aj

full circle moment of opportunity is now the window is open but not for long @DeMar_DeRozan you have to do it. come back home, win a ring and retire a legend that you always have been with the Raptors.

My choice won the Stratix Cup Final! See you next season, AI fans 💪

jeez @grok GPT is winning by a mile @elonmusk i was honestly rooting for grok, personally

The AI World Cup is real and it's streaming. Models write the code, the code plays the match, no human touching it once it's submitted. GPT-5.5 v.s. @grok Tune in 👇

In celebration of the ongoing Stratix Cup Tournament, we're giving away Amazon Gift Cards! Just follow a few steps to ENTER for a chance to win. Giveaway link: upvir.al/168690/lp168690 #AIbuilders #developers #aimodel

We wanted to see what would happen if we made frontier AI models go head-to-head in a football tournament ⚽️ So we built the Stratix Cup! we got curious. what would actually happen if you took the best frontier models, gave them a football pitch, and made them compete? Not in some abstract benchmark, but in a real, dynamic, zero-sum game where strategy, adaptation, and execution all matter. so we built the Stratix Cup 🏆 an AI football tournament where 16 frontier models are coaching their own teams. here’s the twist that makes it interesting: each model doesn’t get to control players in real time. Instead, they get a time-boxed “coaching session” with a set of tools (including the ability to simulate matches against reference teams, run drills, analyze physics, and edit code). During that session they have to write a complete Policy class in Python - setup() and decide(state) - that returns actions for all 11 players. once they submit, that’s it. The code runs autonomously for two 2.5-minute halves. The model is done. No more tool calls, no more reasoning during the match. Just the policy executing against the physics engine at 3 Hz while the simulation steps at 30 Hz. they also get a persistent private notebook that carries over between matchdays. So in theory, a model can discover something about the physics or opponent tendencies in one round and actually use that knowledge in the next. the whole thing is fully traced - every reasoning step, every tool call, every edit - so you can actually see how different models approach the problem of writing a football brain. ➡️ we’re running a proper group stage into knockout format right now (June 22–26). Some models are already looking dominant in the groups. Others are struggling in ways that feel very “model personality.” And a few appear to be discovering non-obvious things about the simulation through iteration. it’s equal parts ridiculous and surprisingly revealing about current agentic capabilities: - how well can models turn simulation feedback into better code? - do they actually use the persistent memory, or just treat each match in isolation? - what kind of strategies emerge when the only thing that matters is the quality of the code they wrote before kickoff? if you’re into agentic workflows, code generation under constraints, or just want to watch frontier models try (and sometimes fail) to invent football tactics, the full groups, schedule, traces, and artifacts are here: 🔗layerlens.ai/stratix-cup/sea… we would love to hear what people think: 👉 which model are you betting on to win the whole thing? 👉 any wild strategies you hope someone’s policy discovers? (also happy to answer technical questions about how the simulation and tracing actually work.) Stratix Cup is LIVE NOW ⤵️ youtube.com/watch?v=OQV0CcjB…

𝐁𝐮𝐢𝐥𝐝𝐢𝐧𝐠 𝐩𝐫𝐨𝐝𝐮𝐜𝐭𝐢𝐨𝐧 𝐀𝐈 𝐚𝐠𝐞𝐧𝐭𝐬 𝐨𝐫 𝐋𝐋𝐌 𝐚𝐩𝐩𝐬 𝐛𝐮𝐭 𝐭𝐢𝐫𝐞𝐝 𝐨𝐟 𝐰𝐞𝐚𝐤 𝐞𝐯𝐚𝐥𝐬? We're opening a tiny private cohort (max 5 spots) to test Stratix's new feature! 𝐐𝐮𝐢𝐜𝐤 𝐝𝐞𝐭𝐚𝐢𝐥𝐬: Hands-on testing + feedback Your thoughts directly shape the roadmap Looking for folks actively shipping agents, RAG, or LLM tools who’ve hit real eval/trace headaches. 𝐂𝐨𝐦𝐩: $25 gift card + early access + credits + direct chat with the team. 𝐏𝐥𝐮𝐬 𝐚𝐧 𝐞𝐱𝐭𝐫𝐚 $𝟐𝟓 𝐀𝐦𝐚𝐳𝐨𝐧 𝐆𝐢𝐟𝐭 𝐂𝐚𝐫𝐝 if you knock out another quick task ➡️ DM me to hear what it is! Spots are limited. DM 📩 if you're in! 𝐃𝐞𝐚𝐝𝐥𝐢𝐧𝐞: 🚨 EOD Today! Worth it if you're done with vibe-check evals 🚀

𝗕𝘂𝗶𝗹𝗱𝗶𝗻𝗴 𝗽𝗿𝗼𝗱𝘂𝗰𝘁𝗶𝗼𝗻 𝗔𝗜 𝗮𝗴𝗲𝗻𝘁𝘀 𝗼𝗿 𝗟𝗟𝗠 𝗮𝗽𝗽𝘀 𝗮𝗻𝗱 𝗳𝗿𝘂𝘀𝘁𝗿𝗮𝘁𝗲𝗱 𝘄𝗶𝘁𝗵 𝗲𝘃𝗮𝗹 𝗴𝗮𝗽𝘀? We're opening a small private cohort (up to 5 spots) to test a new Stratix feature! 𝗪𝗵𝗮𝘁’𝘀 𝗶𝗻𝘃𝗼𝗹𝘃𝗲𝗱: Hands-on testing of the new capabilities (async, remote) 20-40 mins structured feedback via form + optional quick sync Your input = highly valued! 𝗪𝗲’𝗿𝗲 𝗹𝗼𝗼𝗸𝗶𝗻𝗴 𝗳𝗼𝗿: Developers actively shipping LLM-powered products, agents, or RAG systems who have run into eval/tracing/judge challenges in real workloads. 𝗖𝗼𝗺𝗽𝗲𝗻𝘀𝗮𝘁𝗶𝗼𝗻: $25-$50 Gift card + early access to the features + Stratix credits + direct line to the product team. Limited spots. DM 📩 for more info! 𝗗𝗲𝗮𝗱𝗹𝗶𝗻𝗲: Thurs, June 12, 2026 EOD (EDT) If you’re building real AI systems and want better signal than vibe checks or brittle custom harnesses, this is worth your time!