Yoshi

This smart Polymarket trader has a PnL of $330k with $225m in volume! 1) Here's his strategy: - Buy moonshot bets like 'Will Kanye win the elections'. Essentially buying lottery tickets and hoping one hits - You may be asking, but isnt this just gambling? Well yes, on an individual position basis its a coin toss. But in aggregate, he manages his position sizes very efficiently and is able to ensure diversification of bets - This leads to almost VC-esque returns where most bets are expected to go to 0 but the few that hit do a 100x - Very risky strategy IMO but has worked for this trader 2) Here's how you can copy trade him in 20 seconds using @tradefoxai: - Paste his wallet address in the search bar - Follow him and then click copy all trades - Input variables based on your risk tolerance and click 'Copy' TradeFox will now automatically copy every single trade made by Agriculture Secretary within 3 seconds (Link in comments)

Chat no longer dominates AI spend Over the past 30 days, 56.3% of OpenRouter spend went to agents and coding, compared with 33.9% for general-purpose AI. Meanwhile, the market is splitting into modality-specific winners: - Google generated 74% of images and 54% of video hours - OpenAI returned 90% of transcribed characters - ByteDance captured 27% of video hours - Claude leads code generation, debugging, file I/O and frontend/UI Lmk what other data would be interesting to add. I'll be posting these weekly. Data: @OpenRouter

The most trenchant take on the open-source issue so far

Looks like the dixmier conjecture and poisson conjecture are proven false too. Things fall apart; the centre cannot hold.

Has anyone built an OpenRouter or DeFiLlama equivalent for robotics, autonomous systems, and defense? Real-time dashboard tracking funding, contracts, deployments, technical benchmarks, etc.

Most professional “investors” are still blissfully unaware and unprepared for inference to become the largest market in the world. Revenues will be larger than oil and gas, banking, automobiles, etc. Inference will account for multiple percentage points of GDP.

Overheard a 95 year old lady talk about grace hopper racks at breakfast today. I sat there stupefied for a good 15 minutes only to realize she was talking about a grasshopper in trapped in her garden cracks. Going to log off for a bit.

consensus: ai lets us ship software faster tier 2: ai will enable quants to outcompete hardware and eecs types tier 3: prototyping was never the hard part. tolerancing, regs, supply chain, reliability, certification still are tier 4: those are mostly knowledge retrieval combined with checklists and simulation. ai eats those too. endgame: software was the warmup. ai is coming for atoms. every specialist layer between an idea and a shipped object is collapsing, and agency is the last scarce input. the meek and high agency shall inherit the earth

Model Usage Weekly Update Text Tokens Processed 1. @TencentHunyuan Hy3 - 6.13T 2. @XiaomiMiMo MiMo-V2.5 - 5.95T 3. @deepseek_ai V4 Flash - 5.22T 4. @MiniMax_AI M3 - 4.26T 5. @Zai_org GLM 5.2 - 3.19T Total: 52.6T WoW: +13% Image Market Share 1. @GoogleDeepMind 81.0% - 4.32M 2. @OpenAI 7.5% - 399K 3. @SpaceXAI 4.8% - 256K 4. @bfl_ai 3.1% - 163K 5. @BytePlusGlobal 2.5% - 131K Total: 5.33M WoW: -3% Video Request Volume 1. @GoogleDeepMind Veo 3.1 Fast - 42K 2. @BytePlusGlobal Seedance 2.0 - 21K 3. @GoogleDeepMind Veo 3.1 - 18K 4. @GoogleDeepMind Veo 3.1 Lite - 18K 5. @SpaceXAI Grok Imagine Video - 8K Total: 129K WoW: 0% Transcription Request Volume 1. @OpenAI Whisper Large V3 Turbo - 1.02M 2. @OpenAI Whisper Large V3 - 960K 3. @OpenAI GPT-4o Mini Transcribe - 493K 4. @NVIDIAAI Parakeet TDT 0.6B v3 - 350K 5. @Alibaba_Qwen Qwen3 ASR Flash - 257K Total: 3.51M WoW: -51% Notable: - Hy3 debuted at #1 for text usage. - @GoogleDeepMind held the top five image-model spots and three of the top four video-model spots. - @OpenAI held the top three transcription spots. GPT Image 2 was also the fastest-growing top-10 image model at +133%. - Claude Opus 4.7 and 4.8 from @AnthropicAI ranked #7 and #9 for text usage. h/t @OpenRouter for the dashboards. Lmk what other data would be interesting to add.

The neolab trade maps to alt L1s: high-beta proxies on the dominant narrative, rerating faster than the underlying because supply of investable exposure is constrained. Bare metal maps to BTC miners circa 2019-21: operating leverage on a commodity input, providing convexity on the way up. The key is knowing which inning we're in.

Erling Haaland’s perspective on feeling tired:

open source models feel like lewis carroll's red queen. no matter how fast they run, they’re still running in place. moonshot's kimi 2.6 is a fantastic model. it would be the best in the world... if only it had shipped in january.

Anyone have ideas on how to expedite iOS apps getting approved? It says 90% of apps are reviewed within 48 hours and the average review times are much lower on Runway But we're still stuck in "waiting for review"

The AI Infrastructure Stack

i've given codex full access to all my documents and folders

in case you missed it $MSTR is down over 4x since this post you're welcome

the labor market is not ready for what codex is becoming today i gave codex an api key and it: > pulled 100k rows > figured out the endpoints > cleaned the data > answered questions that would’ve taken a $250k data scientist 2-3 business days > sanity checked it it did this in 2-3 minutes btw these were not simple sql queries messy questions where it was unclear what formula to use, what data was missing, and how to iterate based on prior experience working directly with me could see the unhobbling, CoT, and mechanistic interpretability all working in tandem absolutely fantastic 2026 is incredible

nominative determinism can be applied to ai models too.

everyone is salivating over spacex's s1 while 99% would benefit more from reading bending spoon's f-1 their path is more reproducible and less circumstantial

A+ content, interactivity, and information density design