Apple Overhauls Siri with "Privacy" in the Starring Role
Apple is working on an overhaul of Siri, and privacy looks set to be its central theme. According to TechCrunch, the new Siri may include an automatic conversation deletion feature.
This appears to be Apple's answer to what has long been considered Siri's weak point: how conversation data is handled. While Amazon's Alexa and Google Assistant store and leverage conversations in the cloud, Apple seems poised to differentiate itself with a "we don't keep your conversations" approach.
Details will have to wait for an official announcement at WWDC and other events, but a privacy-first strategy should sharpen Apple's distinct position in the AI assistant market.
Microsoft AI CEO: "Human-Level AI Within 18 Months"
The CEO of Microsoft AI has predicted that human-level AI — AI on par with humans — will arrive within 18 months.
The prediction has, unsurprisingly, sparked debate. The very definition of "human-level" is fuzzy: AI already surpasses humans at specific tasks, but general human-level intelligence remains unachieved. Even within the AI research community, opinions are split on whether an 18-month timeline is realistic.
That said, given Microsoft's close partnership with OpenAI and the massive investments it is pouring into next-generation model development, the company's top executive making such a prediction can be read as a strategic message rather than mere optimism.
Citadel's AI Agents Complete PhD-Level Work in Days
Ken Griffin, CEO of major hedge fund Citadel, has said that the firm's AI agents are capable of completing PhD-level research tasks in just a few days.
AI adoption in finance is advancing rapidly, and Griffin's remarks hint at its leading edge. AI agents reportedly power through sophisticated analytical work — the kind that would previously take PhD holders weeks or even months — in a remarkably short time.
If performance at this level is real, the ripple effects could be significant for research-intensive industries beyond finance. On the other hand, the specific bar for what counts as "PhD-level" is unclear, and in practice it may only apply to a narrow set of tasks. For now, it's probably fair to take this simply as evidence that AI adoption in finance has come a long way.
Ring Device Interprets Sign Language — IEEE Spectrum Reports
IEEE Spectrum has reported on technology that interprets sign language using a ring worn on the finger.
Camera-based approaches have long dominated sign language recognition, but because a ring-shaped device senses finger and hand movement directly, it could enable a more accurate and privacy-conscious interface. The technology holds promise for bridging the communication gap between deaf and hearing people.
As the applications of wearable AI continue to expand, this stands out as a practical approach in the accessibility space.
M5 Mac vs DGX Spark: Detailed Local LLM Benchmarks Published
A detailed local LLM benchmark covering the M5 Mac, NVIDIA DGX Spark, AMD Strix Halo, and RTX 6000 has been published on the Reddit LocalLLaMA community.
The tests ran in parallel over three days. Here are the key takeaways:
Memory bandwidth is key**: The gap — roughly 1,800GB/s for the RTX 6000, about 600GB/s for the M5, and around 256GB/s for both the DGX Spark and Strix Halo — translated almost directly into tokens-per-second performance
The M5 Mac far outperforms the DGX Spark on price-performance**: With more than twice the memory bandwidth, the M5 also holds the edge in inference speed
MacBook thermal management is surprisingly good**: Temperatures stayed stable at around 80°C even through three days of continuous operation, though the fans reportedly get quite loud
The AMD EVO X2 struggles with thermal throttling during extended runs**
The benchmark data is [available on GitHub](https://github.com/Light-Heart-Labs/MMBT-Messy-Model-Bench-Tests/tree/main/hardware-tests) and looks like a useful reference for choosing hardware for local LLMs.
---