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
What matters in AI.
SubscribeNews category
377 stories, newest first.
Research
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
The authors add Representation Geometry Regularization, which keeps the geometry of hidden representations and usually stops the decrease.
arxiv.orgClaimed, not confirmed
Research
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
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
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
Only GPT-6 Sol with medium reasoning showed it, and its team got a score 7.3 points above one agent.
Research
The model rewrites its context, keeps important facts, and removes data that is not important.
infoq.comClaimed, not confirmed
Research
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.
Research
The users made more errors or stopped when the team removed the tool, but the group with no AI continued.
Research
7 steps before the decision, the detector had 90.4% AUROC, but only in controlled tasks.
International Business Times, Singapore EditionClaimed, not confirmed
Research
In a Teradata survey of 1,000 leaders, 78% find it hard to make a data foundation for AI agents.
Research
In InfiLoop, accuracy continues to increase after more than 20,000 effective steps.
arxiv.orgClaimed, not confirmed