In partnership with
☀️ TRENDING AI NEWS 🤖 Gemini 3.8 Live: Google launched a Live Avatar for Gemini Enterprise that lip-syncs across 97 languages in real time. 🛠️ Call for Me: Pixel 11 owners can now let Gemini call businesses on their behalf without ever picking up the phone. 🏢 Lovable Revenue: Vibe-coding platform Lovable crossed $600M in annualized revenue as apps on its platform hit nearly 1 billion monthly views. 🚨 Pentagon Polygraph+: The US Department of Defense has requested $30.3M over five years to build an AI-powered lie detector system. |
Something quietly shifted this week - and it has nothing to do with a new model dropping. It's the way AI is reaching into the physical world: scheduling your reservations, calling your dentist, watching your face for micro-expressions. Welcome to Saturday.
🤓 AI Trivia
The US government's new AI-assisted lie detector program is called 'Polygraph+'. But roughly how long has the US military been using traditional polygraph tests?
The answer is hiding near the bottom of today's newsletter... keep scrolling. 👇
| 🤖 Google Gives Gemini a Face - and a Phone | |
Two new Gemini features land in the same week
Google has been busy. This week brought not one but two major Gemini upgrades that push the assistant from a text box into something that feels a lot more like a presence.
First up: Gemini 3.8 Live now comes with a "Live Avatar" - an animated AI persona that lip-syncs in real time during conversations and shows different facial expressions as it responds. It can switch between 97 supported languages without losing continuity. Right now it's gated to Gemini Enterprise customers, but if history is any guide, it won't stay there long.
Gemini Dials In - So You Don't Have To
The second feature, called "Call for Me", is exclusive to Pixel 11 in the US and requires a Gemini subscription. Instead of waiting on hold, you tell Gemini what you need - a restaurant reservation, a stock check, a rescheduled appointment - and it makes the call on your behalf. You don't even need to start the call yourself.
Google is billing this as an "early experiment", but the implications are real. AI agents that can take action in the physical world - placing actual phone calls to actual businesses - represent a meaningful step beyond chatbots.
If you've been following our coverage of AI agents, this is exactly the kind of real-world action-taking capability that makes them genuinely useful - and genuinely risky.
The bottom line
Google is turning Gemini from an assistant you chat with into an agent that acts for you - and these two features together show exactly what that looks like in practice.
Some teams never seem to stop moving. They're on Attio, the agentic CRM.
Every customer signal is captured in one shared context layer, always current and compounding. Agents and workflows build pipeline, chase every buying signal, and move deals forward, an always-on revenue engine running alongside your team.
With Attio, you’ll get:
Leads automatically prioritised and routed to the right rep
Expansion and risk signals caught the moment they land
Follow-ups written in your voice, already there when you arrive
Teams like Parallel, Turbopuffer, and Wordsmith build on Attio. Are you one of them?
| ⚠️ The Pentagon Wants AI to Catch You Lying | |
$30 million for a next-generation polygraph
The US Department of Defense has submitted a budget request for $30.3 million over five years to develop what it's calling "Polygraph+" or "Polygraph Next". The program aims to build AI and machine learning scoring algorithms layered on top of traditional polygraph data - plus a technique called "standoff sensing", which can detect physiological signals without physically wiring someone up.
Traditional polygraphs measure things like heart rate and skin conductance to infer deception. The AI layer would theoretically make the scoring more consistent and the results harder to game by removing human subjectivity from the analysis.
The credibility problem isn't going away
Here's the catch: the scientific community has long debated whether polygraphs actually work. There's no settled consensus that any physiological signal reliably maps to deception. Layering AI on top of a contested methodology doesn't necessarily fix that problem - it can just make flawed conclusions look more authoritative.
The "standoff sensing" piece is particularly worth watching. If the technology can read physiological signals from a distance - without the subject's knowledge - the data privacy and AI ethics implications run well beyond the intelligence community.
The bottom line
AI-powered lie detection is coming whether the science supports it or not - and the government is willing to spend $30M to find out if it works.
| 🚀 Vibe Coding Is Now a $600M Business | |
Lovable's numbers are hard to argue with
Lovable - the AI-powered app builder that lets you build web apps by describing what you want - has crossed $600M in annualized revenue. That's not a projected figure or a funding round valuation. That's actual revenue run rate. Co-founder Fabian Hedin says apps created on the platform are getting close to 1 billion monthly views.
"Vibe coding" - the term for using AI to generate working software through natural language prompts - has gone from a Twitter meme to a legitimate industry segment in under two years. Lovable is the clearest proof point that non-technical founders are not just experimenting with this: they're building real products with real audiences.
The market that wasn't supposed to exist yet
The conventional wisdom was that AI coding tools would first win over developers, then maybe trickle down to non-technical users. Lovable inverted that entirely. Its biggest user base is founders and product people who never wrote a line of code. That's a different product category than GitHub Copilot, and clearly a massive one.
Speaking of building fast - if you're looking to spin up a landing page or product site without the overhead, 60sec.site is an AI website builder that does exactly what the name says. Worth a look if you want something live before the weekend's over.
The bottom line
Vibe coding crossed a threshold this week - at $600M ARR, it's not a trend anymore. It's a market.
| 🛠️ PrismML Puts Tiny LLMs Inside Smart Glasses | |
On-device AI finds a new home on your face
PrismML is bringing its compact language models to Qualcomm-powered smart glasses. The startup's pitch is straightforward: open-weight AI that runs entirely on the device, using the computing power the hardware already has, without routing anything to the cloud.
Smart glasses are having a moment right now. Meta's Ray-Ban lineup is selling well, Google is experimenting in the space, and Qualcomm is positioning its chips as the go-to silicon for wearable AI. PrismML is betting that the next logical step isn't bigger cloud models - it's smaller, faster, local ones that work when you're offline, preserve your privacy, and don't add latency.
For the AI hardware crowd: this is part of a broader push to move inference to the edge. We covered Qualcomm's ambitions to run 30B parameter models on phones earlier this week - PrismML is the application layer that story implied was coming.
The bottom line
On-device AI is moving from phones to glasses - and PrismML is trying to own that transition before the bigger players get there.
| 🚀 Google Is Launching an AI Satellite Next Week | |
Tensor Processing Units go into orbit on a Falcon 9
Google is preparing to launch a satellite equipped with its Tensor Processing Units (TPUs) aboard a SpaceX Falcon 9 rocket. The mission is part of Project Suncatcher - Google's experimental initiative to test whether AI data centers can eventually operate in orbit.
The satellite will test how well Google's AI processors perform in the radiation and temperature extremes of space. If the hardware holds up, the longer-term vision is putting AI compute into orbit - closer to satellite-based sensors, remote infrastructure, and anywhere terrestrial connectivity is limited.
It's an early-stage experiment, and Google is careful to frame it that way. But the combination of AI infrastructure and orbital compute is genuinely novel - and it positions Google at the intersection of two extremely expensive bets: AI and space.
The bottom line
Google is testing whether AI chips can survive space - and if they can, it changes where AI compute can actually live.
| 🌎 Trivia Reveal | |
The answer is over 100 years! The US military began using polygraph examinations in the early 1920s, making it one of the longest-running contested tools in national security history. The DoD has spent a century refining a technique the scientific community still debates - now they want to add AI to the mix.
| 💬 Quick Question | |
Google's new Call for Me feature lets Gemini phone businesses on your behalf. Would you actually use that - or does handing phone calls to an AI feel like a step too far? Hit reply and let me know, I read every response!
That's all for today - catch you Monday with more from the world of AI. And as always, the full archive is waiting for you at dailyinference.com if you want to catch up on anything you missed this week.
| 🎁 Share Daily Inference | |
Know someone who would get value from this? Share your personal link below. One referral gets you the AI Tools Starter Kit, and five unlocks the AI Insider Briefing.