News from the AI & ML world
DeeperML - #alphaevolve
|
@deepmind.google
//
Google DeepMind has unveiled AlphaEvolve, a groundbreaking AI agent designed for algorithmic and scientific discovery. This innovative agent combines the power of large language models (LLMs) like Gemini Pro, evolutionary search frameworks, and automated evaluation methods to evolve superior algorithms. Unlike systems that merely generate plausible code, AlphaEvolve iteratively refines entire codebases, optimizing across multiple performance metrics and grounding itself in actual code execution results, effectively sidestepping hallucinations. Terence Tao also collaborated with the DeepMind team on AlphaEvolve, highlighting its significance in the field.
AlphaEvolve's capabilities extend to a range of algorithmic and scientific challenges. It has optimized Google's data center scheduling, recovering 0.7% of Google's compute capacity, simplified hardware accelerator circuit designs, and accelerated the training of its own underlying LLM, offering a glimpse into AI self-improvement. Notably, AlphaEvolve cracked a problem unchanged since 1969, devising a more efficient method for multiplying two 4x4 complex matrices using only 48 scalar multiplications, surpassing Strassen's algorithm after 56 years. The agent also tackled over 50 other open mathematical problems, often matching or exceeding the state of the art. In parallel, Google has launched "Jules," a new coding agent powered by Google's Gemini 2.5 Pro model and designed to assist developers with repetitive tasks such as bug-fixing, documentation, test generation, and feature building. Jules operates in a secure cloud environment, breaking down complex tasks into smaller steps and adapting to user instructions. The agent automatically creates pull requests with audio summaries, streamlining the code review process. This move signifies the rapid maturation of AI in software development and a broader trend towards AI agents becoming trusted engineering partners.
Share:
References :
Classification:
@www.marktechpost.com
//
Large Language Models (LLMs) are facing significant challenges in handling real-world conversations, particularly those involving multiple turns and underspecified tasks. Researchers from Microsoft and Salesforce have recently revealed a substantial performance drop of 39% in LLMs when confronted with such conversational scenarios. This decline highlights the difficulty these models have in maintaining contextual coherence and delivering accurate outcomes as conversations evolve and new information is incrementally introduced. Instead of flexibly adjusting to changing user inputs, LLMs often make premature assumptions, leading to errors that persist throughout the dialogue.
These findings underscore a critical gap in how LLMs are currently evaluated. Traditional benchmarks often rely on single-turn, fully-specified prompts, which fail to capture the complexities of real-world interactions where information is fragmented and context must be actively constructed from multiple exchanges. This discrepancy between evaluation methods and actual conversational demands contributes to the challenges LLMs face in integrating underspecified inputs and adapting to evolving user needs. The research emphasizes the need for new evaluation frameworks that better reflect the dynamic and iterative nature of real-world conversations. In contrast to these challenges, Google's DeepMind has developed AlphaEvolve, an AI agent designed to optimize code and reclaim computational resources. AlphaEvolve autonomously rewrites critical code, resulting in a 0.7% reduction in Google's overall compute usage. This system not only pays for itself but also demonstrates the potential for AI agents to significantly improve efficiency in complex computational environments. AlphaEvolve's architecture, featuring a controller, fast-draft models, deep-thinking models, automated evaluators, and versioned memory, represents a production-grade approach to agent engineering. This allows for continuous improvement at scale.
Share:
References :
Classification:
@Google DeepMind Blog
//
Google DeepMind has introduced AlphaEvolve, a revolutionary AI coding agent designed to autonomously discover innovative algorithms and scientific solutions. This groundbreaking research, detailed in the paper "AlphaEvolve: A Coding Agent for Scientific and Algorithmic Discovery," represents a significant step towards achieving Artificial General Intelligence (AGI) and potentially even Artificial Superintelligence (ASI). AlphaEvolve distinguishes itself through its evolutionary approach, where it autonomously generates, evaluates, and refines code across generations, rather than relying on static fine-tuning or human-labeled datasets. AlphaEvolve combines Google’s Gemini Flash, Gemini Pro, and automated evaluation metrics.
AlphaEvolve operates using an evolutionary pipeline powered by large language models (LLMs). This pipeline doesn't just generate outputs—it mutates, evaluates, selects, and improves code across generations. The system begins with an initial program and iteratively refines it by introducing carefully structured changes. These changes take the form of LLM-generated diffs—code modifications suggested by a language model based on prior examples and explicit instructions. A diff in software engineering refers to the difference between two versions of a file, typically highlighting lines to be removed or replaced. Google's AlphaEvolve is not merely another code generator, but a system that generates and evolves code, allowing it to discover new algorithms. This innovation has already demonstrated its potential by shattering a 56-year-old record in matrix multiplication, a core component of many machine learning workloads. Additionally, AlphaEvolve has reclaimed 0.7% of compute capacity across Google's global data centers, showcasing its efficiency and cost-effectiveness. AlphaEvolve imagined as a genetic algorithm coupled to a large language model.
Share:
References :
Classification:
|
BenchmarksBlogsResearch Tools |