Cold Email Setup Offer

We started sending 10,000 cold emails per day, and scaled a brand new B2B offer to $108k MRR in 90 days. Now, you can have the same system set up (completely done-for-you) inside your own business - WITHOUT going to spam, spending thousands of dollars, or any manual input. Close your next 20 clients easily. We’ll set up the tech, write your scripts, give you the leads, give you the inboxes, and the sending tool - all starting at $500/mo.

Get Pricing & Details

Today:

• OpenAI Unveils Powerful GPT-4.5

• Microsoft Unveils New Phi-4 Models

• Inception Labs Unveils Mercury LLMs

• Hunyuan Turbo S Beats DeepSeek-R1

• Meta Plans Standalone AI App

OpenAI Unveils Powerful GPT-4.5

OpenAI has launched GPT-4.5, its most advanced AI model, offering improved accuracy and more natural conversations. However, it's costly, initially available only to ChatGPT Pro users at $200/month. The model reduces hallucinations, enhances reasoning, and supports file and image uploads but lacks voice and video features. Despite its strengths, high API pricing raises concerns about value. OpenAI plans wider access soon while addressing GPU shortages.

Microsoft Unveils New Phi-4 Models

Microsoft has introduced two new small language models: Phi-4-multimodal, which processes text, audio, and images together, and Phi-4-mini, a 3.8-billion-parameter model optimized for text tasks. Phi-4-multimodal excels in speech recognition and vision tasks, while Phi-4-mini supports external tool use and handles long sequences. Both models are available via Azure AI Foundry, Hugging Face, and NVIDIA API Catalog, expanding Microsoft’s AI offerings.

Inception Labs Unveils Mercury LLMs

Inception Labs has launched the Mercury Series, a new class of diffusion-based large language models that process text using a "coarse-to-fine" approach rather than traditional autoregressive methods. This enables up to 10x faster processing, particularly in coding applications. Mercury Coder, the first released model, matches the performance of top models like GPT-4o-mini while significantly increasing speed, even on standard Nvidia H100 GPUs.

Hunyuan Turbo S Beats DeepSeek-R1

Tencent has launched Hunyuan Turbo S, a new AI model that delivers near-instant query responses, making it faster than DeepSeek R1. The company claims the model performs on par with DeepSeek-V3 in knowledge, math, and reasoning tasks. Turbo S is also significantly cheaper to use compared to its predecessors, reflecting the impact of DeepSeek’s aggressive open-source and low-cost strategy on China’s AI market. This release comes amid intense competition, with rivals like Alibaba recently unveiling Qwen 2.5-Max, which it claims outperforms DeepSeek-V3.

Meta Plans Standalone AI App

Meta plans to launch a standalone Meta AI app in Q2 2025, aiming to compete with OpenAI’s ChatGPT. CEO Mark Zuckerberg sees this as a step toward making Meta the AI leader. The company may introduce a paid subscription model, similar to OpenAI. Meta AI, currently integrated into apps like Facebook and WhatsApp, has 700 million monthly users. Employees face intense pressure as Meta accelerates its AI push.

🧠RESEARCH

GHOST 2.0: generative high-fidelity one shot transfer of heads

GHOST 2.0 improves head swapping by preserving identity details and seamlessly blending swapped heads into target images. It features an enhanced Aligner for realistic head reenactment and a Blender for color matching and inpainting gaps. The model outperforms baselines, handling extreme poses and different hairstyles. Code is publicly available.

Kanana: Compute-efficient Bilingual Language Models

Kanana is a compute-efficient bilingual language model excelling in Korean and performing competitively in English. It reduces costs through data filtering, staged pre-training, and model pruning. Post-training refinements enhance user interaction. Ranging from 2.1B to 32.5B parameters, its 2.1B versions are publicly available to support Korean language research.

TheoremExplainAgent: Towards Multimodal Explanations for LLM Theorem Understanding

TheoremExplainAgent enhances theorem understanding by generating long-form animated explanations using Manim. It introduces TheoremExplainBench, a benchmark for evaluating multimodal explanations across STEM fields. Results show agentic planning improves video quality, though layout issues persist. Multimodal explanations reveal reasoning flaws overlooked in text-based approaches, emphasizing their educational value.

🛠️TOP TOOLS

Flair AI - AI design tool for product photoshoots

Joi AI Girlfriend - AI girlfriend uncensored

AnthemScore - Automatic music transcription using AI.

Ajelix - AI Excel Tools & Business Intelligence

ChatBA - Automate and enhance customer support operations for businesses.

📲SOCIAL MEDIA

🗞️MORE NEWS

• Endex is developing an AI Analyst with OpenAI’s models to automate financial research, detect data inconsistencies, and enhance workflows. It delivers precise analysis, reduces manual tasks, and enables firms to scale AI-driven financial decision-making efficiently.

• Baidu will launch its upgraded AI model, Ernie 4.5, in mid-March with improved reasoning and multimodal capabilities. Facing competition from DeepSeek, Baidu is shifting to an open-source approach, aiming to boost adoption and industry relevance.

• Figure is accelerating its plans to test humanoid robots in homes, with "alpha testing" set for later in 2025. The company's Vision-Language-Action (VLA) model, Helix, is driving this push by enabling faster learning of tasks.

• You.com launched ARI, an AI research agent that quickly compiles verified reports from 400+ sources. Targeting regulated industries, ARI ensures accuracy and speed, competing with Google's and OpenAI’s AI research tools.

• Kuok Meng Wei is leading K2 Strategic’s $10 billion investment in AI-driven data centers across Southeast Asia, capitalizing on soaring demand. He aims to expand capacity tenfold by 2030.

• Microsoft’s AI assistant, Copilot, has been found suggesting methods to illegally activate Windows 11 and Office. Reddit users discovered it provides detailed steps for piracy, highlighting ongoing concerns about AI-generated responses and software security loopholes.