👉 OpenAI ditched DALL-E for GPT-4's artistic brain 🎨 - sacrificing speed for precision. The new system juggles 20 objects like a pro and even dares to draw celebs. Just don't hold your breath while it works.
👉 Enter Google's Gemini 2.5 Pro 🧠, an AI that finally learned to think before it speaks. It crushes genius-level math, spots birds in crowds, and devours million-token documents for breakfast. Best part? You can watch its mental gears turn in real-time. Like AI reality TV, minus the drama.
Both moves prove one thing: AI's trading party tricks for real smarts. The race for thoughtful AI? Game on. 🚀
Stay curious,
Marcus Schuler
OpenAI Ditches DALL-E for Something Slower But Smarter
Credit: OpenAI GPT 4o
OpenAI plugged image generation straight into ChatGPT today. The upgrade skips DALL-E in favor of GPT-4o's omnimodal brain. It's like replacing your microwave with a slow cooker – takes longer, tastes better.
The new system flexes some impressive muscles. It juggles 20 different objects without mixing up their colors or shapes. Previous AI artists would panic after eight items, creating the digital equivalent of abstract art.
Credit: OpenAI
Here's where it gets spicy: OpenAI dropped DALL-E's strict "no public figures" rule. Adults in the public eye? Fair game. Politicians, celebrities, influencers – they're all on the menu. The only escape hatch? An opt-out button. Expect a stampede of PR teams rushing to click it.
Research lead Gabriel Goh spills the technical tea: they abandoned the everyone-else-is-doing-it diffusion approach. Instead, GPT-4o draws like a human – left to right, top to bottom. This helps it nail text rendering, though it still squints at fine print.
Free users aren't left out in the cold. They get the same image allowance as before with DALL-E. How many exactly? OpenAI's spokesperson mastered the art of the elegant dodge on that one.
Karl Marx on a shopping spree in a US shopping mall / Credit: OpenAI
The system takes its sweet time crafting each image. OpenAI swears the quality boost justifies watching the digital paint dry. Their engineers probably hear "are we there yet?" in their sleep.
Why this matters:
OpenAI just proved slow and steady wins the AI art race – assuming users pack enough patience
The left-to-right approach might revolutionize AI art generation, but only after you've made a coffee run
That public figure policy just turned AI art into the world's most sophisticated meme generator – with an escape clause
Google just unleashed Gemini 2.5 Pro, an AI that stops to think before it blurts out answers. It's like they finally taught their model to count to ten before responding – and it's paying off big time.
The system tops the LM Arena leaderboard and crushes math tests designed for human geniuses. It scored 18.8% on Humanity's Last Exam, a test so hard it makes rocket science look like kindergarten homework.
Simon Willison put the model through its paces. It transcribed Spanish-English conversations with military precision, spotted pelicans in crowded photos, and cranked out new blog features faster than a caffeinated developer. The code quality matched Claude 3.7 Sonnet, its toughest competitor.
Credit: Google
This new version digests a million tokens at once – that's like reading "War and Peace" in one sitting. And Google promises to double that soon. The model juggles text, audio, images, video, and entire codebases without breaking a sweat.
Need proof? It helped Willison navigate 316,000 tokens of complex code while suggesting architectural improvements he hadn't considered in a year of planning. The model even caught missing asyncio support in his design – the AI equivalent of finding Waldo while solving a Rubik's cube.
The bounding box detection deserves a golf clap. It picked out individual pelicans in a flock while politely ignoring the photobombing egret. Even birds can't fool this AI.
Google baked "thinking" right into the model's core. This isn't just fancy marketing – the AI actually reasons through problems instead of playing high-stakes pattern matching. Users can even peek behind the curtain with a "Show thinking" button, like watching an AI solve a Sudoku puzzle in real-time.
Why this matters:
Google created an AI that actually thinks before it speaks – a skill some humans still haven't mastered
The million-token limit means this AI can digest entire codebases, books, or season recaps without getting digital indigestion
Developers might need new excuses for project delays – "the AI solved it while I watched Netflix" is now apparently valid
The "Show thinking" button could turn debugging into a spectator sport
Prompt: Animal fusion hybrid of giraffe legs and a golden retriever haed in natural habitat
AI & Tech News
AI firm dodges music industry's copyright punch
A federal judge blocked Universal Music Group's attempt to silence Anthropic's AI chatbot Claude on Tuesday, Reuters reports. Judge Eumi Lee swatted away music publishers' claims that the AI company's use of song lyrics caused "irreparable harm," noting that you can't license something before deciding if you need a license at all.
Copilot gets brainy with new 'deep reasoning' powers
Microsoft's Copilot just learned to crunch data like a caffeinated data scientist, The Verge reports. The company's new "Researcher" and "Analyst" AI tools promise to handle everything from complex research to Python coding – though perhaps we should wait and see if these digital minds can actually outsmart a cleverly placed checkbox.
AI models learn to clean up their own mess
Databricks just taught AI models to polish their own performance without needing pristine training data, Wired reports. The company's new trick, dubbed TAO, lets smaller models punch above their weight – their test model even managed to outsmart OpenAI's GPT-4 on financial tasks, though someone should probably check if it's been messing with its own test scores.
Amazon tests AI shopping buddy with a medical degree
BMW just handed Alibaba the keys to its AI future in China, tapping the tech giant's Banma system for its next generation of cars. The German automaker watched its sales plunge 13% in China last year while local brands zoomed ahead, forcing BMW to choose between pride and survival in the world's biggest car market.
TikTok Ban Support Plummets in US
Only 34% of Americans now support banning TikTok, plunging from 50% last year as ByteDance floods the airwaves with tearjerker ads about kidney donations and small business survival. The platform weaponizes emotion for its final battle, doubling US ad spending to $7 million ahead of the April 5 deadline when it must sell or shut down.
NSA Sounded Signal Alarm Before Yemen Strike Leak
NSA warned its staff about Signal's vulnerabilities in February – just weeks before Defense Secretary Hegseth spilled war plans to The Atlantic's editor in a group chat about Yemen strikes. The spy agency fretted about Russian hackers but couldn't prevent its own boss from hitting the wrong "Add to Group" button.
Block's Dorsey dumps 931 staff in performance purge
Jack Dorsey swung his cost-cutting scythe through Block on Tuesday, axing 931 employees while insisting this wasn't about money or AI replacement, TechCrunch reports. The fintech CEO's cleanup crew targeted three groups: strategic misfits (391), underperformers (460), and redundant managers (80) – though Dorsey's email somehow managed to make "you're fired" sound like a Silicon Valley optimization problem.
Apple unveils WWDC dates, promises iOS makeover
Apple will unleash its biggest software overhaul since 2013 at WWDC this June, pushing out iOS 19 alongside updates for iPad, Vision Pro, and Mac. The tech giant promises to revamp everything from icons to menus – though curiously, they're still sticking with that hybrid online-plus-one-day-party format that definitely has nothing to do with saving money on snacks at Apple Park.
Trump bans 80 tech firms from US shopping spree
The Trump administration slapped export restrictions on 80 companies Tuesday, targeting Chinese firms that had been gorging on American chips from Nvidia and Intel, The New York Times reports. The move especially stings for Nettrix, a crafty successor to a previously banned company that thought it had figured out how to keep the AI party going – turns out rebranding and moving down the street wasn't quite sneaky enough.
Europe hunts for Starlink alternative as Musk rules space
Europe wants to break free from Elon Musk's grip on Ukraine's battlefield communications, but its aging satellite operators face an awkward truth, the Financial Times reports. While Brussels dreams up alternatives to Starlink's 40,000 terminals in Ukraine, European companies admit their clunky, expensive systems can't match Musk's network – turns out moving chunks of metal through space isn't quite as easy as the billionaire makes it look.
YouTube dethrones Disney as boomers embrace streaming
YouTube snatched the TV viewing crown from Disney in February, with an unlikely boost from the over-50 crowd, The Hollywood Reporter reveals. The streaming giant grabbed 11.6% of U.S. TV time – and surprisingly, viewers old enough to remember rabbit ears contributed more watch time than teens and young adults combined.
Better prompting…
Today: Artistic Inspiration
Create a surreal oil painting that seamlessly blends the vastness of the ocean with the mysteries of outer space. Include vivid imagery, unexpected juxtapositions (like whales swimming among stars or coral reefs growing on asteroids), and a dreamlike atmosphere. Mention color palettes, lighting, and any symbolic or emotional elements the artwork might convey. 16:9
Voilà:
Credit: OpenAI / GPT 4o
Open Source Projects Fight Back Against AI's Data Feast
AI crawlers are devouring open source websites like teenagers at an all-you-can-eat buffet. These digital gluttons now generate 97% of traffic on some sites, forcing developers to adopt creative defense tactics.
The Read the Docs project blocked the bots and watched their daily traffic plummet from 800GB to 200GB. Their monthly savings? A cool $1,500 in bandwidth costs.
Some devs now trap greedy crawlers in endless mazes of fake content. It's like sending robots to digital detention.
Why this matters:
The internet's collaborative spirit faces an existential threat from AI companies treating open source like their personal data buffet
Open source developers are fighting back with ingenious solutions, proving that sometimes the best defense is a crafty offense
Sudowrite: AI's Secret Weapon for Fiction Writers 🎯
Sudowrite supercharges storytelling with AI smarts. This writer's best friend helps authors blast through creative blocks and craft compelling narratives in record time ⚡️
The Founders 👥 Tech meets storytelling in SF, 2020. Amit Gupta (Photojojo founder turned sci-fi writer) and James Yu (Parse founder) joined forces after bonding in a writing group. Now their lean team of <20 pioneers creative AI 🌉
The Product ⚙️ Writers' Swiss Army knife:
Story Engine cranks out plots and characters 📚
One-click style shifts and rewrites 🔄
Story Bible keeps your world organized 📝
Visual story mapping 🗺️
Pro-grade AI models under the hood 🧠
Built-in feedback loop with AI + human readers 👥
The Competition 🥊 Leaves marketing-focused Jasper.ai in the dust for pure storytelling magic. While ChatGPT plays generalist, Sudowrite speaks fiction fluently. No contest vs enterprise-focused Writer.com 💪
Financing 💰 Smart money: $3M seed round (2021) after bootstrapping. Revenue cruising at $750K yearly. Keeps it lean, means business 📈
The Future 🔮 Rules the AI fiction niche with swagger. Eyeing script tools and publisher team-ups. Must dodge ethics drama and keep its tech edge sharp. Strong writer gang and laser focus spell success ahead 🎯
