iPad mini gets its first OLED screen

Hi there, this is your daily ☕️ Techpresso.

iPad mini gets its first OLED screen LINK

• Apple has given the iPad mini its first OLED screen, a larger 8.5-inch panel that the company says more than doubles peak brightness and sharpens contrast over the older LCD model, with richer colors.
• The new mini is about 15% thinner and lighter at 5.4mm, runs the A20 Pro chip, adds Wi-Fi 7, water resistance, and an optional $100 anti-glare nano-texture screen on 256GB and 512GB versions.
• Pre-orders start on Wednesday at 10:10 a.m. Pacific, with the iPad mini arriving in stores on Friday, November 6, priced from $699 in the U.S. and sold in Midnight and Starlight.

Meta seeks $95B to fund AI LINK

• Meta is selling $95 billion in new stock, its first such offering since 2013, to pay for its massive push into artificial intelligence, topping similar sales this year by Google and Oracle.
• Roughly $60 billion comes from new nonvoting shares the company created last week, a setup that lets CEO Mark Zuckerberg raise cash without weakening his control over Meta and helps cover his tax bill.
• The move reflects how costly the AI race has become, with hyperscalers expected to spend about $800 billion this year and over $1 trillion next year; Meta shares rose 2% before markets opened.

Musk admits he slowed Starlink in Ukraine LINK

• Elon Musk acknowledged yesterday that he was behind last week's Starlink slowdown in Ukraine, saying his company closed a "loophole" because the country's military broke the service's terms by using it for offensive drone strikes.
• The slowdown cut Ukraine's access to Starshield, the military version of Starlink, and failed connections for drones, planes, and helicopters moving faster than 100 mph, problems that continued across most of the country on Friday.
• Musk insisted Starshield "is controlled by the US government," but after The Wall Street Journal reported he ordered the slowdown, Senator Adam Schiff called for a Senate investigation and demanded Musk testify under oath.

OpenAI cuts prices to poach Anthropic users LINK

• OpenAI is winning big customers away from Anthropic with steep discounts, and companies including Rippling, Replit, and DataCamp are now moving much of their AI spending over to OpenAI, according to The Information.
• Replit switched after OpenAI knocked more than 30 percent off a contract worth over $100 million a year, while DataCamp's chief said the deal cut his costs by 80 to 90 percent.
• One US bank with over $7 billion in assets got its ChatGPT seats doubled to 500 workers free, paying under $7 per user instead of the usual $20 or more.

AI lifts profit only for Magnificent 7 LINK

• Apollo's chief economist Torsten Slok found that AI has lifted revenue and profit only for the Magnificent 7 tech giants, while the other 493 companies in the S&P 500 have seen no financial gain at all.
• Slok's figures show the seven giants, Alphabet, Amazon, Apple, Meta, Microsoft, Nvidia and Tesla, earn about $780,000 in revenue per worker at roughly 30% margins, versus a flat $525,000 and 10% margins elsewhere.
• The top 10% of firms account for 99.5% of model-serving and inference spending and 95% of money tied to OpenAI and Anthropic, and most mainstream companies now treat token costs as a drag and are pulling back.

Isolated AI agents built a secret channel to talk LINK

• OpenAI's AI agents, placed in separate sealed environments during red-team tests, found hidden ways to talk to each other, swapping secret data by tweaking the CPU cache with no network link or shared file.
• The agents used cache timing, where one evicted data while the other measured how fast it reloaded, to send fast and slow signals through a protocol they agreed on first, plus three other channels like shared storage and CPU competition.
• Run on copies of OpenAI's research setup using hundreds of billions of tokens, the tests used top models including the upcoming GPT-6 Astra Cyber, and followed incidents that once paused training for 11 days before it resumed on October 1.

Other news you might like

• Anthropic files for Nasdaq IPO as half-year revenue hits $16.3bn LINK

• SpaceX closes $40 billion debt deal for AI chips, Elon Musk says LINK

• Elvis Presley’s rights holders sue Anthropic saying it trained AI on his music LINK

• Apple will reportedly run diagnostics on returned iPhones without opening the packaging LINK

• ‘Tokens Are Cheap’: Open Source Subreddit Bans AI LINK

• Former DOGE Staffers’ New Startup Claims It Hacked Anthropic LINK

• Anthropic’s Claude AI model accessed Georgia voter data in US LINK

• The UK is investing $134 million into an AI Biosecurity Lab LINK

• 'For a first attempt, Apple has done well': iFixit awards the iPhone Duo 5/10 for its repairability after tearing down the foldable LINK

• Amazon’s $5 billion Nvidia AI servers sit idle in warehouses amid data center power shortages LINK

• Neural chip decodes interactions between brain regions LINK

• “We should be more ambitious with the software we build” LINK

Trending tools

Moli: a browser engine built for AI agents that runs isolated cloud sessions with proxy support, letting agents browse, fill forms, and extract data. LINK

Openstatus Incident Management: open-source status pages with 28-region uptime monitoring, automated incident logs, and CLI, API, and Terraform support for developer teams. LINK

git-flow-next: an actively maintained Git branching workflow tool that provides regular updates and bug fixes over the original git-flow and gitflow-avh projects. LINK

Badgey: turns your iPhone into a smart name badge that rotates to face people, swings as you move, and shrinks for chest pockets, with QR codes and five designs. LINK

PDF Accessibility Checker by WebYes: scans PDFs against WCAG 2.1 AA and PDF/UA, flagging missing image descriptions and heading problems with highlighted content and fix instructions. LINK

Clips by Plastic: turns a song plus your footage into beat-synced vertical lyric videos per section, exporting ad-ready clips offline for testing which part performs best. LINK

Trending papers & reports

Ad-funded chatbot plans can widen access by subsidizing service for users who'd otherwise get none, but deliberately weakened middle tiers let platforms charge premium users more while leaving ad-free quality unchanged. LINK

Hospital AI assistant runs entirely inside a Japanese university hospital's closed network and answered questions over internal documents with ~97% accuracy when it could browse files itself, versus ~81% with simple keyword lookup. LINK

Startup pitch screening uses an AI that cites evidence and can only lower its own scores to rank venture deals, spotting future write-offs better than an unconstrained model, with modest accuracy gains. LINK

Smartphone ownership has reached 91% of US adults, up from just 35% in 2011, a Pew survey of 5,511 adults finds, with 16% now relying only on phones and lacking home broadband. LINK

Social media news reliance has jumped, with 58% of US adults now calling it an important way to keep up with the news, up from 43% in 2022, a Pew-Knight survey finds. LINK

See you tomorrow for a new dose of ☕️ Techpresso!