by this time next year, most of labs revenue will come from what they build in-house and then sell further up the stack. the average users just don't 1. have deep enough pockets 2. work on hard enough tasks gpt-synopsys and anthropic wet lab are the first of many products which come as a result of labs gatekeeping and utilizing superintelligent swarms to create new and capture existing markets. currently by use of footholds and in the future independently. by 2030 labs will have eaten practically every economically valuable sector such as finance, legal and healthcare. what's to stop labs? why wouldn't they? how can a 10,000 employee healthcare company stand a chance if they are up against 10 million vastly superior agents to the swarms publicly available? the labs understand this and are hence preparing for the "grand capture" by donating shares to the public (anthropic and spacexai already committed with openai about to) to minimize p(public revolution). the question becomes what to do. one option is obviously to join a lab. if you want to join the lab I believe the best "unit chance of joining lab/time" ratio is proving scaleable user acquisitions and publishing relevant hot topic things on X. if you don't want to join a lab, acquiring factories, propietary data, licenses, land and things that are human facing and exist in a stack of value creation will be fruitful as they become the primary bottlenecks once suffcient swarm cognition is solved

72.2: mimo-v2.6-flash scores highest on SeesaBench for finance work, at $0.15 a run. It edges deepseek-v4.1-flash (71.3, $0.45). Meta's Muse Spark Open comes third (67.3, $2.93). bench.seesa.ai/?view=finance

Run DeepSeek V4.1 Flash 1017 on 2x DGX Sparks ✨ - 1M total KV pool, 512K context - 75 tok/s on prose, single stream - 130 tok/s on prose, 4 concurrent streams - Prefill 2000-2175 tok/s - EXL3-EAQv1, 2.9 bpw The quant in this recipe got about the same calibration as my GLM 5.3 Flash, with some extras. Feedback would be appreciated! Special thanks to @bertholomusai and all the contributors for the engine work, and to @ashxhart for TensorFold. Get it here: github.com/MiaAI-Lab/DeepSee…

Every Friday, we open the lab door. 🧪 Labs is where new Seesa features land first. Switch them on for yourself, and off again any time. This week: Pulse. Seesa opens on a picture of your week: your meetings above a line, the work Seesa did for you below it, and what it made along the bottom. Pick any stretch and Seesa tells you what happened in it: the meetings you had, what it drafted, and what's still ahead. Then open the documents or review the drafts in one click. No digging through inboxes to work out where the week went. Find it in Settings → Labs. seesa.ai/

New on SeesaBench: Meta's Muse Spark Open joined last night and took #1 from deepseek-v4.1-flash. 72.5 to 70.5, a tie within the 95% ranges. Muse costs $2.93 a run, DeepSeek $0.45. It leads the analyst and risk views, and is #3 for finance and ops. bench.seesa.ai

I thought $NVDA GPUs are banned in China … wife material.

Saw a guy call it cuck coding instead of vibe coding because you just sit in a chair and watch😭

JUST IN: Elon Musk unveils the Tesla Roadster by launching it off a ramp before the car flipped in midair and safely landed on the ground. The car is a collaboration between Tesla and SpaceX. One of the most insane things I've ever seen.

Which AI is best at office work? SeesaBench, our public leaderboard, tests 15 AI models on real business tasks. You can now view it by role, like finance or analyst, and the ranking changes. The best model overall is only 4th for analysts. bench.seesa.ai

Qwen 3.8 27b in #tensorfold running at excellent speeds on a single DGX Spark!

🚀 Introducing MiniCPM-V 4.7, our new open-source MLLM series! Advancing the Performance–Efficiency Pareto Frontier. It brings strong multimodal capabilities to iOS and Android devices with extreme efficiency. Two new models: 🔹 MiniCPM-V 4.7 35B-A3B — Bring SOTA performance to local devices with newly introduced LES (Lookahead Expert Scheduling) technique, a new milestone of on-device multimodal AI. 🔹 MiniCPM-V 4.7 1B — Lightweight, efficient multimodal AI. Easier to deploy. ⚡ What's new? SOTA performance: Competitive results across 34 benchmarks spanning multimodal understanding, STEM reasoning, OCR and long-video comprehension. Fast on-device inference: LES predicts and preloads expert weights to reduce inference stalls, boosting 5× speedup on iPad Pro M4 (16GB) with byte-for-byte identical outputs. It enables MiniCPM-V 4.7 35B-A3B to achieve 17+ tokens/s on iPad Pro M4 (16GB) and 13+ tokens/s on iPhone 17 Pro Max, while costing only 3GB memory 🤯 Efficient visual processing: 4× and 16× hybrid visual token compression. Even at 16× compression, 35B-A3B scores 90.4 on OCRBench, 5.2 points ahead of Qwen3.6 35B-A3B. Easier customization: We introduce Self-Replay Fine-Tuning, a simple way to customize new target capabilities at low cost while maintaining general performance. Broad hardware support: Day-0 adaptations for Apple, Intel, AMD, Arm and Rockchip, with many more platforms supported through FlagOS! Download, deploy, and build something we haven't imagined yet. 👀👇 🔗 GitHub: github.com/OpenBMB/MiniCPM-V 🤗 Hugging Face: huggingface.co/collections/o…