☀️ TRENDING AI NEWS
🤖 Grok barely registers in US federal AI records despite Musk's government ties
🚨 AI used to reconstruct dead pilots' voices from cockpit recordings - NTSB temporarily blocks docket access
🏢 VCs and founders are using inflated ARR figures to crown AI startups - and investors know it
🛠️ Anthropic's Code with Claude event shows a future where most code is AI-written
Something about the Grok story feels almost poetic. The chatbot that Elon Musk built partly to influence government is - according to a new Reuters report - almost completely absent from actual government AI usage records. While xAI is burning through billions and Musk is as plugged into Washington as anyone alive, the product itself simply isn't being used. That tension between narrative and reality runs through a lot of today's stories.
🤓 AI Trivia
Anthropic's Code with Claude event took place alongside another major AI event this week. Which conference was running at the same time, on the same day?
💻 Microsoft Build
🔬 AWS re:Invent
🌐 Google I/O
🤖 Meta Connect
The answer is hiding near the bottom of today's newsletter... keep scrolling. 👇
🤖 Grok Is Missing From the Government It Was Supposed to Influence
A new Reuters report found that Grok barely appears in federal records of how the US government used AI last year. That's a striking finding given that Elon Musk has been at the center of the current administration and xAI has raised billions pitching Grok as a serious enterprise product.
Usage Numbers Don't Match the Hype
The Verge's analysis points out this isn't the only sign of trouble. Despite Musk positioning xAI at the heart of what could be the largest IPO in American history via SpaceX, Grok's actual adoption metrics remain thin. Federal AI usage tends to skew toward OpenAI and Google products - and that gap appears to be significant.
The broader takeaway: there's a real difference between having political proximity and building a product people actually choose to use. Grok may be getting pushed hard in some circles, but the usage data tells a different story.
🚨 AI Is Resurrecting the Voices of Dead Pilots
This one is genuinely unsettling. People have been using AI on spectrogram images of cockpit voice recordings to reconstruct the audio of pilots who died in crashes - effectively bringing back their voices from accident investigations. The situation got serious enough that the National Transportation Safety Board (NTSB) temporarily blocked access to its docket system to prevent further reconstruction.
Spectrograms to Speech - A New Frontier in Audio AI
The technique works by converting a visual representation of sound - a spectrogram - back into audio using AI models. What was once a one-way transformation is now reversible with enough compute. This has obvious implications beyond aviation: any archived spectrogram could potentially become reconstructable audio.
The ethics here are layered. These are recordings from fatal accidents, involving people who never consented to having their voices reconstructed. It's a direct collision between what audio AI is technically capable of and what should arguably be off-limits. Expect this to become a live regulatory debate fast.
🏢 How VCs Are Inflating ARR to Crown AI Darlings
If you've been watching AI startup valuations with raised eyebrows, this TechCrunch investigation gives you the receipts. Some AI companies are stretching traditional Annual Recurring Revenue metrics in ways that would make a traditional SaaS CFO wince - and their venture capital backers are fully aware.
ARR as a Marketing Tool, Not a Metric
The report details how some founders are counting committed spend, letters of intent, and projected usage as "ARR" - figures that wouldn't pass muster under traditional definitions. The motivation is clear: in a market where headline revenue numbers drive the next funding round, there's enormous pressure to show growth curves that justify eye-watering valuations.
What makes this particularly interesting is that investors know it's happening and are playing along. It's not fraud exactly - it's a collective agreement to use a shared language that everyone knows is stretched. The risk is when those numbers get treated as real by downstream investors who haven't done the work.
Speaking of building fast - if you're launching an AI project or startup and need a site up quickly, 60sec.site is an AI website builder that gets you from idea to live site in under a minute. Worth bookmarking.
🛠️ Anthropic's Vision for a World Where AI Writes Almost All Code
MIT Technology Review got a front-row seat at Anthropic's Code with Claude event in London - a two-day developer event that kicked off on May 19, the same day as Google I/O in Palo Alto. The atmosphere apparently had strong vibes, and the central message was clear: the future of software development involves AI writing most of the code.
Pull Requests Written Entirely by Claude
Attendees were asked how many had shipped a pull request in the last week that was completely written by AI. The hands that went up tell you where this is heading. Claude's coding capabilities have been a major focus for Anthropic, and the event was designed to show developers what building in an AI-first workflow actually looks like day to day.
If you want to go deeper on the coding AI landscape, we've been tracking this closely at Daily Inference - including our recent coverage on OpenAI's math problem breakthrough which showed what frontier reasoning models can now do beyond just code.
⚠️ London Blocks Palantir's £50M Met Police AI Deal
London Mayor Sadiq Khan has blocked a £50 million contract between the Metropolitan Police and Palantir, the controversial US data analytics company. The deal would have used Palantir's AI technology to automate intelligence analysis in criminal investigations. Palantir has fired back, accusing Khan of "putting politics above public safety."
Scotland Yard vs. City Hall
The Met itself has criticized the decision as "disappointing" and warned it could hit frontline policing effectiveness. The tension here reflects a wider debate about public sector AI procurement - specifically whether the efficiency gains from AI surveillance tools outweigh the civil liberties concerns attached to who builds them.
Palantir has a long history of controversy around its government contracts - from US immigration enforcement to military applications. For Khan, blocking the deal is a political statement as much as a procurement decision. For the Met, it means going back to square one on a capability they clearly wanted.
🔬 DeepMind's CEO Says We're Standing at the Foothills of the Singularity
During Google I/O this week, Google DeepMind CEO Demis Hassabis made a striking claim: we are currently "standing in the foothills of the singularity." MIT Technology Review unpacked what that actually means for AI-driven science - and it's more substantive than it might first sound.
From AlphaFold to Autonomous Scientific Discovery
The piece traces a clear shift in how AI is being applied to science. It's no longer just accelerating existing workflows - tools like AlphaFold have already changed structural biology permanently, and DeepMind is pushing toward systems that can generate and test scientific hypotheses with minimal human input. The path Hassabis is describing is one where AI doesn't just assist researchers - it starts to become the researcher.
Whether you find that thrilling or alarming probably says a lot about your overall relationship with this technology right now.
🌎 Trivia Reveal
The answer is Google I/O! Anthropic's Code with Claude event in London kicked off on May 19 - the exact same day as Google's developer conference in Palo Alto. Anthropic staffers were quick to tell MIT Tech Review it was a coincidence, not a flex. Sure it was.
💬 Quick Question
The Grok story today raises an interesting question about the gap between narrative and adoption. So here's mine for you: which AI chatbot or tool do you actually use week-to-week - not the one you feel like you should be using, but the one you genuinely reach for? Hit reply and tell me - I read every single response and I'm genuinely curious whether usage patterns in this community match the mainstream headlines.
That's all for today - see you tomorrow with more. If you want to catch up on anything you missed this week, the full archive is right here. And if someone forwarded this your way, you can subscribe at dailyinference.com.