Microsoft’s recent AI with big math skills

Welcome, AI enthusiasts.

While OpenAI and Google place giant presents under the AI tree, Microsoft’s Phi AI model family just proved that good things can are available small packages too.

The brand new Phi-4 is outperforming the industry’s giants with only a fraction of the parameters — hinting at a shiny future for small, specialized models. Let’s get into it…

In today’s AI rundown:

• Microsoft releases small, powerful Phi-4

• ChatGPT gains ‘Projects’ for chat organization

• Practice job interviews with ChatGPT

• Pika drops major 2.0 video upgrade

• 5 recent AI tools & 5 recent AI jobs

• More AI & tech news

Read time: 4 minutes

LATEST DEVELOPMENTS

MICROSOFT

🦾 Microsoft releases small, powerful Phi-4

The Rundown: Microsoft just released Phi-4, a 14B parameter small language model that outperforms massive competitors like GPT-4o and Gemini Pro 1.5 in areas like mathematical reasoning despite a drastic size difference.

The small print:

• Phi-4 outperforms models like Gemini Pro 1.5 on several math and complicated reasoning benchmarks despite being a fraction of the dimensions.

• Phi-4 even surpasses its teacher model, GPT-4o, on graduate-level STEM Q&A and math competition problems.

• Microsoft trained Phi-4 totally on synthetic data, using AI to generate and validate roughly 400B tokens of high-quality training material.

• The model also features an upgraded mechanism that may process longer inputs of as much as 4,000 tokens, double the capability of Phi-3.

• Phi-4 is obtainable in a limited research preview on Azure AI Foundry, and a wider release is planned for Hugging Face.

Why it matters: Microsoft’s Phi models proceed to challenge the ‘larger is best’ trend in AI, showing that smaller models can match or exceed the capabilities of larger ones — particularly in specialized areas. The AI future might not be about raw size but smarter architecture and training approaches that do more with less.

TOGETHER WITH RABBIT

🐰 One orange gift that’s fit for all

The Rundown: Bring one of the best gift to your holiday festivities this yr with the rabbit r1 — an AI companion designed to grow more delightful over time because of continuous updates and recent features that keep it fresh and interesting.

Whether young, young at heart, or in-between, gift r1 to:

• Feed curiosity with unlimited answers to any questions at a moment’s notice

• Elevate travels with bi-directional translation in 100+ languages and personalized itineraries

• Nerd out with a totally customizable UI, the flexibility to create custom agents, and a playground of out-of-the-box AI features

Benefit from the one discount of the yr and secure r1s for everybody this holiday season.

OPENAI

🗂️ ChatGPT gains ‘Projects’ for chat organization

The Rundown: OpenAI launched Projects for ChatGPT on Day 7 of its ‘12 Days of OpenAI’ event, a brand new organizational system that lets users group conversations, files, and custom instructions into individual workspaces with shared context.

The small print:

• The feature introduces project-specific folders where users can bundle related chats, documents, and custom AI instructions across conversations.

• Each Project robotically leverages GPT-4o while maintaining access to core features like Canvas, DALL-E, and web search capabilities.

• The system is rolling out first to Plus, Pro, and Teams subscribers, with Enterprise and Education users gaining access in January.

• Projects may be created and managed through the net interface and Windows app, while mobile and Mac users can view and chat with existing Projects.

Why it matters: While this isn’t essentially the most groundbreaking feature (Anthropic released Projects for Claude in June), it is important for user workflows — avoiding the dreaded have to refresh entire context and instruction prompts when starting recent chats.

AI TRAINING

💼 Practice job interviews with ChatGPT

The Rundown: ChatGPT’s Advanced Voice Mode may be changed into a personalised interview coach, conducting mock interviews and providing real-time feedback.

Step-by-step:

1. Open ChatGPT’s Advanced Voice Mode in your mobile device.

2. Arrange your specific interview scenario and industry context.

3. Engage in a practical mock interview with industry-focused questions.

4. Get immediate feedback in your responses and presentation.

Pro tip: Should you need more time to formulate your responses, you possibly can customize how the AI responds in Custom Instructions.

PRESENTED BY INNOVATING WITH AI

🤝 Turn AI passion right into a consulting profession

The Rundown: Innovating with AI’s recent program, AI Consultancy Project, transforms AI enthusiasts into skilled consultants — tapping right into a market projected to succeed in $54.7B by 2032.

The 6-month program delivers:

• Proven frameworks for client acquisition and repair delivery

• A step-by-step path to six-figure consulting income

• Students who land their first AI client in as little as 3 days

Click here to request early access to The AI Consultancy Project.

PIKA LABS

🎬 Pika drops major 2.0 video upgrade

The Rundown: Pika Labs just released version 2.0 of its AI video generator, introducing a brand new ‘Ingredients’ tool that lets users incorporate their very own images into AI-generated videos — alongside improved motion, prompting, and animation features.

The small print:

• A brand new ‘Scene Ingredients’ system allows users to upload and blend characters, objects, and backgrounds that the AI robotically recognizes and animates.

• Pika’s updated model shows impressive realism, smooth movement, and prompt/image adherence, giving users more control over outputs.

• The brand new video generator also contains a significant update to text alignment, showcasing the flexibility to craft realistic branded scenes and promoting content.

• Pika has already attracted over 11M users and secured $80M in funding, and the new edition follows its viral ‘effects’ launch in October.

Why it matters: Pika’s recent upgrades are wild, continuing to maneuver video outputs out of the ‘slot machine’ luck phase right into a more customizable, personalized experience. While we patiently waited for Sora, the AI video scene leveled up in a significant way — with Pika, Luma, Runway, Kling, Hailuo, and others dulling the impact of OpenAI’s latest release.

NEW TOOLS & JOBS

Trending AI Tools

• 🔎 Reddit Answers – AI-powered search tool that enables you to find human perspectives, recommendations, and info powered by Reddit communities

• 😷 Doctronic AI – Quick, accurate care from home with an AI consultation followed by video visits with licensed doctors

• ✍️ Paperguide AI author – Easily write well-researched articles and academic papers with AI

• 👕 iMerch AI – Next-gen AI e-commerce tool offering intelligent product recommendations and personalized product lists

• ⚙️ Depth AI – Answer complex questions on large and messy codebases, onboard recent engineers quickly, and ship code faster

Latest AI Job Opportunities

• 📋 Findem – Technical Program Manager

• 📞 Cresta – SDR Manager, East

• 🗣 Palantir Technologies – Communications Strategist

• 🤖 Meta – Partnerships Operations Lead, AI

• 💊 Hippocratic AI – Business Development Executive, Pharma

QUICK HITS

xAI rolled out an upgraded version of Grok-2 to all X platform users, featuring tripled speed, improved multilingual capabilities, and integration of web search and advanced image generation features.

Meta’s FAIR released a set of latest AI research projects including Meta Motivo for embodied agent control and Meta Video Seal for video watermarking, alongside improved models for memory scaling and social intelligence.

OpenAI cofounder Ilya Sutskever warned that AI has reached ‘peak data’ through the NeurIPS conference, predicting a shift from current training methods to more autonomous, reasoning-based systems that can develop into increasingly unpredictable.

Google unveiled NotebookLM Plus with interactive audio features and Gemini 2.0 Flash integration, allowing users to verbally engage with AI hosts during Audio Overviews and access expanded enterprise capabilities.

OpenAI published recent email correspondences and a timeline of events with Elon Musk, claiming that Musk initially wanted the corporate to be a for-profit entity despite lively lawsuits.

DeepSeek released VL2, a brand new vision-language model family leveraging MoE architecture that performs similarly to rival models despite smaller sizes.

Anonymous-chatbot has returned to the LM Arena, which was previously used to check GPT 4o, sparking rumors of a possible GPT 4.5 or upgraded OpenAI model coming soon.

THAT’S A WRAP

SPONSOR US

Get your product in front of over 800k+ AI enthusiasts

Our newsletter is read by hundreds of tech executives, investors, engineers, managers, and business owners all over the world. Get in contact today.

See you soon,

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