In partnership with

Welcome to AI Entrepreneurs

Explore the transformative power of AI across industries in this issue! From AlphaFold 3’s groundbreaking open-source release revolutionizing drug discovery to Anthropic and OpenAI’s insights on the future of AGI, we cover the latest advancements shaping the fields of healthcare, coding, video tech, and more. Dive into our curated highlights to see how these AI-driven innovations are setting new standards and redefining possibilities for the future.

🔬 AlphaFold 3 Open-Sourced: A New Era for Drug Discovery & Biology

Google DeepMind’s groundbreaking protein prediction model is now available for academic research.

Google DeepMind has open-sourced AlphaFold 3, its Nobel Prize-winning AI model for protein prediction, providing academic researchers full access to both the code and model weights for the first time. Released with restrictions on commercial use, AlphaFold 3 promises to reshape the field of molecular biology.

Key Highlights:

• Molecular Interactions: Predicts interactions between proteins and molecules like DNA, RNA, and drug compounds.

• Academic Access: Available for non-commercial use; commercial rights retained by Isomorphic Labs.

• Broad Impact: The model has already mapped over 200 million protein structures.

• Industry Influence: Other tech giants, including Baidu and ByteDance, are developing similar models based on AlphaFold’s specs.

• Commercial Partnerships: Isomorphic Labs holds exclusive commercial rights, with $3 billion secured in pharmaceutical deals.

The open-source release of AlphaFold 3 levels the playing field for researchers worldwide, accelerating breakthroughs in biology and medicine by giving scientists outside of large pharmaceutical companies access to cutting-edge AI technology. This move opens new doors for disease research, drug discovery, and more, marking a major advancement for AI-powered science.

🚀 The Road to AGI: Perspectives from Anthropic & OpenAI Leaders

Inside the Views of Anthropic’s Dario Amodei and OpenAI’s Sam Altman on Achieving AGI and Its Impact

The path toward Artificial General Intelligence (AGI) is a highly anticipated milestone in AI research. AGI, often defined as an AI with human-level capabilities across a wide range of tasks, is still theoretical. Yet leaders in the field, like Dario Amodei of Anthropic and Sam Altman of OpenAI, suggest that it may be closer than expected.

Dario Amodei’s Views:

• Timeline: Amodei predicts that AGI could be achieved by 2026 or 2027, assuming no major obstacles arise.

• Progress: He notes the rapid advancement of AI capabilities, moving from high school to PhD level in a few years.

• Challenges: Potential issues include data limitations, scaling difficulties, and geopolitical risks, but he believes these are unlikely to cause significant delays.

• Optimism: Amodei is optimistic, stating that the number of convincing blockers to AGI has been decreasing rapidly.

Sam Altman’s Views:

• Imminence: Altman also believes that AGI is on the horizon and could be realized soon.

• Transformative Potential: He emphasizes the transformative impact AGI will have on various sectors, including the need for careful management and ethical considerations.

• Management: Altman stresses the importance of responsible development and deployment of AGI to ensure it benefits humanity.

Both leaders are optimistic about AGI's near-term arrival and stress addressing its challenges and ethics. AGI could revolutionize fields like medicine and climate science but poses ethical and regulatory challenges. Experts call for careful development and oversight to ensure societal benefits and manage risks to jobs, privacy, and global stability.

🚀 Qwen2.5-Coder Challenges Claude: The New SOTA Code LLM

Bringing Power, Diversity, and Practicality to Open Source Code Generation.

The Qwen team has introduced Qwen2.5-Coder, an advanced open-source code language model series, raising the bar in coding capabilities across various programming needs. The series, especially the flagship 32B model, boasts state-of-the-art (SOTA) coding capabilities, impressive multi-language support, and practical applications in real-world scenarios.

Key Highlights:

• SOTA Performance: Qwen2.5-Coder-32B-Instruct matches GPT-4o in coding benchmarks like EvalPlus and LiveCodeBench.

• Diverse Model Sizes: Six sizes, from 0.5B to 32B, meet diverse developer needs and computing resources.

• Multi-Language Proficiency: High scores in over 40 languages, including Haskell and Racket, through McEval and MdEval benchmarks.

• Code Repair Excellence: Strong repair capabilities, scoring 75.2 on MdEval, reducing the cost of learning unfamiliar languages.

• Human Alignment: Enhanced preference alignment with human preferences, tested via the Code Arena benchmark.

Qwen2.5-Coder models offer multi-language support, human preference alignment, and code reasoning for generating and repairing code. Available in base and instruct versions, the 32B model uses Fill-in-the-Middle mode for precise code completion. These models enhance coding workflows and support non-native developers. The open-source release provides a scalable platform for developers and researchers, fostering innovations in code-based AI applications.

🎥 ReCapture: Google’s AI Tool for Retrofitting Camera Moves in Video

Professional Camera Moves, No Pro Skills Required

Google’s ReCapture offers a groundbreaking way for anyone to add professional-level camera movements to videos, even after they’re filmed. Traditional methods have struggled with this, but ReCapture uses an AI-driven, two-phase process to make it possible. First, the tool creates an "anchor video" with the new camera angles, though this initial draft may still have some visual inconsistencies.

The second phase enhances the anchor video through "masked video fine-tuning" using LoRA layers—one temporal and one spatial—to ensure smooth, realistic motion that stays true to the video’s original detail. While promising, ReCapture is still in the research phase, with commercial applications potentially years away.

🚀 FrontierMath Exposes AI’s Math Limits

Top AI models face a new challenge in advanced problem-solving

A groundbreaking benchmark called FrontierMath, crafted by over 60 top mathematicians, reveals a key weakness in current AI models: their struggle with complex mathematical reasoning. Despite excelling at simpler tests, leading AI systems like GPT-4o, Claude 3.5, and Gemini 1.5 Pro solved less than 2% of FrontierMath’s problems, which span intense number theory and abstract algebraic geometry—tasks that can take expert mathematicians hours or days to complete.

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

🚨Learn to Build Agentic Memory with LLMs

Unlock advanced memory capabilities in LLM applications with hands-on tools and expert guidance.

DeepLearning.AI and Letta AI have launched a free course on building agentic memory into applications with Large Language Models. Taught by Letta’s founders, Charles Packer and Sarah Wooders, this course explores the cutting-edge MemGPT framework, focusing on advanced memory management within LLMs.

Learn to create agents with self-editing memory, customizable memory blocks, and multi-agent collaboration using Letta’s open-source tools for equipping LLMs with persistent, efficient memory.

Enroll for free here ➡️

🎶 The Beatles’ Final Song Restored with AI Earns Grammy Nod

‘Now and Then’ takes on Beyoncé, Taylor Swift, and Billie Eilish

The Beatles’ newly completed song “Now and Then” has earned two Grammy nominations nearly 50 years after the band split. Restored using AI, this final track—originally a John Lennon demo from the late 1970s—was finished by Paul McCartney and Ringo Starr with machine learning tech that isolated Lennon’s vocals.

Now, it's up for Record of the Year and Best Rock Performance, competing against today’s biggest names like Beyoncé and Taylor Swift. 🎸🏆 The awards on February 2nd will reveal if this timeless piece resonates with a new era of music fans.

🏥AI in Healthcare

AIHealthTech Insider: Issue #22

Stay ahead with the latest AI-driven healthcare advancements! This issue dives into groundbreaking biomarker discoveries for cardiovascular health and AI tools for diagnosing complex conditions like long COVID and PCOS. Discover how artificial intelligence is reshaping modern medicine for better outcomes.

🔔AI in Social Media

Startup Success: Instant Access to AI Entrepreneur Courses!