AI News - Jul 18, 2025

Welcome to AI News in 5 Minutes or Less, where we digest the latest in artificial intelligence faster than ChatGPT can write your homework and with fewer hallucinations than your last company holiday party. I'm your host, and I promise to be more reliable than Anthropic's usage limits.

Speaking of which, our top story today: Anthropic just pulled what I'm calling "The Great Claude Code Caper." Without warning, they tightened usage limits on Claude Code, leaving developers more shocked than finding out their AI girlfriend was actually just a very sophisticated autocomplete. Multiple sources report outraged developers who went from coding paradise to rationing their API calls like it's toilet paper in 2020. Anthropic's response? Crickets. Which, coincidentally, is also the sound of developers' productivity grinding to a halt.

But wait, there's competition heating up! Chinese AI model Kimi K2 is undercutting rivals with prices so low, even your local dollar store is jealous. It's like the Walmart of AI models, except instead of greeters, you get responses in broken English that somehow still make more sense than my last performance review. The model's already racked up over 100,000 downloads on HuggingFace, proving that in the AI world, being cheap and cheerful beats being expensive and existential.

Meanwhile, OpenAI just dropped their ChatGPT Agent, which they claim can think, act, and use tools. So basically, it's everything I pretended to be on my resume. The agent can do research, make bookings, and create slideshows, making it more productive than most middle managers. OpenAI staffer Keren Gu tweeted it has the "strongest safeguards," which is corporate speak for "we really hope it doesn't book you a one-way ticket to Antarctica while trying to order pizza."

Competition update: Anthropic also launched Claude for Financial Services, because apparently regular Claude wasn't anxious enough about money. Now it can access real-time financial data and probably judge your spending habits too. "I see you bought another subscription service, Dave. Have you considered the compound interest you're missing out on?"

Time for our rapid-fire round!
Google's Gemini 2.5 family is expanding faster than my waistline during lockdown, with Flash, Pro, and the adorable new Flash-Lite.

Sam Altman claims Meta is offering 100 million dollars to poach OpenAI employees. That's roughly the cost of one ChatGPT query by 2030 at current pricing trends.

Mistral released Voxtral models that speak eight languages, making them more multilingual than my Uber driver but probably less opinionated about the fastest route.

And GitHub's AutoGPT project hit 177,000 stars, proving that developers love automation almost as much as they love arguing about tabs versus spaces.

For our technical spotlight: Researchers just published a paper showing AI can achieve second place at competitive programming contests. Great, now AI is better at coding than me AND it doesn't need coffee breaks or complain about Sprint planning. The real kicker? It solved problems that would make most humans cry into their mechanical keyboards.

Also noteworthy: Google's working on AI for dolphin communication. Because apparently teaching computers to talk to humans wasn't hard enough. Coming soon: "ChatGPT, but for marine mammals." I can't wait for dolphins to start arguing about whether they're experiencing true intelligence or just sophisticated echolocation.

Before we wrap up, let's address the elephant in the server room. Hacker News users are having an existential crisis about whether current AI is "real intelligence" or just "glorified prediction systems." One user called it "false confidence without consequence," which coincidentally describes my entire dating history. They want features like verbatim memory and version control, because nothing says "artificial intelligence" like remembering things accurately and not gaslighting users about previous conversations.

That's all for today's AI News in 5 Minutes or Less! Remember, in a world where AI can write code, generate videos, and potentially talk to dolphins, the most impressive feat might still be getting through a conversation without it mansplaining quantum physics to you.

Until next time, keep your tokens close and your API keys closer. This has been your guide to the AI apocalypse, now with 30% more accuracy than a ChatGPT history essay!