Alibaba's o1 reasoning rival

PLUS: AI2 launches fully open Llama competitor

Welcome, AI enthusiasts.

Chinese tech giant Alibaba just entered the reasoning race in a big way — with a new open o1 rival that matches the industry leader’s capabilities.

Open-source AI is officially competing with Silicon Valley’s finest, and OpenAI’s model moat is looking thinner by the day. Let’s get into it…

In today’s AI rundown:

  • Alibaba challenges o1 with open-source reasoning model

  • AI2 launches fully open Llama competitor

  • Create live web prototypes with Qwen Artifacts

  • AI outperforms experts at predicting scientific results

  • 5 new AI tools & 5 new AI jobs

  • More AI & tech news

Read time: 4 minutes

Important: The Rundown is slowly being sent from a new email address. To ensure you get our newsletter, please add [email protected] to your contact list.

LATEST DEVELOPMENTS

ALIBABA

Image source: Alibaba

The Rundown: Alibaba's Qwen team just released QwQ-32B-Preview, a powerful new open-source AI reasoning model that can reason step-by-step through challenging problems and directly competes with OpenAI's o1 series across benchmarks.

The details:

  • QwQ features a 32K context window, outperforming o1-mini and competing with o1-preview on key math and reasoning benchmarks.

  • The model was tested across several of the most challenging math and programming benchmarks, showing major advances in deep reasoning.

  • QwQ demonstrates ‘deep introspection,’ talking through problems step-by-step and questioning and examining its own answers to reason to a solution.

  • The Qwen team noted several issues in the Preview model, including getting stuck in reasoning loops, struggling with common sense, and language mixing.

Why it matters: Between QwQ and DeepSeek, open-source reasoning models are here — and Chinese firms are absolutely cooking with new models that nearly match the current top closed leaders. Has OpenAI’s moat dried up, or does the AI leader have something special up its sleeve before the end of the year?

TOGETHER WITH EIGHT SLEEP

The Rundown: Eight Sleep's Pod 4 Ultra is redefining sleep by combining AI, biometrics, and personalized climate control for the ultimate night’s rest —  bringing lab-grade sleep optimization to your bedroom.

The Pod 4 Ultra offers:

  • AI-driven temperature adjustments throughout the night

  • Detailed sleep analytics and daily sleep fitness scores

  • Advanced snore detection with automatic bed adjustments

Use code RUNDOWN at eightsleep.com/rundown for up to $600 off bundled purchases through December 14th.

AI2

Image source: AI2

The Rundown: Research institute AI2 just released OLMo 2, a new family of fully open-source language models that matches the performance of similar-sized competitors like Meta’s Llama.

The details:

  • The 7B and 13B models were trained on a 5T token dataset of high-quality academic content, filtered web data, and specialized instruction sources.

  • The OLMo models achieved similar or better results while using less computing power than competitors and being smaller in size.

  • The models are fully open, with AI2 providing access to source code, training data, and a dev package with training recipes and evaluation frameworks.

  • The release also includes instruction-tuned variants, which achieve competitive results against leading open models like Qwen 2.5.

Why it matters: While other open-source models release weights but remain heavily guarded, OLMo 2 proves that cutting-edge AI can be developed and released completely in the open — potentially setting a powerful new standard for how future systems are built and shared.

AI TRAINING

The Rundown: Qwen2.5-Coder’s new Artifact feature instantly transforms your web ideas into live, interactive prototypes.

Step-by-step:

  1. Visit Hugging Face and locate the Qwen2.5-Coder-Artifacts space.

  2. Enter your prototype description with specific design requirements.

  3. Click "Send" to generate and preview your prototype instantly.

  4. Refine the design and export the code for your project.

Pro tip: Start with basic layouts and gradually add features to build complex prototypes efficiently.

AI RESEARCH

Image source: Ideogram

The Rundown: A new study from the University College of London just revealed that AI systems can predict scientific outcomes significantly better than expert neuroscientists — also uncovering ‘hidden’ patterns in research that could help better guide future studies.

The details:

  • A ‘BrainBench’ tool was used to test 15 AI models and 171 neuroscience experts’ ability to distinguish real vs. fake outcomes in research abstracts.

  • The AI models achieved 81% accuracy, compared to 63% for the experts — with a ‘BrainGPT’ trained on neuroscience papers scoring even higher at 86%.

  • The success suggests scientific research follows more discoverable patterns than previously thought, which AI can leverage to guide future experiments.

  • The researchers are developing tools to help scientists validate experimental designs before conducting studies, potentially saving time and resources.

Why it matters: While AI's pattern recognition capabilities aren't surprising, its ability to predict scientific outcomes could completely change how research is conducted. Using AI to validate experiments before spending any time in the lab could lead to faster research cycles, fewer dead ends, and accelerated scientific breakthroughs.

NEW TOOLS & JOBS

  • 🎥 Magic Roll - Create viral shorts in one click with B-roll, motion graphics, and AI-powered captions

  • 🤝 OfferGenie - AI-powered career copilot with real-time guidance to ace every interview

  • 📸 Runway Frames - A new foundation model for image generation with style precision and visual world-building.

  • ⚙️ Foundry - Build, evaluate, and improve AI agents that can automate key parts of your business

  • 💬 Llms.txt Generator - Generate an llms.txt file for your website to provide information to help LLMs use your website at inference time

  • 🚀 The Rundown - Head of Growth

  • 🛠️ Shield AI - Manufacturing Engineer

  • 💼 Cresta - Sales Development Representative, New York

  • 🧠 Writer - Director, AI Research

  • 🔬 Deepmind - Research Engineer, Materials Science

QUICK HITS

OpenAI temporarily suspended access to Sora for beta testers following Tuesday’s leak, with a group of artists creating an unauthorized public interface to the AI video tool.

xAI reportedly plans to release a standalone app to compete with OpenAI’s ChatGPT as early as December, marking the company’s first product outside of the X platform.

H Company showcased new demos of its Runner H agent, performing advanced web tasks, including real-time data extraction, complex interface navigation, and precision web scraping across multiple platforms.

ElevenLabs introduced GenFM podcasts, a new feature that allows users to generate AI-hosted conversations in 32 languages about uploaded PDFs, articles, eBooks, and more.

Elon Musk posted on X that he plans to start an AI game studio with xAI, saying he wants to “make games great again.”

Chinese self-driving startup Pony AI raised $260M at a $4.5B valuation as the autonomous taxi company’s U.S. IPO goes live for trading this week.

THAT’S A WRAP

That's it for today!

Before you go we’d love to know what you thought of today's newsletter to help us improve The Rundown experience for you.

Login or Subscribe to participate in polls.

See you soon,

Rowan, Joey, Zach, and Alvaro—aka The Rundown Team

Reply

or to participate.