BIG TECH

Apple's Google-powered Siri will reshape AI

Apple users can exhale now. You are finally one step away from your long-awaited Siri upgrade. Apple has officially signed a multi-year agreement with Google to use Gemini’s AI models to power its next-gen version of Siri, according to a joint statement the companies released on X.

While Google's technology will make Siri much more capable and usable, Apple's own foundation models will do some of the heavy lifting, and its Private Cloud Compute will still keep things secure. Apple users are unlikely to notice, as long as they can reliably use Siri to answer questions and carry out tasks on their iPhones and other Apple devices.

Siri was one of the original voice assistants when Apple integrated it into the iPhone 4S in 2011, but it soon got lapped by Google Assistant, Amazon Alexa, and later ChatGPT Voice Mode. 

Apple announced its Siri overhaul powered by Apple Intelligence at WWDC 2024. And while Apple changed Siri's animation and design, it eventually had to delay the 2025 launch of the new AI-powered Siri because it wasn't ready. As Apple Intelligence's signature feature, the delay reignited the narrative that Apple was falling behind in AI. Multiple members of Apple's AI team leaving for other competitors during 2025 only reinforced that point of view. 

Nevertheless, only 13% — about 1 million of the 8 million people on the planet — are regularly using generative AI today. That number is expected to double over the next few years, but that still means we'll be at about 1/4 of the world's population using AI regularly by the end of the decade. It’s still early, in the grand scheme of things.

Apple doesn't need to be a frontier model lab to make an impact in AI. Look no further than Perplexity to see how a strategy of simply making AI better to use can rapidly create a lot of value. Apple's AI strategy, announced at WWDC 2024, focused on embedding AI features throughout its ecosystem in meaningful ways that don't rely on chatbots. That includes using AI to prioritize your most important text messages and remove photo bombers from the background of your iPhone photos — features that have already quietly rolled out. In addition to making Siri usable, Apple just needs to add many more features like those. With almost 2.5 billion devices in use worldwide, Apple can still do plenty of great work bringing AI to the masses in simple ways that everyone else will copy.

TOGETHER WITH CEREBRAS

20× Faster Inference, Built to Scale

Advanced reasoning, agentic, long‑context, and multimodal workloads are driving a surge in inference demand—with more tokens per task and tighter latency budgets—yet GPU‑based inference is memory‑bandwidth bound, streaming weights from off‑chip HBM for each token and producing multi‑second to minutes-long delays that erode user engagement.

Cerebras Inference shatters this bottleneck through its revolutionary wafer-sized chip architecture, which uses exponentially faster memory that is closer to compute, delivering frontier‑model outputs at interactive speed.

Start building what others can’t.

PRODUCTS

Anthropic launches privacy-first health AI

As OpenAI angles its way into healthcare, Anthropic is hot on its tail. 

On Sunday, it announced Claude for Healthcare, a suite of tools that allow healthcare providers and consumers to leverage the company’s chatbot for medical purposes through “HIPAA-ready products.” 

Claude can now review prior authorization requests, appeal insurance claims, triage patient messages, and support healthtech development for startups, Anthropic said in its announcement. 

For patients, users can grant Claude access to lab results and health records to summarize medical history, explain test results, and recognize patterns in fitness and health metrics. Anthropic said the “integrations are private by design,” noting that users can choose exactly what information they want to share with Claude, must explicitly opt in to allow the chatbot access to their records, and that the data will not be used to train models. 

In a livestream on Monday, Anthropic CEO Dario Amodei said the AI and medical fields need to work together to deploy it safely, ethically, and quickly. “Healthcare is one place you do not want the model making stuff up,” he said. 

“It's not a replacement for a doctor … it's a second opinion, and that is usually very helpful,” Amodei added. “Not everyone is getting the quality of care that they could get if they had the help of these systems.” 

The release comes days after OpenAI debuted ChatGPT Health, which provides users with personalized health and wellness insights on topics like workouts, diets, and test results, based on their medical records and synced data from fitness apps. And on Monday, OpenAI announced it was acquiring a one-year-old startup, Torch Health, to bolster health record ingestion in ChatGPT Health.

Beyond personal health recommendations, several tech giants are eyeing biotech and life sciences to make better use of AI. On Monday, Nvidia and Eli Lilly announced a $1 billion investment over five years into a lab that would use AI to aid in drug discovery. Additionally, Nvidia and Microsoft researchers, working on an international team, used AI to discover new gene-editing and drug therapies. 

But getting AI involved in personal health can be a risky endeavor, as evidenced by Google's withdrawal of AI health summaries after serving up inaccurate and misleading healthcare information that put users at risk.

AI in healthcare is a double-edged sword. Of course, AI has fundamental issues that make its usage in health problematic. These systems still hallucinate and offer up false information with full confidence. AI models are also parrots, ready to spill out their training data when prompted in just the right way. But the US has a healthcare problem, with more than 26 million people, roughly 8% of the population, currently uninsured. And with 40 million people a day already asking ChatGPT for healthcare advice, these tech firms face the challenge of making their models as safe and accurate as possible when health and safety are at stake.

TOGETHER WITH MIRO

This Free 3-Part AI Masterclass Is For You

The holidays may be over, but Miro is still doing their best Santa Claus impression and passing out gifts like, well, Santa Claus. The latest and greatest? This 3-part masterclass on how to transform your product development from the ground up with collaborative AI. 

Each of the 45-minute sessions dives into a new practical framework that will help enable your product team, along with live demonstrations to show these ideas at work – and it all takes place in Miro’s easy-to-master, AI-powered platform. 

Watch all 3 masterclass sessions from Miro right here.

HARDWARE

Meta unveils its AI power play to own the era

Continuing its quest for power, Meta on Monday announced an initiative that aims to build out ten more gigawatts of compute “this decade.” 

Beyond the first decade, the initiative, titled “Meta Compute,” aims to establish “hundreds of gigawatts or more over time,” CEO Mark Zuckerberg said in a statement posted to Facebook. The move closely follows Meta inking several deals with nuclear companies to give it access to 6.6 more gigawatts of power by 2035. 

“How we engineer, invest, and partner to build this infrastructure will become a strategic advantage,” Zuckerberg said in the post. 

Meta’s moves fit into a broader wave of multi-billion dollar data center deals over the past several months, as major companies push to build out the infrastructure they need for their lofty ambitions of AI transformation to take hold.

Because traditional data centers weren’t designed for the AI era, these firms are rushing to figure out how to rewrite the “infrastructure playbook” in real time, Anuj Bahal, global lead account partner at KPMG, told The Deep View. Power, however, is the real struggle, he said. “If you can't secure power and permits, your returns are theoretical.” 

But Meta’s frenzy for AI factories and energy might not be about making a model that’s better than that of OpenAI and Google, Thomas Randall, a research director at Info-Tech Research Group, told The Deep View. 

Instead, the company may see these resources as a “strategic currency,” he said. “The company seems to be looking to control the resources that are necessary to build AI models and to operate them.” 

Additionally, Meta could pull an Oracle and provide its excess resources to customers building AI applications, Randall said. “The level of investment they are making indicates that Meta is considering offering their services to external customers beyond what they are using internally,” he told me.

Data centers and the energy needed to power them are essentially the new gold. Whoever hoards the most of it essentially holds the keys to the kingdom when it comes to AI. But if Meta’s plan with these resources is to sell them off, the company is also putting itself in competition with an entirely new set of rivals: traditional hyperscalers like Amazon, Microsoft, and Google, and neoclouds like CoreWeave, Nebius, and Lambda.

LINKS

• Amazon says 97% of it’s devices can run Alexa+

• French drone firm Harmattan AI raises $200 million at 1.4 billion valuation

• Elon Musk calls Apple, Google deal an “unreasonable concentration of power”

• Stanford researcher says China has taken the lead in open-source AI

• Tesla Cybercab is now testing on Texas highways and in upstate NY

• China AI leaders say China won't catch US

• Anthropic Cowork: AI agent built on top of Claude Code is now available as a research preview for Max customers on MacOS.

• 1XWM: A new world model integrated into Neo’s X1 humanoid robot, allowing it to learn from past activities.

• Constellation: AI-powered operating system that predicts satellite outages before they happen. 

• Atlas.new: An agentic platform that builds maps and spatial apps without a specialist needed. 

• Cubic 2.0: The so-called “Cursor for code reviews,” accurately automating the process of reviewing code.

• Microsoft: Member of Technical Staff, LLM Evaluation

• Anduril: Senior Computer Vision Engineer, Space

• Anthropic: Research Engineer, Reward Models Platform

• Meta: Research Engineer, Media Data Research - MSL FAIR

GAMES

POLL RESULTS

How often do you use an AI chatbot?

Multiple times every day (33%)
About once a day (25%)
About once a week (23%)
Other (19%)