New research finds AI models often fabricate step-by-step explanations that look convincing but don't reflect their actual reasoning. 25% of recent papers incorrectly treat these as reliable—affecting medicine, law, and safety systems.
AI meeting assistants are transforming how teams capture and act on discussions. From free tools like Fathom to advanced analytics from Avoma, these 10 apps automatically transcribe, summarize, and extract action items from calls.
In what might be called the tech equivalent of a midlife crisis, Baidu just discovered the joy of sharing. The Chinese search giant unveiled Ernie X1, a new reasoning-focused AI model, while making its entire AI suite free - several weeks ahead of schedule and years after anyone asked.
The timing tells a story of disruption. Once China's undisputed AI leader, Baidu has watched upstart DeepSeek rewrite the rules by matching premium AI performance at Walmart prices. ByteDance and Moonshot AI piled on, stealing users despite Baidu's head start in the ChatGPT race.
Credit: Baidu
Now Baidu claims its upgraded Ernie 4.5 beats OpenAI's latest offering. But the real headline isn't the technology - it's the strategy. By June 30, Baidu will open-source its AI models, abandoning its walled garden approach. They're even integrating rival DeepSeek's R1 model into their search engine, a move that probably hurt their pride as much as it helps their product.
The financials tell the rest: while AI boosted cloud revenue 26% last quarter, weak ad sales cast a shadow. Baidu just freed up $1.6 billion from its YY Live acquisition for AI investment - because sometimes staying relevant means breaking open the piggy bank.
This dramatic shift reveals a deeper truth about tech: yesterday's market leader can become tomorrow's fast follower. In China's AI landscape, even giants must adapt or risk irrelevance. Baidu's transformation from gatekeeper to open-source advocate shows just how thoroughly DeepSeek has shaken up the status quo.
Why this matters:
Shows how a single disruptive player (DeepSeek) can force industry giants to fundamentally rethink their business
Demonstrates that in AI, technical superiority matters less than accessibility and developer adoption
Grammarly bought email app Superhuman for an undisclosed sum, part of its plan to build an AI productivity empire. With $1 billion in fresh funding, the grammar company wants to put AI agents at the center of your workday.
While Congress debates TikTok's future, ByteDance quietly built America's #2 education app. Gauth helps 200 million students cheat on homework by solving problems from photos. Same company, same data concerns, zero scrutiny.
Programming computers in English sounds impossible. But Andrej Karpathy built working apps without knowing code, using only natural language prompts. He calls it Software 3.0. These AI systems think like humans, complete with superhuman memory and distinctly human mistakes.
Chinese AI startup MiniMax just trained a frontier model for $534,700 - a fraction of what competitors spend. Their open-source M1 beats DeepSeek while using 75% less computing power. The breakthrough suggests advanced AI no longer requires massive budgets.