Long-running agentic workflows are becoming practical through better orchestration, token budgeting and self-evolving policies. Use cases range from trading and scientific discovery to automated hardware debugging.
An experiment with a 27 billion parameter LLM in 4-bit quantization on an RTX 3090, for 63 hours, does not solve Riemann but shows the endurance of self-hosted agentic pipelines. The value is not the theorem: it is process stability, memory managemen...
A framework autonomously revises the textual policy of an LLM trading agent using decision traces and portfolio feedback, keeping the model fixed. Tests frequently show improved Sharpe Ratio and cumulative return, with analyses linking allocation cha...
A researcher let a 27B Qwen 3.8 model in 4-bit quantization with a 100K-token context window run for 63 hours on a single RTX 3090. It generated over 50 million tokens without solving the Riemann hypothesis, but showed organized memory, spontaneous c...
A report suggests an enthusiast used a bot built on GPT-6 Astra and Python to repeatedly beat Balatro's highest difficulty, Gold Stake with the Black Deck. Technical details are scarce, but the case prompts reflection on LLM numerical reasoning limit...
Ant International has listed 10 Alipay+ wallets and seven acquirers for Phase I of its Agentic Mobile Protocol in 2026. No partner has given a go-live date and Ant has not disclosed transaction volumes since April. The stated commitments focus on ide...
BudgetBench proposes a protocol and harness for comparing memory strategies in local LLM agents, treating the per-call token budget as the independent variable. Pilot tests on qwen2.5:1.5b, Qwen3 30B-A3B, and LongMemEval expose budget violations, non...
A multi-agent framework generates QUBO formulations from natural-language descriptions and uses test cases to correct them. On QUBOBench, 100 problems across 12 domains, it reaches 68% accuracy, beating a single-call baseline by 22%. The result shift...
A public dataset records 558 complete AI scientific agent trajectories with thoughts, tool calls, errors, and confidence. Traditional benchmarks look only at final outputs; here comparable frontier models reveal different error profiles. Claude Opus ...
Nvidia has released SoL-Pi, a standalone extension for the Pi framework. It adds four opt-in mechanisms, disabled by default, born from auto-research loops: they reduce repeated model turns, context replay, oversized observations, and unnecessary log...
An experiment with an old MacBook shows an AI agent observing the screen through a webcam and mirror, writing AMD drivers and checking its own progress in real time. The case signals an 'agent-first' approach to driver development and local debugging...
