AI Right Now — The 5 Things that Matter.
Anthropic developed a model that was too dangerous to release.
AI AUTOMATION
Jyotsna
4/16/20261 min read
No fluff. No jargon. Just the stories shaping artificial intelligence this week.
No one obviously wins - and that is different.
01 . With a single hair Anthropic leads. xAI, Google, and OpenAI are right behind.
DeepSeek of China is hardly any older. The difference between the optimal and second-best AI model is as small as ever. The race has shifted to who is the smartest to who is cheapest and most reliable.
02. Anthropic developed a model that was too dangerous to release.
Its Mythos AI was capable of breaching 80 percent of the software systems that it attacked, with no human assistance. It was silently limited by Anthropic. It is likely at least months before other labs can do the same. Cybersecurity has forever altered.
03. OpenAI would like to place advertisements in your chatbot.
100 billion in advertising revenue in 2030. That's the bet. Customers enter what they want in the chat - ideal targeting. The danger: as soon as you think that the AI is selling you something, you lose trust in it in all other aspects.
04. Genius at physics. Useless at dishes.
The best AI models are now able to answer over 50 percent of questions in PhD level exams, as compared to only 8.8 percent last year. Meanwhile, robots can manage tasks in the house with only 12% success. Researchers call it "jagged intelligence." It is the weirdest thing in technology at the moment.
05. 75% of AI's money goes to 20% of companies
PwC researched 1,200 executives in 25 industries. The winners are not only cutting down on costs using AI but they are also reinventing their complete business model using AI. All other people are flying pilots that have no destinations.
