GB with useful context. Also, I have found that “uncensored” model variants don't mean capability is proven unchanged. For example, their published comparison doesn't yet cover coding, math, multilingual, or vision quality. 😁

🚨 BREAKING 🇺🇸 THE FED WILL OFFICIALLY ANNOUNCE INFLATION DATA TODAY AT 8:30 AM ET! IF CPI > 4.0% → MARKET DUMPS HARD IF CPI = 3.8%-4.0% → MARKET STAYS FLAT IF CPI < 3.8% → MARKET GOES PARABOLIC ALL EYES ON THE RELEASE!!

claude opus velocity (version gain / 100 days) 4.1 █ 0.13 4.5 ████ 0.36 4.6 █ 0.14 4.7 █ 0.14 4.8 ██ 0.24 5 ████ 0.35 5.5 ████████ 0.83 4 → 4.8 took a year. 4.8 → 5.5 took 4 months.

and here’s the grok 4.7 model card. a few jumps vs 4.6 that stood out: - Terminal-Bench: 20.3% → 38.0% - SWE-Marathon: 31.9% → 46.0% - HealthBench Pro: 48.5% → 56.7% - Legal Agent: 15.8% → 19.6% - EEBench: 60.0% → 66.0% for the same price as 4.6!

TL;DR: Three failure modes of long-horizon agents — compounding errors, context rot, and task-state loss — solved structurally through a Manage-Execute-Audit (MEA) loop. WeaveBench PassRate: 51.8% → 80.7%. LongHorizon-Harness: Advancing Long-Horizon Agents for Real-World Tasks Key points: 🔧 Manager: Maintains explicit task state outside the execution trajectory. Constructs subtask contracts specifying goals, acceptance criteria, constraints, and evidence. ⚡ Executor: Runs each subtask in a fresh, budget-bounded context — no prior trajectory history passed in. 🔍 Auditor: Post-execution read-only inspection, independent of the executor. Reports completion, integrity, and state updates; serves as persistent cross-round memory. 🖥️ GUI/CLI hybrid: Manager routes tasks to the right interface; AgentAdapter swaps in Claude Code, Codex CLI, Hermes Agent without modifying native loops. 📊 WeaveBench: PassRate 51.8% → 80.7%; Design +60pp, Spatial/3D +50pp. 🤖 OSWorld 2.0: Qwen 3.7-Plus 2.8% → 8.3% (3×); Claude Opus 4.7 20.6% → 35.3%. 💻 Terminal-Bench: 69.7% → 77.2%, with 24% fewer tokens consumed. 💡 Manager overhead: only 2–8% of total tokens. Auditor carries 19–38% — the primary cost of reliability. "Agent capability is a property of the complete model–harness system, not the model alone" — that framing changes how you should think about long-horizon agent design. #AIAgents# #LLM#

Hyperliquid Whales Are Betting Big on ethereum:0x514910771af9ca656af840dff83e8264ecf986ca and hyperliquid:native Hyperliquid whale positioning shows a clear split across major assets. 🔹 LINK: 9.45 long/short ratio → Super Bullish 🔹 HYPE: 2.83 → Bullish 🔹 XRP: 2.08 → Bullish 🔹 BTC: 1.8 → Slightly Bullish But whales are taking the opposite side on several assets: 🔻 ETH: 0.72 🔻 SOL: 0.52 🔻 ZEC: 0.51 🔻 LIT: 0.46 → Bearish The standout is clearly ethereum:0x514910771af9ca656af840dff83e8264ecf986ca, where whale longs heavily outweigh shorts. Overall, whales aren't making a broad market bet. They're showing strong conviction in specific assets while staying defensive on others.

If you grew up sleeping with the lights on, this week's for you. → 2002 Neo Destiny Dark Gengar 1st Ed Holo PSA 8 → 2014 Phantom Forces Gengar EX Holo PSA 10 → 2021 Fusion Strike Gengar VMAX Full Art PSA 10 → 2021 Fusion Strike Gengar VMAX Alt Art Secret PSA 10 → 2020 Wobbuffet V Full Art PSA 10 Four Gengars across two decades. Ready to add them to your binder? Start trading →