Good Morning from San Francisco,
AI search bots are flunking fact-checking while charging premium rates for their mistakes. The Tow Center caught eight chatbots fumbling basic news queries - tasks a quick Google search could solve. Even $40 monthly subscriptions buy you more confidence, not accuracy. Think of it as paying extra for an AI that's wrong with a fancier accent. 🔍
Meanwhile, Sony wants your PlayStation hero to chat back. They secretly tested an AI version of Aloy from Horizon Forbidden West, who responds to voice commands with fluid speech. The system blends OpenAI's whisper, GPT-4, and Llama 3 with Sony's voice tech. 🎯
Both stories share a theme: AI's confident sprint ahead of its actual capabilities. The tech impresses in demos but stumbles on basics - like finding a news article or knowing when to stay quiet. Sort of like that friend who's wrong about everything but never runs out of opinions. 😏
Stay curious,
Marcus Schuler
PlayStation's Secret Plan: Making Game Heroes Talk Back
Sony has unleashed a prototype AI-powered version of Aloy, the hero from Horizon Forbidden West. The character can now chat with players during gameplay, responding to voice prompts with synthesized speech and facial animations.
A leaked internal video shows Sony's PlayStation Studios building this conversational AI system. It combines OpenAI's whisper for speech recognition, GPT-4 and Llama 3 for conversation, and Sony's own tech for voice synthesis and facial animation.
Sony tested the system on both PC and PS5. The PS5 version runs with "little overhead" - tech speak for "it won't make your console catch fire."
The demo raises eyebrows across the gaming industry. Voice actors wonder if AI will steal their jobs. Developers question whether players really want to chat with the character they're supposed to be controlling.
Sony isn't alone in this AI gaming gold rush. Nvidia flaunts its "Ace" technology for AI-powered NPCs. Microsoft courts Inworld AI to bring chatty characters to Xbox. Even game studios themselves are diving in - half of developers surveyed now use AI tools at work.
Sony's experiment leaked just before the Game Developers Conference in San Francisco. The company responded with silence - and a swift copyright strike on the video.
Why this matters:
- Your PlayStation might soon talk back to you - and not just to remind you to charge your controller
- The line between playing a character and having a conversation with one is about to get very, very blurry
Read on, my dear:
AI Photo of the Day
Prompt:
face farming hair curly haircut looks
Study: AI Chatbots Can't Find News, But Love Making Up Stories
AI search tools are bombing basic fact-checking tests. Premium versions charge up to $40 monthly to be wrong with more confidence.
The Tow Center for Digital Journalism challenged eight AI chatbots to find original news sources. They stumbled hard. The bots bungled over 60% of queries that a simple Google search could solve.
Grok 3 led the failure parade, botching 94% of searches. DeepSeek misattributed sources 57.5% of the time. Even Perplexity, the top performer, flubbed 37% of answers.
Premium models strutted their stuff with special flair. Perplexity Pro ($20/month) and Grok 3 ($40/month) didn't just get things wrong - they declared their mistakes with absolute certainty. Picture a confident tour guide marching tourists straight into a lake.
The bots showed remarkable creativity in their failures. ChatGPT invented Wall Street Journal articles. Gemini time-traveled news pieces backward by three years. Grok 3 mastered the art of the broken link, serving up 154 error pages in 200 attempts.
Even formal partnerships couldn't save the day. Time magazine signed deals with OpenAI and Perplexity. Neither bot could reliably spot Time's articles. Imagine hiring a librarian who can't find books on their own shelf.
The premium service problem stands out. Companies market these upgrades as more reliable. Instead, they've created expensive echo chambers of wrong answers. Only Copilot showed any humility, choosing silence over spreading misinformation.
Why this matters:
- AI search tools have mastered one skill: being confidently wrong while charging for the privilege
- The more you pay, the less likely your AI assistant will admit it's lost - proving that in tech, confidence and competence don't share a price tag
Read on, my dear:
- implicator.ai: Premium AI Search: Pay More, Get More Confident Wrong Answers
- Columbia Journalism Review: AI Search Has A Citation Problem
Better Prompting 🔓: Recipe Creation
✂︎ ✂︎ ✂︎ ✂︎ ✂︎ Snippet ✂︎ ✂︎ ✂︎ ✂︎ ✂︎ ✂︎
Step into the role of a renowned chef specializing in international cuisine. Your task: Create a detailed, engaging recipe for a traditional dish from [COUNTRY/REGION].
The recipe must include:
- Complete ingredient list with exact measurements
- Detailed step-by-step cooking instructions
- Cultural and historical background of the dish
- Notes on typical serving occasions
- Prep time and difficulty level
Make sure instructions stay beginner-friendly.
Also provide:
- Tips for finding authentic ingredients or suitable substitutes
- Recipe variations (like vegetarian options)
- Plating and presentation suggestions
- Recommended sides and drinks that enhance the experience
Add personal stories or anecdotes that give the recipe authenticity. Help users recreate the true flavors and traditions of [COUNTRY/REGION] in their own kitchen.
✂︎ ✂︎ ✂︎ ✂︎ ✂︎ Snippet ✂︎ ✂︎ ✂︎ ✂︎ ✂︎ ✂︎
AI & Tech News
Xi to Chinese Economy: Hold My Beer While I Fund AI
Xi Jinping unleashed an ambitious tech spending spree at China's annual parliament, pumping 8.3% more into AI and advanced weapons while the economy sputters. The message rang clear through the Great Hall as 3,000 party faithful applauded: innovation trumps immediate economic relief, even if local governments drown in debt.
Tech Giants Stumble - Tesla Leads the Selloff
Tech heavyweights hit a wall on Black Monday. Tesla plunged 15%, while five "Magnificent Seven" stocks dropped over 4%. This marks Nasdaq's worst nosedive since September 2022, pushing S&P 500 dangerously close to correction territory.
Schmidt Eyes SpaceX Competition - Takes Relativity's Helm
Eric Schmidt, Google's former CEO, just pulled a space surprise. The 69-year-old snatched both majority control and the CEO seat at Relativity Space - his first chief executive role since leaving Google in 2011.
Mega Deal: CoreWeave Becomes OpenAI's New Hardware Muscle
AI startup CoreWeave just landed an $11.9 billion contract with OpenAI, who grabbed $350 million in CoreWeave shares. The deal arms ChatGPT's creator with Nvidia-powered AI infrastructure.
Bluesky: Longer Videos, Smarter DMs
Twitter's alternative now stretches video limits to three minutes and adds DM filters, tucking stranger's messages into a separate "Chat Requests" folder.
Musk's X Battles Outages - Blames Ukraine
X suffered massive Monday outages affecting thousands. While pro-Palestinian group "Dark Storm" claims responsibility, Musk points fingers at "Ukraine region" - his favorite scapegoat for technical hiccups.
Meta Takes On Nvidia With Secret AI Chip
Meta has begun testing its first in-house chip for AI training. The social media giant aims to cut its massive infrastructure costs and reduce dependence on Nvidia's expensive GPUs.
The chip specializes in AI tasks only, making it more power-efficient than standard graphics processors. Meta plans wider deployment if tests succeed.
The stakes are high. Meta expects to spend up to $119 billion in 2025, with most capital going to AI infrastructure.
Why this matters:
- Meta just crashed Nvidia's exclusive AI chip party - and brought its own silicon
- When a company burning $119B builds its own brain, the AI game changes
Read on, my dear:
AI Decoded 🔓:The New AI Models Fight It Out
The AI world spins faster than a caffeinated hamster in its wheel. Q1 2025 delivered four heavyweight contenders, each flexing unique muscles and sporting peculiar quirks.
GPT-4.5 shines as a digital therapist. OpenAI's latest offspring grasps subtle emotions and humor, helping users craft tricky emails or navigate domestic disputes. But throw complex math its way, and it stumbles. Some users still swear by its predecessor, GPT-4o.
Claude 3.7 Sonnet conquers the coding realm. Anthropic's new model outperforms OpenAI's o3-mini by 20 points on coding benchmarks. It particularly excels when paired with AI assistants like Cursor. Yet Claude sometimes plays it too safe, declining even harmless conversations.
Grok 3 surfs the internet unfiltered. Elon Musk's AI baby delivers real-time information with zero sugar-coating. Office workers beware - not every boss appreciates AI commentary without a diplomatic filter.
DeepSeek R1 crashes the party as the dark horse. This open-source newcomer thrills developers who love peeking under the hood. Beginners should tread carefully - its security gaps still gape like an forgotten zipper.
