MAROKO133 Update ai: OpenAI experiment finds that sparse models could give AI builders the

📌 MAROKO133 Update ai: OpenAI experiment finds that sparse models could give AI bu

OpenAI researchers are experimenting with a new approach to designing neural networks, with the aim of making AI models easier to understand, debug, and govern. Sparse models can provide enterprises with a better understanding of how these models make decisions. 

Understanding how models choose to respond, a big selling point of reasoning models for enterprises, can provide a level of trust for organizations when they turn to AI models for insights. 

The method called for OpenAI scientists and researchers to look at and evaluate models not by analyzing post-training performance, but by adding interpretability or understanding through sparse circuits.

OpenAI notes that much of the opacity of AI models stems from how most models are designed, so to gain a better understanding of model behavior, they must create workarounds. 

“Neural networks power today’s most capable AI systems, but they remain difficult to understand,” OpenAI wrote in a blog post. “We don’t write these models with explicit step-by-step instructions. Instead, they learn by adjusting billions of internal connections or weights until they master a task. We design the rules of training, but not the specific behaviors that emerge, and the result is a dense web of connections that no human can easily decipher.”

To enhance the interpretability of the mix, OpenAI examined an architecture that trains untangled neural networks, making them simpler to understand. The team trained language models with a similar architecture to existing models, such as GPT-2, using the same training schema. 

The result: improved interpretability. 

The path toward interpretability

Understanding how models work, giving us insight into how they're making their determinations, is important because these have a real-world impact, OpenAI says.  

The company defines interpretability as “methods that help us understand why a model produced a given output.” There are several ways to achieve interpretability: chain-of-thought interpretability, which reasoning models often leverage, and mechanistic interpretability, which involves reverse-engineering a model’s mathematical structure.

OpenAI focused on improving mechanistic interpretability, which it said “has so far been less immediately useful, but in principle, could offer a more complete explanation of the model’s behavior.”

“By seeking to explain model behavior at the most granular level, mechanistic interpretability can make fewer assumptions and give us more confidence. But the path from low-level details to explanations of complex behaviors is much longer and more difficult,” according to OpenAI. 

Better interpretability allows for better oversight and gives early warning signs if the model’s behavior no longer aligns with policy. 

OpenAI noted that improving mechanistic interpretability “is a very ambitious bet,” but research on sparse networks has improved this. 

How to untangle a model 

To untangle the mess of connections a model makes, OpenAI first cut most of these connections. Since transformer models like GPT-2 have thousands of connections, the team had to “zero out” these circuits. Each will only talk to a select number, so the connections become more orderly.

Next, the team ran “circuit tracing” on tasks to create groupings of interpretable circuits. The last task involved pruning the model “to obtain the smallest circuit which achieves a target loss on the target distribution,” according to OpenAI. It targeted a loss of 0.15 to isolate the exact nodes and weights responsible for behaviors. 

“We show that pruning our weight-sparse models yields roughly 16-fold smaller circuits on our tasks than pruning dense models of comparable pretraining loss. We are also able to construct arbitrarily accurate circuits at the cost of more edges. This shows that circuits for simple behaviors are substantially more disentangled and localizable in weight-sparse models than dense models,” the report said. 

Small models become easier to train

Although OpenAI managed to create sparse models that are easier to understand, these remain significantly smaller than most foundation models used by enterprises. Enterprises increasingly use small models, but frontier models, such as its flagship GPT-5.1, will still benefit from improved interpretability down the line. 

Other model developers also aim to understand how their AI models think. Anthropic, which has been researching interpretability for some time, recently revealed that it had “hacked” Claude’s brain — and Claude noticed. Meta also is working to find out how reasoning models make their decisions. 

As more enterprises turn to AI models to help make consequential decisions for their business, and eventually customers, research into understanding how models think would give the clarity many organizations need to trust models more. 

🔗 Sumber: venturebeat.com

📌 MAROKO133 Update ai: Researchers from PSU and Duke introduce “Multi-Agent System

Share My Research is Synced’s column that welcomes scholars to share their own research breakthroughs with over 2M global AI enthusiasts. Beyond technological advances, Share My Research also calls for interesting stories behind the research and exciting research ideas. 

Meet the author
Institutions: Penn State University, Duke University, Google DeepMind, University of Washington, Meta, Nanyang Technological University, and Oregon State University. The co-first authors are Shaokun Zhang of Penn State University and Ming Yin of Duke University.

In recent years, LLM Multi-Agent systems have garnered widespread attention for their collaborative approach to solving complex problems. However, it’s a common scenario for these systems to fail at a task despite a flurry of activity. This leaves developers with a critical question: which agent, at what point, was responsible for the failure? Sifting through vast interaction logs to pinpoint the root cause feels like finding a needle in a haystack—a time-consuming and labor-intensive effort.
 
This is a familiar frustration for developers. In increasingly complex Multi-Agent systems, failures are not only common but also incredibly difficult to diagnose due to the autonomous nature of agent collaboration and long information chains. Without a way to quickly identify the source of a failure, system iteration and optimization grind to a halt.
 
To address this challenge, researchers from Penn State University and Duke University, in collaboration with institutions including Google DeepMind, have introduced the novel research problem of “Automated Failure Attribution.” They have constructed the first benchmark dataset for this task, Who&When, and have developed and evaluated several automated attribution methods. This work not only highlights the complexity of the task but also paves a new path toward enhancing the reliability of LLM Multi-Agent systems.
The paper has been accepted as a Spotlight presentation at the top-tier machine learning conference, ICML 2025, and the code and dataset are now fully open-source.

Paper：https://arxiv.org/pdf/2505.00212
Code：https://github.com/mingyin1/Agents_Failure_Attribution
Dataset：https://huggingface.co/datasets/Kevin355/Who_and_When
 
 
Research Background and Challenges
LLM-driven Multi-Agent systems have demonstrated immense potential across many domains. However, these systems are fragile; errors by a single agent, misunderstandings between agents, or mistakes in information transmission can lead to the failure of the entire task.

Currently, when a system fails, developers are often left with manual and inefficient methods for debugging:
Manual Log Archaeology : Developers must manually review lengthy interaction logs to find the source of the problem.
Reliance on Expertise : The debugging process is highly dependent on the developer’s deep understanding of the system and the task at hand.
 
This “needle in a haystack” approach to debugging is not only inefficient but also severely hinders rapid system iteration and the improvement of system reliability. There is an urgent need for an automated, systematic method to pinpoint the cause of failures, effectively bridging the gap between “evaluation results” and “system improvement.”

Core Contributions
This paper makes several groundbreaking contributions to address the challenges above:
1. Defining a New Problem: The paper is the first to formalize “automated failure attribution” as a specific research task. This task is defined by identifying the

2. failure-responsible agent and the decisive error step that led to the task’s failure.

Constructing the First Benchmark Dataset: Who&When : This dataset includes a wide range of failure logs collected from 127 LLM Multi-Agent systems, which were either algorithmically generated or hand-crafted by experts to ensure realism and diversity. Each failure log is accompanied by fine-grained human annotations for:
Who: The agent responsible for the failure.
When: The specific interaction step where the decisive error occurred.
Why: A natural language explanation of the cause of the failure.

3. Exploring Initial “Automated Attribution” Methods : Using the Who&When dataset, the paper designs and assesses three distinct methods for automated failure attribution:
All-at-Once: This method provides the LLM with the user query and the complete failure log, asking it to identify the responsible agent and the decisive error step in a single pass. While cost-effective, it may struggle to pinpoint precise errors in long contexts.
Step-by-Step: This approach mimics manual debugging by having the LLM review the interaction log sequentially, making a judgment at each step until the error is found. It is more precise at locating the error step but incurs higher costs and risks accumulating errors.
Binary Search: A compromise between the first two methods, this strategy repeatedly divides the log in half, using the LLM to determine which segment contains the error. It then recursively searches the identified segment, offering a balance of cost and performance.
 
Experimental Results and Key Findings

Experiments were conducted in two settings: one where the LLM knows the ground truth answer to the problem the Multi-Agent system is trying to solve (With Ground Truth) and one where it does not (Without Ground Truth). The primary model used was GPT-4o, though other models were also tested. The systematic evaluation of these methods on the Who&When dataset yielded several important insights:

• A Long Way to Go: Current methods are far from perfect. Even the best-performing single method achieved an accuracy of only about 53.5% in identifying the responsible agent and a mere 14.2% in pinpointing the exact error step. Some methods performed even worse than random guessing, underscoring the difficulty of the task.
• No “All-in-One” Solution: Different methods excel at different aspects of the problem. The All-at-Once method is better at identifying “Who,” while the Step-by-Step method is more effective at determining “When.” The Binary Search method provides a middle-ground performance.

• Hybrid Approaches Show Promise but at a High Cost: The researchers found that combining different methods, such as using the All-at-Once approach to identify a potential agent and then applying the Step-by-Step method to find the error, can improve overall performance. However, this comes with a significant increase in computational cost.

• State-of-the-Art Models Struggle: Surprisingly, even the most advanced reasoning m…

  Konten dipersingkat otomatis.

  🔗 Sumber: syncedreview.com

  ---