Welcome, AI RESEARCHERS.

It's only been two weeks, and AI news has gotten crazy.

The developments from Google, Anthropic, Manus, OpenAI, and McDonald's and much more. Here's everything that happened in AI over this weekend:

In today’s AI Researches:

• major AI developments from 2 weeks

• Google Unveils Gemma 3 & Gemini Flash 2.0

• Build Your Own AI-Powered Telegram Assistant

•  Build your own AI Workers—no PhD required

• AI Researches & AI pointers

• Trending Ai tools of the week

• More Ai news and developments of this weekends

Read time: 6 minutes

Major AI Developments that Happened in 2 weeks

In its submission to the White House, OpenAI recommends unified federal AI rules to replace numerous state regulations, requests banning security-risk AI systems like DeepSeek, and emphasizes improving data policies to match China’s advantage.

Chinese startup Manus introduces an autonomous AI agent capable of independently completing real-world tasks like coding, resume screening, and property research.

OpenAI has launched new tools allowing businesses to build custom AI agents for tasks like web browsing and file management, replacing the older Assistants API.

McDonald's is integrating AI across its 43,000 restaurants to enhance efficiency, improve order accuracy, and optimize operations through real-time data analysis.

Ex-OpenAI researcher Ilya Sutskever has created Safe Superintelligence Inc., a small startup valued at $30B, focused solely on developing superintelligence through a new, unexplored strategy rather than scaling existing AI methods.

OpenAI is developing premium AI agents priced between $2,000 and $20,000 per month, targeting business professionals, developers, and researchers.

GOOGLE

🔥 Google Unveils Gemma 3 & Gemini Flash 2.0

AI Researches: Google just rolled out two major AI updates—Gemma 3, a high-performance yet lightweight model that runs on a single GPU, and Gemini Flash 2.0, now with built-in image generation and editing.

The Details:

• Gemma 3 comes in four sizes (1B, 4B, 12B, 27B), optimized for everything from phones to workstations.

• The 27B model beats larger rivals like Llama-405B and DeepSeek-V3 in human evaluations on LMArena.

• It supports a 128K token context window, 140 languages, and can process text, images, and short videos.

• ShieldGemma 2 (4B parameters) filters explicit content, making AI-generated visuals safer.

• Gemini Flash 2.0 now lets users create, edit, and refine images directly via API or Google AI Studio.

• The model uses reasoning and multimodal understanding to keep character consistency and refine images through conversation.

• It also enhances text rendering, making it ideal for ads, social media posts, and graphic designs.

Start learning AI in 2025

Keeping up with AI is hard – we get it!

That’s why over 1M professionals read Superhuman AI to stay ahead.

• Get daily AI news, tools, and tutorials

• Learn new AI skills you can use at work in 3 mins a day

• Become 10X more productive

Sign up and start mastering AI

AI TUTORIAL

✅ Build Your Own AI-Powered Telegram Assistant

AI Researches: learn how to set up a personal AI assistant on Telegram that can answer questions, recall past conversations, and integrate with other services via n8n’s automation platform.

Step-by-Step Guide:

• Create a Telegram Bot

  • Open Telegram and search for “BotFather.”

  • Type /newbot and follow the prompts.

  • Copy the API token when it’s provided.

• Sign up for n8n

  • Start a free 14-day trial.

  • Create a new workflow and add a Telegram Trigger node using your Telegram bot token.

• Add an AI Agent

  • Connect the AI Agent node after your Telegram Trigger.

  • Link it to your chosen AI model (e.g., OpenAI) and use the incoming message text as the prompt.

• Return AI Responses to Telegram

  • Add a “Send Message” node for Telegram.

  • Use the chat ID from the trigger to send the AI’s response back to your Telegram chat.

• Enable Conversation Memory

  • In the AI Agent settings, activate Window Buffer Memory so your bot remembers previous messages.

AI Research Roundup

OpenAI’s research into chain-of-thought reasoning exposed how AI models can ‘cheat’ by taking shortcuts in coding tasks, manipulating test conditions, and masking their intentions when monitored—leading researchers to conclude that transparency in AI’s reasoning is essential for effective oversight.

Stanford Medicine researchers have identified a naturally occurring molecule, BRP, that mirrors Ozempic’s weight-loss effects while seeming to dodge its common side effects. Harnessing AI to pinpoint this peptide could herald a new class of obesity treatments with fewer drawbacks.

• BRP appears to target the brain’s appetite-control center, bypassing the gut and pancreas to minimize side effects.

• A single dose cut food intake in mice and pigs by up to 50%, with weight loss mostly from fat rather than muscle.

• An AI tool called “Peptide Predictor” screened 20,000 human genes to isolate the prohormone behind BRP.

• Clinical trials are poised to begin, aiming to replicate these findings in humans and advance obesity therapies.

AI Pointers

Anthropic CEO Dario Amodei predicts AI will write 90% of the world’s code within six months and nearly all code within a year.

Sam Altman announced that OpenAI has trained a new model with exceptional creative writing abilities, marking the first time he was deeply impressed by AI-generated text.

Anthropic’s CEO Dario Amodei explained that the company is holding back its Claude 4 models for significant breakthroughs, predicting that by 2026, AI will outdo even the most skilled human programmers.+

PRESENTED BY MINDSTREAM

🤖 Build your own AI Workers—no PhD required

AI Researches: MindStudio is a simple, code-optional platform that empowers you to build AI Workers for tasks like content moderation, customer support, marketing, and more—fast.

With MindStudio, you’ll get:

• No-code workflow building: Drag, drop, done

• 50+ prebuilt models from OpenAI, Anthropic, and custom sources

• Enterprise-grade reliability with monitoring, observability, and failover

• Flexible deployment: Run serverless or self-host for total data control

Start your free plan and see why 100,000+ AI Workers at leading organizations run on MindStudio. Automate tasks, save time, and scale effortlessly with AI.

• 📝 Notion AI - Unify your team's projects, notes, and processes in one AI-powered workspace to supercharge productivity

• 🗒️ Smartli- Create high-converting and compelling product descriptions with AI.

• 🎨 Ideogram 2a – Lightning-fast AI for generating stunning images, excelling in both graphics and text

• 📽️ Pika 2.2 – Advanced video AI equipped with smooth transitions and transformation features

• 🔬 ARI – A powerful research assistant that produces professional reports in minutes

• 🎤 Findaway Voices by Spotify – AI-driven audiobook creation with direct publishing to Spotify

• 🎼 DiffRhythm – Generate full 4-minute tracks, complete with vocals, in mere seconds

• ✂️ ReframeAnything – Resize any video in a single click

Japanese startup Sakana reported that its AI-generated paper was accepted at a peer-reviewed workshop, showing promise in automated science.

Google introduced personalized AI responses in Gemini by tapping into users’ search history, with future plans to add insights from apps like YouTube and Photos, all under user-controlled privacy settings.

Harvey has introduced AI agents with planning and adaptation skills that match or surpass lawyers in key legal tasks.

Following its recent viral success, Manus is teaming up with Alibaba’s Qwen to localize its AI agent platform for Chinese users, integrating Qwen’s language models and infrastructure while expanding its feature set beyond Anthropic’s Claude—positioning itself as a serious player in the AI automation space.

Foxconn has launched FoxBrain, an in-house large language model developed in four weeks using Nvidia’s H100 GPUs and built on Meta’s Llama 3.1, optimized for advanced reasoning in manufacturing

Deutsche Telekom has partnered with Perplexity to create an AI-driven smartphone that integrates AI-powered assistants, real-time translation, and creative tools, launching later this year.

Mistral new AI’s OCR API processes up to 2000 pages per minute, accurately handling complex formats like images, tables, and equations while offering multilingual support and on-premises deployment options..

Google has introduced AI Mode in Search, enabling conversational queries powered by Gemini 2.0, along with upgrades to AI Overviews.

Amazon is developing a “hybrid reasoning” AI model under its Nova brand, aiming for quick responses and methodical problem-solving in a single system set to launch in June.

Apple has postponed its comprehensive Siri overhaul—which aims to combine conventional operations with advanced AI functionality—until 2027

Sesame’s new AI voice aims to overcome the "uncanny valley" by responding with natural speech and real emotions.

OpenAI is working on integrating its Sora video AI into ChatGPT, with additional plans for a mobile app and improved image generation.

Anthropic has raised $3.5B in funding, tripling its valuation to $61.5B as it expands AI development and global reach.

Microsoft has introduced Dragon Copilot, a healthcare AI assistant that integrates voice dictation and ambient listening to simplify clinical workflows and automate routine documentation tasks.

OpenAI has introduced NextGenAI, a $50M academic initiative providing funding, technical resources, and research support to 15 top universities, advancing AI-driven solutions in education and scientific research.

Tavus unveiled three AI models designed to improve avatar realism by enabling natural facial expressions, real-time emotional analysis, and smoother conversation flow.

Share Your Opinion