Get tomorrow's brief in your inbox
Today: Harvard shows AI models beat ER doctors at initial diagnoses. Google Gemini now handles multi-turn conversations without repeating "Hey Google." Denmark pauses data center construction as grid capacity hits a wall.
AI outperforms emergency room doctors in diagnostic accuracy
A Harvard Medical School study published in Science shows OpenAI's o1 model delivered more accurate emergency room diagnoses than attending physicians. Researchers tested the model on 76 real ER patients at Beth Israel Deaconess Medical Center, comparing AI-generated diagnoses against those from two internal medicine doctors. At the critical first triage point (when information is most limited and urgency highest), o1 delivered exact or very close diagnoses 67% of the time, versus 55% and 50% for the two human physicians. The models received no pre-processing, just the raw data from electronic medical records at each diagnostic checkpoint.
This matters for small businesses in healthcare, urgent care, or telemedicine. If you run a clinic or offer medical services, AI diagnostic assistance is moving from experimental to practical. The study's lead author emphasized these findings show "an urgent need for prospective trials" in real patient care settings, meaning regulatory frameworks and liability structures are still being worked out. The models only handled text-based information, not imaging or physical examination data. If you're evaluating AI tools for clinical decision support, look for solutions that integrate with your EHR system and focus on triage or second-opinion use cases rather than final diagnosis. The business opportunity is reducing misdiagnosis rates and speeding up initial assessments, which translates to better patient outcomes and lower malpractice risk.
Google Gemini adds Continued Conversation for multi-turn voice commands
Google rolled out a Continued Conversation feature for Gemini, its voice assistant for the home. After the initial "Hey Google" activation, the microphone stays active for a few seconds (indicated by pulsing lights), letting you follow up without repeating the wake phrase. The upgrade includes contextual memory (Gemini remembers prior questions in the conversation), multilingual support, better side-talk detection (fewer false triggers in noisy environments), and whole-home access for staff or guests.
If you run a retail store, service business, or any operation where staff interact with customers while multitasking, this makes voice commands more practical. You can ask "What's today's schedule?" then immediately follow with "Email the team the update" without re-activating. Setup is straightforward: open the Google Home app, go to Home Settings, and enable the feature under Gemini voice assistant. Free to use if you already have Google Home devices. The downside is potential misinterpretation in busy environments and the need to train staff on new tech. Data privacy is a concern if you're collecting customer info via voice, so review Google's data handling policies before rolling this out in client-facing settings.
More on Gemini Continued Conversation
Denmark pauses data center construction as power grid maxes out
Denmark's state-owned grid operator Energinet introduced a temporary pause on new grid connection agreements after capacity requests hit 60 GW, nearly 9x the country's peak electricity demand of 7 GW. Data centers account for 14 GW of the requests (about 23%). The moratorium lasts three months or until new prioritization criteria are established. No political decisions yet, as Denmark is forming a new government. The U.S. is seeing similar pushback: Maine nearly banned data center construction, and Pennsylvania, Virginia, and Oklahoma are considering moratoriums.
If you're planning cloud infrastructure investments or considering colocation deals, factor in the possibility of capacity constraints. Regions that were previously attractive for cheap renewable energy (Nordics, Pacific Northwest) may not have grid availability when you need it. If your business depends on low-latency access to data centers (gaming, real-time analytics, video processing), diversify your geographic footprint now rather than betting on a single region. The trend is toward stricter access criteria based on "societal value" and project maturity, so enterprise customers with long-term contracts and clear use cases will get priority over speculative deployments.
Full story on Denmark's data center pause
Big Tech earnings show massive AI infrastructure spending pays off
Alphabet, Amazon, Apple, Microsoft, and Meta reported earnings this week, revealing $180 billion to $200 billion in annual data center spending. Alphabet's Google Cloud grew 63% to an $80 billion annual run rate, the fastest growth among cloud providers. Amazon Web Services grew 28% (fastest in 15 quarters) to $150 billion annual run rate. Microsoft's Azure is up 40% to $90-95 billion, though the stock fell 2.4% on uncertainty about OpenAI's contribution to those numbers. Meta's stock dropped 9.8% despite $125-145 billion in planned spend, likely due to lack of clear ROI on internal AI training.
For small businesses, this confirms the cloud providers are not pulling back. If you've been hesitating to commit to a cloud platform because of "bubble" fears, those fears are overblown. The spending is translating to revenue growth, which means these platforms will be around and improving. If you're choosing between AWS, Google Cloud, and Azure, note that Google Cloud is growing fastest (potential for more aggressive pricing and feature releases) while AWS has the largest base and most mature enterprise tooling. Apple is mostly free-riding on Google's Gemini for its 2.5 billion devices, which keeps its spending low but doesn't help you unless you're building iOS apps that tap into Apple Intelligence features.
Analysis of Big Tech AI spending
"This is fine" meme creator says AI startup stole his art
KC Green, creator of the famous "This is fine" dog meme, accused AI startup Artisan of using his copyrighted art in a subway ad campaign without permission. The ad showed the dog saying "my pipeline is on fire" with a message to "Hire Ava the AI BDR." Green posted on Bluesky that the art was "stolen like AI steals" and urged people to vandalize the ads if they see them. Artisan said it has "a lot of respect for KC Green" and scheduled a call with him. This follows previous controversy over Artisan's "Stop hiring humans" billboard campaign. Green told TechCrunch he's looking into legal representation but said it "takes the wind out of my sails" to spend time on lawsuits instead of creating comics.
If you're using AI image generation tools (Midjourney, DALL-E, Stable Diffusion) for marketing materials, this is a reminder that copyright risk is real. Memes are not public domain just because they're widely shared. If you're creating ads, social media posts, or branded content, either license images properly, commission original work, or use tools that indemnify you against copyright claims (Adobe Firefly, Shutterstock's AI generator). If you're running an AI startup, don't use recognizable copyrighted art in your campaigns, even if you modify it. The legal downside (settlements, injunctions, brand damage) far outweighs the attention you'll get.
Details on the Artisan controversy
Accounting software choices for startups
If you're launching a business and need accounting software, here are the options that matter. QuickBooks Online starts at $35/month and offers 650+ integrations, making it the default for venture-backed startups that need to scale quickly. Wave Accounting is free for basic services (invoicing, expense tracking), ideal for very small operations with simple needs; you pay only for add-ons like payroll. Xero is user-friendly with strong cash flow management and 800+ integrations, though pricing is similar to QuickBooks. Brex combines expense management, banking, and business credit cards with no monthly fees, solving multiple problems in one platform. Sage Intacct is for complex needs (multi-entity management, custom reporting), suitable if you're growing fast and need enterprise-grade features.
The decision comes down to your growth trajectory and complexity. If you're bootstrapping and doing under $100K in annual revenue, start with Wave. If you're raising capital or planning to scale quickly, use QuickBooks or Xero. If you're managing multiple entities or have complex reporting needs (franchise, multi-location, international), evaluate Sage Intacct. Brex makes sense if you need working capital and want to consolidate expense tracking with banking services. Don't overthink it: pick one, set it up correctly from day one (chart of accounts, expense categories), and you can always migrate later if you outgrow it.
Full comparison of accounting software
Use accounting software from day one, not "when you need it"
Most startups delay setting up proper accounting until they hit a pain point (tax season, investor due diligence, cash flow crisis). By then, you're backfilling months of transactions, which is expensive and error-prone. If you just started a business, spend two hours this week setting up Wave (free) or QuickBooks ($35/month). Create your chart of accounts, connect your bank account, and categorize your first transactions. This habit takes 10 minutes per week once it's running and saves you thousands in accounting fees and stress later. If you're already behind, hire a bookkeeper to clean up the past six months and commit to maintaining it yourself going forward.
AI diagnostic tools are moving from research papers to real clinical workflows, but expect a slow regulatory path before they're standard practice. Voice assistants are getting smarter at multi-turn conversations, which makes them more viable for business operations (customer service, staff coordination, inventory checks). Power grid constraints in data center hubs signal that cloud capacity isn't infinite; if you're planning infrastructure-heavy projects, diversify your geographic strategy and lock in commitments early. Big Tech's massive AI spending is producing revenue growth, which means the tools and platforms you're using today will keep improving rather than disappearing in a "bubble burst."