self-improvement

Ornith-1.5 is a new AI model that introduces end-to-end self-improvement through reinforcement learning, optimized for efficient deployment on single GPUs and edge devices.

Warp introduces Warp Factories, an open, flexible infrastructure for building cloud software factories with features like code configuration, support for any model, evals, and self-improvement.

Ornith-1.5-35B-A3B is a new AI foundation model that achieves superior performance on coding and agentic benchmarks by employing end-to-end self-improvement, activating only about 3 billion parameters per token.

Ornith-1.5-35B-A3B is a mixture-of-experts AI model that activates only 3B parameters per token and outperforms similar-sized models like Qwen and Gemma in coding and agentic benchmarks.

Ornith-1.5-9B is a 9B dense AI model that advances foundation model building through end-to-end self-improvement, optimizing task generation, scaffold construction, and solution rollouts via reinforcement learning. It demonstrates competitive performance on various benchmarks compared to other models like Qwen and Gemma.

AutoMem is a text-gradient recursive self-improvement framework for automated memory architecture search in LLM agents, which discovers task-adaptive architectures that outperform human-designed baselines with improved accuracy and efficiency.

This paper proposes model-harness co-evolution as a fundamental principle for recursive self-improvement in AI agents, introducing HELIX, a source-traceable system that improves both execution and learning by generating structured training signals from verified trajectories.

Zetta introduces a closed-loop embodied harness that evolves runtime critics and recovery skills to govern physical execution in real-time, achieving state-of-the-art success on robotics benchmarks with significant inference speedup and self-evolution.

This tweet advises indie programmers not to focus solely on coding but to learn marketing, recommending the book 'Traction' and emphasizing the importance of letting others find you.

The essay applies the concept of activation energy from chemistry to explain initial barriers in contexts like physics, neuroscience, and personal relationships, emphasizing the importance of low energy for sustaining actions.

The paper introduces DIVE, a diversity-driven framework that enables frozen LLMs to self-improve by evolving persistent natural-language skills from task experience and verifier feedback, without parameter updates. It outperforms existing methods on math and logical reasoning tasks and transfers across model scales.

Introduces SBCO, a self-supervised verifier-grounded harness optimizer for planning agents that improves agent outputs via approximate block coordinate ascent, matching or exceeding self-modifying baselines with far less compute.

Naval Ravikant discusses how following your natural intellectual obsessions leads to success, citing examples like Bryan Johnson and Balaji Srinivasan.

Prime Agent is a coding agent that can refine its own harness, launched on Product Hunt.

This paper introduces Macaron-V1, an open continual learning agent-model family using Mixture-of-LoRA to compose specialist adapters on frozen base models, with recursive self-improvement and model-harness co-design.

The author proposes using AI as a personal coach, discovering cognitive blind spots and strengths through regular review of interactions with AI, to achieve continuous growth.

The paper introduces Hierarchical Self-Improvement (HSI), a framework that enhances frozen LLM agents by evolving task-specific harnesses through hierarchical self-modification, achieving substantial gains on moderate tasks while being limited by feedback quality and backbone capabilities.

Introduces a GitHub open-source project with 57k stars, byoungd/up, as a life advancement guide for ordinary people, covering English learning, hands-on AI tools, and personal retrospectives; it also includes information on the high-priced sale of the 2.ai domain portfolio.

The author shares a list of books that inspired him to earn 500k through apps and pursue freelancing, including The Almanack of Naval Ravikant, Soft Skills, Low-Risk Entrepreneurship, etc., emphasizing creating products with low marginal cost and passive income.

Saboo Shubham announced the open-source release of a template for building long-horizon agent harnesses that can dream and self-improve in the background.