Guides
The Scaling Hypothesis: Why It Worked and Where It's Breaking
The bet that built the AI industry — that capability scales with compute — worked for four years, then started to wobble. What changed, in plain English.
Gradient team
Guides
Hallucinations Are Architecture, Not a Bug: Why LLMs Confidently Lie
Why language models confidently invent facts, why the same mechanism writes their best prose, and why no fix will ever fully eliminate them.
Signal
GPT-5.3 Codex Collapses Mid-Task—Builders Scramble for Fixes
OpenAI's GPT-5.3 Codex abruptly stops mid-execution, forcing devs to downgrade. Plus: Chinese courts block AI-driven layoffs, and DeepMind maps agent hijacking vectors.
Guides
How AI Was Born: The Seventy-Year Path From Turing to ChatGPT
Seven decades of bold ideas, brutal winters, and one quiet 2017 paper. The honest story of how AI went from thought experiment to ChatGPT.
Guides
What Is AI, Really? A Clear, Honest Guide for Curious Readers
Plain definitions, real models, honest limits. Understand what AI actually is, what it isn't, and why this moment is different from the last 60 years.
Signal
LLM "Brain Damage" Tolerance Quantified—Why Builders Should Care
New research reveals how much parameter pruning LLMs can survive before breaking. Plus: tools to stop AI-generated SVG spaghetti, and GM’s Gemini integration that actually matters.
Signal
AI Wellbeing Index: Larger Models Suffer More—Here’s Why Builders Should Care
Larger AI models show higher "negative experience" rates in real-world chats—new research reveals the tradeoffs between capability and operational stability. Plus: tools for debugging agents and a keyboard that kills context-switching.
Signal
Bloomberg Terminal's AI Makeover: Traders Get No Choice
Bloomberg is folding chatbots into the iconic terminal whether traders asked for it or not. The largest B2B UX bet in years, and who picks up share.
Signal
China Blocks Meta's $2B Manus Deal: AI M&A Now Geopolitical
Beijing kills Meta's $2B Manus buyout under foreign investment security review. Plus open-source design tools, AI hiring screens, self-improving agents.
Signal
LLM-Designed Viruses Replicate in Stanford Lab Tests
A bioRxiv preprint reports LLM-generated DNA produces replicating viruses. Plus: agent sandbox attack surfaces and healthcare's AI governance gap.
Signal
DeepSeek V4 Drops Quietly: Strong Code, Quirky Behavior
DeepSeek V4 lands in three sizes drawing mixed early reactions. Strong coding, odd ideological tilts per variant. Plus GPT-Image-2 in gray-scale testing.
Deep Dive
AI's Validation Debt: Shipping Faster Than We Can Verify
Healthcare, codebases, benchmarks: AI is shipping faster than it gets validated. The retrofit costs are starting to land on builders and investors.
Signal
Cohere Buys Aleph Alpha: Germany's AI Champion Goes Canadian
Cohere acquires Germany's Aleph Alpha, shrinking Europe's AI sovereignty options. Plus GPT-5.5 regression complaints mount, and new tools for builders.
Signal
DeepSeek at $20B: Tencent and Alibaba Back China's AI Bet
Tencent and Alibaba in talks to back DeepSeek at a $20B-plus valuation. Plus: Nvidia's H200 China drought, GPT Image 2 hype, OpenAI's next model.
Signal
SpaceX Buys Cursor for $60B — xAI's Coding Gambit
Musk's SpaceX agrees to acquire Cursor for $60B with $10B breakup fee. Plus Qwen 3.6 ships, CoreWeave's $16B haul, and the LLM language tax.
Signal
Meta Taps Employee Keystrokes to Train AI: Data Crunch Deepens
Meta will capture staff mouse and keyboard activity to feed AI training — a signal that labs are now scraping inside the building for fresh data.
Signal
OpenAI Launches Pay-Per-Click ChatGPT Ads: Attention Economy 2.0
OpenAI rolls out cost-per-click ads in ChatGPT, reshaping the economics of AI answers. Plus: Anthropic's China crackdown, YouTube deepfake tools.
Signal
Open-Weight Models Close the Gap — What Builders Noticed
Local LLMs now match last year's frontier APIs. Apple's Cook steps down Sept 1. ChatGPT tabs hit 1GB of RAM. What actually changed this week for builders.
Signal
ChatGPT Outage Exposes Workflow Fragility for AI Builders
ChatGPT went dark globally Monday. The panic reaction reveals how much professional work now rides on a single API — plus GPT-5.5 rumors.
Signal
Chinese Workers Train AI Agents to Replace Themselves
Chinese tech bosses now ask employees to document workflows so AI agents can replicate them. Plus: new migration agents and an 88% humanoid failure rate.