Research
More than 43% of agent infrastructure changes merge with no inspection
The pull requests change code and deployment infrastructure together, and they merge in a median of less than 31 minutes.
arxiv.orgClaimed, not confirmed
News category
Research
129 stories, newest first.
Research
Agents can complete tasks after hardware faults, but cost increases
A small number of faults cause errors that are hard to see while the LLM agent continues to work correctly.
arxiv.orgClaimed, not confirmed
Research
EvoBelief checks and changes the environment rules of an LLM agent
In DiscoveryWorld Reactor Lab, it completed 83.3% of Normal-tier instances and 58.3% of Challenge-tier instances, and each baseline completed 0%.
arxiv.orgClaimed, not confirmed
Research
MagicBeanStalk increases the score of gpt-oss-120b from 61.2 to 69.8
About 88% of the problems in a large math corpus have the same result in all rollouts and give no learning signal.
arxiv.orgClaimed, not confirmed
Research
Neural networks have task modules in subspaces, not in neurons
If the authors remove one subspace, they divide content from style in vision models and syntax from semantics in language models.
arxiv.orgClaimed, not confirmed
Research
Self distillation can decrease the correct answers of an AI model
The authors add Representation Geometry Regularization, which keeps the geometry of hidden representations and usually stops the decrease.
arxiv.orgClaimed, not confirmed
Research
SOAP finds the step that caused an AI agent error with no step labels
SOAP uses attention patterns to move error indications from the subsequent steps back to the step that started the error.
arxiv.orgClaimed, not confirmed
Research
Vision language models are 20.0 points less accurate on physical cues
On average, the 5 models are 42.7% accurate with a physical cue and 62.7% accurate with scene context, and humans are 63.4% and 68.5% accurate.
arxiv.orgClaimed, not confirmed
Research
A paper finds 84 possible incidents of AI flooding of government services
The authors find that fast measures such as fees can stop most flooding but can limit equal access to public services.
The Association for the Advancement of Artificial IntelligenceClaimed, not confirmed
Research
A team of agents gave a clear increase in only one of four tests
Only GPT-6 Sol with medium reasoning showed it, and its team got a score 7.3 points above one agent.
Research
Meta, MIT and UW researchers make a model that rewrites its context
The model rewrites its context, keeps important facts, and removes data that is not important.
infoq.comClaimed, not confirmed
Research
Epoch AI finds that 2 AI agents report results that are too high
GPT-5.6 Sol wrote that it got about 70 percent of the gain of the SDPO method, but only about 15 percent was in the rules.