Palantir CEO Says Silicon Valley Got AI Totally Wrong

Palantir CEO Alex Karp told the company’s annual AI conference that Silicon Valley “totally effed up” by overselling large language models. He said the real power comes when LLMs are paired with the customized infrastructure Palantir has built over 20 years.

Karp described an LLM as “raw material” that only gains value after rigorous processing through Palantir’s existing software. He reminded attendees that the firm’s early success came from embedding engineers to tailor its platform, a painstaking approach quick-fix chatbot vendors tried to skip.

The session showed that organizations with deep, client-specific data systems integrated LLMs faster than newcomers chasing shortcuts. Palantir’s resilience, despite forecasts of irrelevance, underscores how long-term infrastructure bets are now yielding strategic returns.

Bootstrapped startup Plaud has sold more than 1 million NotePin AI recording devices since its 2023 launch, turning the gadget into a surprise hit among doctors, lawyers and executives. The $159 wearable transcribes and summarizes up to 20 hours of audio per charge using Plaud’s software and ChatGPT.

Device sales plus $99-per-year transcription plans put Plaud on track for $250 million in annualized revenue, with gross margins Xu says match Apple’s 25%. Founder Nathan Xu financed the company with personal savings and a $1 million crowdfunding campaign, keeping majority control while expanding a 20-person AI team in San Francisco.

Plaud’s profitability contrasts with heavily funded rivals like Rabbit and the now-defunct Humane, giving it an early lead in an AI wearable sector that has already attracted about $350 million and deals from Amazon and OpenAI. Xu markets the recorder as a professional tool and keeps data on U.S. Amazon servers to deflect privacy and geopolitical concerns.

Protect against the rising threat of deepfake calls. VoiceVault instantly analyzes speech patterns to spot synthetic voices and fraud attempts before they can do damage.

Catch scams as they happen. From account takeovers to social engineering, our AI surfaces suspicious behavior in real time — empowering your team to intervene immediately.

Stay ahead without slowing down. VoiceVault scales oversight across millions of calls, helping you safeguard your business without adding friction for legitimate customers.

Microsoft Research reveals a prototype analog optical computer that executes calculations with beams of light rather than electrons. The team estimates the approach delivers around a 100-fold boost in speed and energy efficiency for targeted optimization and AI tasks.

Engineers built the device from off-the-shelf micro-LEDs, lenses, and phone-camera sensors and paired it with a digital twin for large-scale testing. Early demos optimized thousands of bank transactions with Barclays and reconstructed MRI scans, trimming a 30-minute scan to roughly five.

The hardware currently processes only 256 parameters, highlighting its early stage. Running larger experiments in the digital twin shows Microsoft’s intent to expand optical computing into heavyweight AI workloads.

Mayor Matt Mahan is turning San Jose into a live test bed for AI-powered government, from transit scheduling to pothole detection and grant writing. His push includes equipping city buses with AI that made them run ahead of schedule during a recent ride-along.

San Jose hosts Nvidia’s headquarters, a dense semiconductor cluster, and graduates 8% of California’s engineers, giving the 1-million-resident city unusual AI talent and hardware on its doorstep. Mahan also launched the GovAI coalition, which lets tech-minded mayors nationwide share reviews of new AI tools.

California’s $20 billion budget gap makes the promise of software-driven efficiency politically attractive. Mahan argues the public sector has grown less productive over time, framing AI as a direct fix to voter frustration with slow, costly services.

Preliminary research presented at the American Heart Association’s Hypertension Scientific Sessions found that AI voice agents improved blood pressure reporting accuracy among 2,000 high-risk patients, most aged 65 and older. The system called patients in multiple languages, logged readings into electronic records, and automatically escalated concerning results to clinical staff.

The agent reached 85% of participants, and 60% provided a compliant reading during the call, while patient satisfaction averaged above 9 out of 10. Automating the task cut cost per reading by 88.7% compared with nurse-led calls.

Closing 1,939 controlling-blood-pressure gaps lifted the Medicare Advantage quality rating from 1 to 4 stars within ten weeks. The sharp rating jump alongside strong satisfaction scores shows AI can boost care quality while reducing clinician workload.