AIMode.newsSearch
Live

Section

Technology

Gadgets, platforms, software, and the companies building them.

Advertisement
Technology

421 bugs in Microsoft's Patch Tuesday release, and the Norks have already attacked one

Related coverage: 421 bugs in Microsoft's Patch Tuesday release, and the Norks have already attacked one The Register Shattering the Dream – When a Job Offer Becomes a Zero-Day Attack Check Point Research Microsoft Plugs Nearly 400 Security Holes Krebs on Security Windows 11 Gets Dynamic File Sizes in File Explorer, Low Latency Profile for Apps, and More TechPowerUp Microsoft's Patch Tuesday Deluge Continues With August Updates Dark Reading

The Register · 16 hours ago

Technology

An FCC filing points to new Sonos headphones coming soon

It looks like we'll be getting a successor to the Sonos Ace headphones sometime this fall, which was hinted at by CEO Tom Conrad during a third quarter earnings call. As reported by What Hi-Fi?, an FCC filing by Sonos, Inc reveals details about upcoming wireless headphones with the model number S52A. In addition to the S52A designation, one of the submitted documents shows an oval diagram - likely the ear cup - marked with the name Sonos Ace Ultra, although that name isn't listed anywhere else. It follows the naming scheme Sonos has used in the past with its Arc and Arc Ultra soundbars. The documents also list color options we'll likely see … Read the full story at The Verge.

The Verge · 16 hours ago

Technology

Emergent Introspective Awareness in Large Language Models

We investigate whether large language models can introspect on their internal states. It is difficult to answer this question through conversation alone, as genuine introspection cannot be distinguished from confabulations. Here, we address this challenge by injecting representations of known concepts into a model's activations, and measuring the influence of these manipulations on the model's self-reported states. We find that models can, in certain scenarios, notice the presence of injected concepts and accurately identify them. Models demonstrate some ability to recall prior internal representations and distinguish them from raw text inputs. Strikingly, we find that some models can use their ability to recall prior intentions in order to distinguish their own outputs from artificial prefills. In all these experiments, Claude Opus 4 and 4.1, the most capable models we tested, generally demonstrate the greatest introspective awareness; however, trends across models are complex and sensitive to post-training strategies. Finally, we explore whether models can explicitly control their internal representations, finding that models can modulate their activations when instructed or incentivized to "think about" a concept. Overall, our results indicate that current language models possess some functional introspective awareness of their own internal states. We stress that in today's models, this capacity is highly unreliable and context-dependent; however, it may continue to develop with further improvements to model capabilities.

Hacker News · 16 hours ago