GPT-5 Arrives in Microsoft Copilot

Hey there 👋

We hope you're excited to discover what's new and trending in AI, ML, and data science this week.

Here is your 5-minute pulse...

print("News & Trends")

After GPT-5’s Debut: OpenAI's Next Moves Revealed (4 min. read)

Over dinner in San Francisco, Sam Altman admitted GPT-5’s debut felt anticlimactic compared to GPT-4, prompting OpenAI to reinstate GPT-4o and the model picker after user complaints about tone and performance standing roughly even with peers. Yet API usage doubled and GPU demand soared. Altman hinted that OpenAI is aiming to evolve into something much broader than just ChatGPT as he teases bets on hardware, data centers, and an AI-powered social platform.

GPT-5 Arrives in Microsoft Copilot with Smart Agent Routing (4 min. read)

Microsoft just dropped GPT-5 into Microsoft 365 Copilot and Copilot Studio, giving you real-time agent routing so your AI picks between a speed-boosted or deep-reasoning model based on task complexity, enhancing multi-turn coherence, accuracy, and tool orchestration. This is a global rollout across web, desktop, and mobile.

Claude Adds Feature to End Harmful Conversations (4 min. read)

Anthropic now gives Claude Opus 4 and 4.1 the power to end conversations in rare, extreme cases, think repeated requests for sexual content involving minors or violent instructions, after multiple refusals and redirections. This experimental feature, part of their “model welfare” initiative, lets the AI step away from unproductive or harmful interactions without affecting other chats, unless the user is at risk of self-harm or harming others. Most users will likely never encounter it.

Meta Unveils DINOv3 And Sets a New Standard for Self-Supervised Vision (5 min. read)

DINOv3 is Meta’s new self-supervised vision foundation model that learns rich, dense features without any labels, using carefully scaled data and model design alongside a new technique called Gram anchoring to prevent feature degradation during long training. It supports resolution-robust representations and outperforms previous state-of-the-art models across vision tasks without fine-tuning, and comes in a family of models from lightweight to massive for flexible deployment.

print("Applications & Insights")

Fast Prototyping of GenAI Apps with Streamlit (Course)
This self-paced DeepLearning.AI course teaches you to go from just a few lines of Python into a fully interactive GenAI app using Streamlit and Snowflake, so you can prototype chatbots in hours not days. It guides you through MVP workflows, prompt engineering, RAG, and deployment on Snowflake or Streamlit Community Cloud, all built around fast feedback loops, secure data, and practical iteration.

Agentic AI: On Evaluations (16 min. read)
The playbook outlines how evaluating agentic AI requires new approaches beyond static benchmarks, focusing on dynamic tasks like RAG, chatbots, and multi-step planning. It highlights frameworks such as DeepEval, RAGAS, and OpenAI Evals, and stresses the need for human-in-the-loop judgment, task-specific metrics, and continuous evaluation pipelines. The key message is that effective assessment must evolve alongside agent capabilities to ensure reliability, safety, and real-world utility.

Using GPT-5 to Use a Computer (7 min. read)
Archon is a lightweight desktop copilot that makes your computer self-driving by combining GPT-5’s multi-step reasoning with a custom, fine-tuned vision model. A tiny bar at the bottom of your screen takes natural-language commands, snaps screenshots, plans actions with GPT-5, and then executes clicks or keystrokes with speed and precision. In a racing game demo it interpreted “start playing,” recognized the interface, and navigated the track—showing how strong instruction following can unlock powerful UI control.

The State of AI 2025 (10 min. read)
AI startups are growing like nothing we've ever seen - some hit $100M ARR in their first year. Bessemer studied 20 companies and found two types: Supernovas that explode to $125M by year two but have 25% gross margins and fragile retention, and Shooting Stars that quadruple yearly with 60% margins.

print("Tools & Resources")

TRENDING MODELS

text-generation
google/gemma-3-270m
⇧ 6.49k Downloads
A compact large language model developed by Google, ideal for text generation tasks and lightweight deployments. Suitable for research and downstream applications requiring efficient inference.

image-text-to-text
zai-org/GLM-4.5V
⇧ 9.5k Downloads
GLM-4.5V is a large multimodal AI model that can process images and text to generate text-based responses. It's used for vision-language tasks, combining image understanding with natural language.

image-to-video
tencent/Hunyuan-GameCraft-1.0
⇧ 382 Downloads
Hunyuan-GameCraft-1.0 generates video from images, focusing on advanced game environments and real-world simulations. It's designed for synthetic media content creation and video AI research.

text-generation
openai/gpt-oss-20b
⇧ 3.65M Downloads
GPT-OSS-20B is OpenAI's 20B parameter open-source language model for advanced conversational AI and text generation. It supports a wide range of applications in research and enterprise.

image-to-image
Qwen/Qwen-Image-Edit
⇧ 347 Downloads
Qwen-Image-Edit excels in image-to-image transformation and editing with deep learning. It enables precise editing capabilities for photos and digital images within creative and technical workflows.

TRENDING AI TOOLS

• 🕸 Comet Browser: An AI-powered browser that acts as your assistant for research and daily tasks.

• 🤖 Genspark AI Developer: An all-in-one AI workspace for building and iterating with custom agents.

• 🆕 GenFlow 2.0: Baidu’s platform that runs 100+ AI agents in parallel for complex tasks.

📣 Before you go, a quick message from our sponsor:

Unlock the Power of AI With the Complete Marketing Automation Playbook

Discover how to scale smarter with AI-driven workflows that actually work. This playbook delivers:

• A detailed audit framework for your current marketing workflows

• Step-by-step guidance for choosing the right AI-powered automations

• Pro tips for improving personalization without losing the human touch

Built to help you automate the busy work and focus on work that actually makes an impact.

Steal the Playbook

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 pulse experience for you.

See you soon,

Andres