🚀 OpenAI drops GPT-5 with a unified thinking system

Welcome back to the AI Business Summary newsletter!

OpenAI just dropped GPT‑5 ⚡, cutting hallucinations sixfold and matching expert performance in half of test scenarios.

Here’s everything else you need to know this week in AI...

In Today’s Issue:

🚀 OpenAI drops GPT-5 with a unified thinking system

🤝 OpenAI partners with Broadcom for 10-gigawatt chip deployment

💻 NextSilicon reveals processor to challenge Intel and AMD

🎙️ gpt-realtime goes production-ready with phone calling support

👓 Amazon tests AI smart glasses for delivery drivers

🚗 GM announces eyes-off driving and Google AI integration

🧬 UK fast-tracks seven new AI medical tools

⚖️ Reddit sues Perplexity for large-scale data scraping

AI Systems & Infrastructure

🚀 OpenAI drops GPT-5 with a unified thinking system

• GPT-5 responses ~45% less likely to contain factual errors than GPT-4o

• Routes between fast responses and deep reasoning automatically

• Comparable to experts in roughly half of the tested scenarios

• Six times fewer hallucinations than O3 on factuality benchmarks

🤝 OpenAI partners with Broadcom for 10-gigawatt chip deployment

• Custom AI accelerators designed specifically for OpenAI models

• Deployments start second half of 2026, completed by 2029

• Broadcom handles manufacturing and Ethernet infrastructure scaling

• Targets surging global demand across partner data centers

💻 NextSilicon reveals processor to challenge Intel and AMD

• $300 million-funded startup targets the scientific computing market

• Uses the open RISC-V standard, competing with the ARM architecture

• Sandia National Labs has been evaluating prototypes for three years

• Promises faster performance with less power consumption

AI Interfaces & Tools

🎙️ gpt-realtime goes production-ready with phone calling support

• New speech model scores 82.8% on reasoning benchmarks

• Enables image inputs and remote tool connections

• Direct phone network integration via SIP protocol

• 20% price cut versus gpt-4o-realtime-preview model

👓 Amazon tests AI smart glasses for delivery drivers

• Hands-free package scanning and turn-by-turn walking directions

• Computer vision detects hazards and guides building navigation

• Emergency button and prescription lens compatibility included

• Currently trialing in North America before wider rollout

AI in Industries

🚗 GM announces eyes-off driving and Google AI integration

• Gemini AI assistant launches in vehicles next year

• Lidar-based autonomous system debuts on Escalade IQ 2028

• New centralized computing platform across all models

• An energy home system available via leasing starting in 2026

🧬 UK fast-tracks seven new AI medical tools

• MHRA AI Airlock program enters second phase

• Tools promise minutes instead of weeks for test results

• Includes cancer diagnostics and genetic eye disease detection

• Regulatory sandbox approach ensures safety before deployment

AI Regulation & Ethics

⚖️ Reddit sues Perplexity for large-scale data scraping

• Alleges circumvention of data protection measures for training

• Claims 40x increase in Reddit citations after cease-and-desist

• Targets three data-scraping companies plus Perplexity directly

• Action: audit your training-data licenses now

🌱 Bonus Thought

Financial regulators just issued their clearest warning yet: AI delivers massive efficiency gains while concentrating risk in ways we've never seen. The Financial Stability Board's latest report highlights how third-party dependencies mean one AI vendor failure could cascade across multiple banks. Market correlations amplify when algorithms make similar decisions simultaneously. Cyber risks multiply as attack surfaces expand.

Meanwhile, Reddit sues Perplexity for data scraping while GM promises hands-off driving. Two sides of the same challenge: who controls the data that powers these systems? The pattern spans finance, transport, and healthcare. Speed scales faster than safety frameworks. Third-party concentration versus operational efficiency. Innovation velocity versus governance maturity.

Companies racing ahead without proper vendor audits and failure-mode testing aren't just risking compliance. They're betting their entire business on systems they don't fully control or understand. The FSB report makes this clear: existing frameworks may not be sufficient for AI-driven financial stability risks.

Audit your AI dependencies before they audit you.

📝 Business Prompt to Try

Why It Works

• Data-driven optimization: Uses historical performance to pinpoint what drives or kills engagement.

• Fast payoff: Produces ready‑to‑deploy experiments that can show measurable lift within two weeks.

• Smart personalization: Keeps messaging on‑brand while applying AI’s pattern recognition to tone, timing, and phrasing.

• Built for iteration: Encourages continuous learning via A/B cycles rather than one‑off guesses.

💡 Quote of the Week