• AI Academy
  • Posts
  • 🤖 Anthropic Breaks the Mold with Hybrid Reasoning Model

🤖 Anthropic Breaks the Mold with Hybrid Reasoning Model

Plus: Google's AI Scientist and Microsoft's Quantum Gamble

February 27, 2025

Hello AI Enthusiast,

Anthropic's Claude 3.7 Sonnet introduces a hybrid approach that lets users choose between quick answers or visible step-by-step reasoning, while Google's AI Co-Scientist system shows promising results in real-world lab experiments. Meanwhile, Microsoft makes a bold claim in quantum computing with a chip design they say could revolutionize the field. Let's dive into what these news might mean.

Before we get to the news, a quick reminder that our 11th edition of the 7-week GenAI Project Bootcamp kicks off next week! If you're looking to learn how to identify and build AI projects, this is your chance. Some of our graduates have even transitioned into AI consulting roles in their fields of expertise after completing the program. This hands-on course will give you the skills to not just create one AI project, but to develop a framework for building many more in the future.

The Big Picture 🔊

Anthropic Launches Claude 3.7 Sonnet

Anthropic has released Claude 3.7 Sonnet, their most intelligent model yet that uniquely combines fast responses with visible step-by-step thinking in a single system. The model shows particularly strong coding capabilities and comes alongside Claude Code, a new terminal-based tool that allows developers to delegate engineering tasks directly from their command line. Available on all Claude plans with the same pricing as previous versions, this approach differs from competitors who offer separate models for reasoning tasks.

Helin Yontar
Helin YontarCPO and Polyglot

Anthropic's hybrid approach makes sense from a user experience perspective as users can now choose when to activate reasoning mode. This is different from what Sam Altman hinted about GPT-5, where the model itself might decide which capability to use based on the request.

The focus on real-world coding tasks seems to be paying off, especially considering coding assistance is one of AI's most profitable markets right now. With Cursor being the fastest company to reach $100M in AR, Claude Code is clearly aimed at capturing part of this lucrative market.

Google Rolled Out AI Co-Scientist

Google has introduced "AI co-scientist," a multiagent system built with Gemini 2.0 designed to function as a virtual scientific collaborator. The system uses specialized agents that work together to generate novel research hypotheses and proposals by mimicking the scientific method. Google validated its effectiveness through laboratory experiments in three biomedical applications, including drug repurposing for leukemia and explaining mechanisms of antimicrobial resistance.

Gianluca Mauro
Gianluca MauroFounder and AI Rockstar

What's interesting is that this system creates novel research hypotheses rather than just compiling existing information. However, it's not actually a new model, it's essentially a series of prompts working with Gemini 2.0. When they describe all these "agents" (generation, reflection, ranking), they're basically using different prompts that talk to each other in a structured workflow.

The approach is clever, one prompt generates ideas, another critiques them, another ranks them, but calling this an "agent" stretches the definition.

Microsoft Unveils Revolutionary Quantum Chip

Microsoft has introduced Majorana 1, a groundbreaking quantum chip with a new Topological Core architecture. The chip uses a special material called a "topoconductor" that creates more stable qubits - the building blocks of quantum computers. Microsoft claims this design could lead to a million qubits on a single palm-sized chip, potentially enabling quantum computers to solve complex problems that today's computers can't handle.

Gioele Mottarlini
Gioele MottarliniCOO and Image Addict

The contrast between AI's openness and quantum computing's secrecy is striking. Microsoft claims a major breakthrough but shares minimal technical details, making scientists skeptical.

In AI, we expect transparency. When OpenAI, for example, doesn't explain their technology, we protest. This openness is why AI advances so quickly.

Bits and Bobs 🗞️

  • Microsoft Research has developed Muse, a generative AI model that creates complex gameplay sequences by learning from massive amounts of human gameplay data.

  • Perplexity has open-sourced R1-1776, a post-trained version of DeepSeek-R1 designed to remove censorship and provide unbiased, factual responses while maintaining strong reasoning capabilities.

  • Researchers have created an AI tool that diagnoses various infections by analyzing gene sequences, offering a comprehensive view of the immune system's disease record.

  • OpenAI introduces SWE-Lancer, a benchmark of 1,400 real-world freelance software engineering tasks worth $1 million, evaluating whether AI models can complete them.

  • Spotify is now accepting audiobooks narrated by digital voices, thanks to a collaboration with ElevenLabs.

  • Google has revealed that its AI video model, Veo 2, will cost $0.50 per second of video, totaling $30 per minute.

  • Perplexity is developing its own web browser, aiming to reinvent the browsing experience.

From Our Channels 🤳

Check out Gianluca's TikTok discussing a study that found labeling products as "AI-powered" doesn't actually change how people view them. What truly matters is whether the technology solves real problems. The same principle at the heart of our Bootcamp.

@gianluca.mauro

How customers perceive the value of AI #Ai #learnontiktok #artificialintelligence #business #machinelearning #product #ux #entrepreneur

LOLgorithms 😂

How about a breakthrough in fixing Teams?

That's a wrap on our newsletter! Before you go, here’s a quick recap of our offerings:

Catch you next week! 👋