Welcome, AI RESEARCHERS.

It's only been ONE week, and AI news has gotten crazy.

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

In today’s AI Researches:

• Anthropic Begins Research Into AI Emotions and Awareness

• Use AI in NotebookLM to organize your research better

• AI Researches

• AI TIPS of this week

• Crazy sayings from Anthropic's Chief Information Security Officer and DeepMind CEO on AI

• Best performing Ai tools of the week

• More Ai news and developments of this weekends

Read time: 5 minutes

 iMajor AI Developments that Happened in this week i

OpenAI is now offering its powerful image generator, gpt-image-1, to developers through an API, letting apps like Figma and Canva create detailed, editable AI images

Microsoft introduced new Copilot agents for research and data analysis, revealed tools for building autonomous agent systems, and reported that companies using AI are seeing better performance and employee outlooks.

Two Korean undergraduates built Dia, an open-source speech AI model that reportedly outperforms top commercial tools like ElevenLabs—without any funding.

Mechanize, a new AI startup, aims to fully automate white-collar jobs by training AI agents in virtual work simulations. Its goal is to replace human workers entirely.

OpenAI released o3 and o4-mini, advanced reasoning models with full tool access and image-based thinking, along with Codex CLI, an open-source coding agent for terminal use.

 IN LATEST 

🔥 Anthropic Begins Research Into AI Emotions and Awareness

AI Researches: Anthropic is kicking off a New research program focused on “model welfare,” asking one of AI’s biggest questions — could future AI systems be conscious, and if so, how should we treat them?

The Details:

• The new program will explore how to measure AI consciousness, identify signs of preferences or distress, and figure out how to respond if they exist.

• Kyle Fish, Anthropic’s first AI welfare researcher, believes there’s about a 15% chance that today’s models could already be conscious in some form.

• The research builds on a recent report co-authored by Fish, which argues that AI consciousness could be a near-term reality, not just sci-fi.

• Anthropic stressed that there’s no consensus in science yet — but the more advanced these models get, the harder the questions become.

Why this matters:
If AI ever crosses the line into consciousness, we’ll be forced to rethink ethics, rights, and even the definition of life — and the scary part? We won’t know where that line is until we’ve already crossed it.

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

✅ Use AI in NotebookLM to organize your research better

AI Researches: NotebookLM’s new “Sources” tool lets you enter a topic, shows up to ten helpful links with brief notes, and saves the ones you choose for easy summaries, questions, or audio.

Step-by-Step Guide:

• 1 Access NotebookLM & Upload Materials
  Visit notebooklm.google.com → sign in → Create new notebook → attach starter links, PDFs, Docs, or pasted text.

• 2 Open Discover Sources
  In the Sources panel, press Discover to launch the web-search dialog.

• 3 Enter a Precise Query
  Describe your focus clearly—e.g., “I want to dive more into the HQC algorithm”—to guide the scan.

• 4 Review & Add Key Sources
  NotebookLM lists up to 10 annotated recommendations → preview, then click Add to Notebook for the best items.

• 5 Take Notes & Generate Briefing Docs
  Highlight important passages or use Briefing Doc for an instant, structured overview of combined sources.

• 6 Chat for Quick Answers
  Ask follow-up questions in the chat pane to clarify terms, pull stats, or compare viewpoints without leaving the notebook.

• 7 Create Audio or Other Outputs
  Select Audio Overview for a podcast-style summary, or export content as PNG/PDF for easy sharing.

• 8 Refine with Discover Again
  Return to Discover, tweak the query, and repeat to fill gaps or branch into related sub-topics—research stays current with minimal effort.

Anthropic analyzed over 300,000 real conversations with its AI model Claude and identified 3,307 values grouped into five categories like Practical and Social. The study shows that Claude’s responses shift based on context, revealing that its moral judgments are flexible, not fixed.

In a new paper, DeepMind experts argue that AI has reached the limits of learning from human data and should now learn from continuous real-world experiences. Using signals like health data or test results, this method could help AI discover knowledge beyond human understanding while staying adaptable and safe.

Profluent’s new AI model, ProGen3, shows that larger models and datasets improve protein design, confirming AI scaling laws in biology. It can create novel antibodies and compact gene editors, speeding up drug discovery and gene therapy development.

Meta’s FAIR lab unveiled five new AI research projects focused on improving machine perception, video understanding, and team-based reasoning. Highlights include a state-of-the-art visual encoder, 3D object locator, and a benchmark for collaborative AI performance.

1.OpenAI is secretly adding hidden watermarks to AI text using invisible characters.

These characters look like normal spaces but are slightly different. You won’t notice them in regular apps, but they are used to mark long AI-generated texts—like essays that students might try to copy.

How to find out if your AI text is watermarked:

1. Use a special character viewer online or

2. Open the text in a code editor like Sublime Text or VS Code.
   These tools can show you hidden characters, and you'll see a unique pattern if the watermark is there.

How to remove the watermark:

• Just find and replace those special spaces with normal ones using any editor that shows hidden characters.

2.OpenAI’s new o3 and o4-mini models can now understand and think through images, not just text.

This means you can upload a screenshot, photo, or diagram, and ChatGPT will look at it and solve problems step-by-step, just like it reasons with words.
It’s similar to voice mode, but now with pictures.

How to use this feature properly:

• Clearly tell ChatGPT your goal:
  For example, say “Summarize this chart,” “Find mistakes in this code,” or “Translate this handwriting.”
  Clear instructions help it give the right answers.

• Use images when things are hard to explain with words:
  If your question depends on visual details like layouts, drawings, handwritten notes, or charts, uploading an image will work better than just typing.

• Don’t worry about bad images:
  Even blurry, rotated, or messy pictures can be understood. The model is designed to handle imperfect images.

Anthropic's Chief Information Security Officer, Jason Clinton, expects that within a year AI-powered virtual employees, capable of autonomous actions and memory retention, will become common on corporate networks, posing new cybersecurity risks around account management, access control, and liability.

DeepMind CEO Demis Hassabis said AI could eliminate all disease within 10 years by accelerating drug discovery. He also predicted AGI could arrive in 5–10 years, with AI systems already showing early reasoning and perception skills.

• 🗨️ ChatFlow – Speed up your growth and manage more customers with AI chatbots that automate support and boost engagement.

• 🧩 Fabric – Gather all your ideas, files, and links into one searchable, smart second brain that keeps everything organized.

• ⚙️ Agents by Athena AI – Create powerful AI agents that can handle complex tasks and make your workflows smoother.

• 📝 Kin – Reflect deeper, plan smarter, and stay connected with an AI assistant that learns and grows with you.

• 🎯 AIApply – Get interview-ready with a real-time AI coach that listens, reviews, and gives personal feedback based on your resume.

• 📬 Tanka – Chat smarter with the first AI messenger that remembers your conversations and helps you craft better replies.

• 📢 Cluely AI – Close more sales with AI that handles objections, suggests smart answers, and helps you seal the deal faster.

• 🧭 Rabbitholes AI – Explore deep conversations with different AI models on an endless canvas made for discovery.

Perplexity has launched its Perplexity Assistant app for iOS, enabling users to perform agent-like tasks, browse the web, and more through voice commands on mobile.

ByteDance’s Dreamina unveiled Seedream 3.0, a powerful text-to-image model now ranked second on Artificial Analysis’ Image Arena Leaderboard—just behind GPT-4o.

OpenAI is reportedly projecting revenue of $125 billion by 2029 and $174 billion by 2030, driven by AI agents, upcoming products, and expanding API and user adoption.

BMW will begin integrating DeepSeek’s AI models—developed by the Chinese startup—into its upcoming vehicles in the region, with rollout expected later this year.

Google launched Gemini 2.5 Flash, an AI model with strong reasoning and STEM skills, and a new “thinking budget” to control quality vs cost. It lets users toggle deeper reasoning for complex tasks while keeping simple responses efficient and affordable.

Adobe expanded its Firefly AI platform at MAX London by launching two upgraded image models, adding third-party model integrations, releasing new text-to-video and text-to-vector tools, introducing Firefly Boards for collaboration, and announcing a mobile app, all while maintaining IP-safe standards.

Google DeepMind expanded its Music AI Sandbox by launching the upgraded Lyria 2 model, adding new Create, Extend, and Edit features for musicians, introducing real-time music generation, and widening access to more U.S.-based creators.

OpenAI is expected to release an open-source reasoning model this summer that outperforms other open models on benchmarks and will come with a permissive usage license.

U.S. President Donald Trump signed an executive order to create an AI Education Task Force and launch the Presidential AI Challenge, aiming to bring AI into K–12 classrooms.

Loveable introduced Loveable 2.0, an updated app-building platform featuring multiplayer workspaces, an improved chat agent, a refreshed user interface, and additional enhancements.

Anthropic announced that Claude now features autonomous web and file search capabilities, along with Google Workspace integration, aimed at providing more accurate and context-rich responses for users.

Share Your Opinion