AI News
Latest news and trends from the world of artificial intelligence
Inside Celosphere 2025: Why there’s no ‘enterprise AI’ without process intelligence
AI adoption is accelerating, but results often lag expectations. Celosphere 2025 will tackle the AI ROI challenge head-on by highlighting enhancements to the Celonis Process Intelligence (PI) Platform that help enterprises harness ‘enterprise AI,’ powered by PI, to continuously improve operations, creating measurable business value at scale. The event will spotlight how global enterprises are building “future-fit” operations and will also focus on agentic AI, and navigating tariffs and supply chain shocks. Celonis’ edge lies in treating process intelligence not as an add-on, but as the foundation of the enterprise stack.
How LeapXpert uses AI to bring order and oversight to business messaging
LeapXpert uses AI to bring order and oversight to business messaging.
How Lumana is redefining AI’s role in video surveillance
For all the progress in artificial intelligence, most video security systems still fail at recognising context in real-world conditions. Lumana, an AI video surveillance company, believes the fault in these systems lies deep in the foundations of how they are built. Instead of layering AI on top of old video security frameworks, Lumana rebuilt the infrastructure itself with an all-in-one platform that combines modern video security hardware, software, and proprietary AI. The company’s design also prioritises privacy. Lumana’s next stage of development aims to move from detection and understanding to predicting.
Google airs AI-generated ad without AI label, citing viewer apathy
Google has launched a new ad for its AI search, made entirely with its AI video tool Veo 3, but without disclosing the use of AI. The spot airs on TV from today and expands to cinemas and online media on Saturday. To avoid criticism of fake-looking people, the video uses stylized, toy-like characters. Robert Wong from Google Creative Lab said most viewers don’t care if AI was involved. Google treats AI like any other creative tool, such as Photoshop. A Christmas version is already planned.
Google leans on token metrics, not revenue, adding to bubble talk about AI growth
Google earns solid revenues from its cloud services and AI chips, but it's still unclear if generative AI has produced any significant income so far. Instead of disclosing concrete revenue numbers, Google uses the token consumption of its largest clients as a performance measure, though this figure doesn't offer much insight into actual use, costs, or business impact. Current estimates suggest generative AI brings in $3 to $4 million per major customer each year, which is only a small portion of Google's overall cloud revenue.
Accidental AI Safety experiment by PewDiePie
Accidental AI Safety experiment by PewDiePie: He created his own self-hosted council of 8 AIs to answer questions. They voted and picked the best answer. He noticed they were always picking the same two AIs, so he discarded the others, made the process of discarding/replacing automatic, and told the AIs about it. The AIs started talking about this "sick game" and scheming to prevent that. This is the video with the timestamp:
Bill Gates calls AI "the biggest technical thing ever in my lifetime," but warns of a bubble
Bill Gates calls AI "the biggest technical thing ever in my lifetime," but warns of a bubble
Meta researchers open the LLM black box to repair flawed AI reasoning
Researchers at Meta have developed a new technique, Circuit-based Reasoning Verification (CRV), that can predict the correctness of an LLM's reasoning and intervene to fix mistakes by monitoring its internal "reasoning circuits."
Why Did MiniMax M2 End Up as a Full Attention Model?
After M2 release, we’ve been receiving many queries from the community on “Why did you turn back the clock and go with full attention with MiniMax M2?” This article provides a detailed explanation of the trade-offs and challenges in choosing between full and efficient attention mechanisms for LLMs, with a focus on MiniMax M2. It discusses the limitations of efficient attention, the importance of evaluation, real bottlenecks, and future directions. The post also includes an addendum about SWA code.
Emergent Introspective Awareness in Large Language Models
We investigate whether large language models can introspect on their internal 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. Models demonstrate some ability to recall prior intentions in order to distinguish their own outputs from artificial prefills. 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.
Archive
About Sources
AI news are automatically downloaded from various sources and translated using AI. Updates occur twice a day.