A Lawsuit And A Suspicious 96% Average

Grok 4.5 came out last week and immediately started a knife fight over price.

xAI’s new model dropped on July 10 with token costs a fraction of its rivals, roughly $0.13 per task on one public leaderboard versus $1.57 for Anthropic’s top model. Musk asked Tesla engineers to test it alongside their existing tools and email him critical feedback. Around the same time, Tesla capped employee spending on outside AI at $200 a week.

Musk is not pretending Grok is the best. “Fable is definitely better than Grok 4.5,” he wrote on X, “but most tasks don’t require Fable-level capability.” That is the whole pitch in one sentence: not the smartest model, the one that is good enough for most work at a fraction of the cost.

Meanwhile Sam Altman spent the week talking up GPT-5.6 and telling reporters Musk is “obsessed” with him. On the benchmarks, Grok 4.5 sits mid-pack, behind several models from OpenAI, Anthropic, and Google. The two richest feuds in tech are now also a spreadsheet argument about cost per token.

For you, the takeaway is not the drama. It is that “which model is best” is no longer the only question companies ask. “Which model is good enough for 60 percent less” is the one that decides what you’ll actually use at work.

Apple took OpenAI to federal court, and the details read like a heist movie.

Apple filed a lawsuit on July 10 accusing OpenAI of stealing confidential AI trade secrets. The complaint centers on employees who left Apple for OpenAI, alleged secret recruiting conversations, and at least one company laptop that reportedly never came back.

This is the messy underside of the AI talent war. Every company is poaching every other company’s engineers, and those engineers carry knowledge in their heads that no NDA fully contains. When the people move this fast, “what walked out the door with them” becomes a courtroom question.

You do not need to work at Apple for this to matter. If you have ever switched jobs in the same field, you have carried skills, context, and habits to the next employer. The line between “my experience” and “their trade secret” is exactly what this case is going to test.

A Brown economics professor gave one take-home midterm in twenty years. The average score was 96 percent. He is pretty sure most of it was AI.

Roberto Serrano’s welfare economics class normally drew 30 students. This spring it drew 86, an enrollment jump he ties directly to the promise of take-home exams. Historic averages ran 65 to 80 percent on a test he made harder than usual. The class posted a 96.

So he made the final in person. More than a dozen students dropped. Even more failed. The gap between the take-home 96 and the in-person reality is the whole story, and Brown’s leadership response, in Serrano’s word, was “meek.”

This is the AI-in-school debate stripped of theory. The tools are good enough to fake mastery on paper, which means the only credential that survives is the one you can prove in the room, live, on the spot.

Some Japanese companies have given up on nagging employees to adopt AI. Now they just pay them.

Honda, ANA, and FamilyMart are among the firms rewarding workers with cash for using AI on the job, and some are baking AI adoption directly into performance evaluations. The reason is blunt: adoption there has been slower than in other economies, and the carrot is working better than the stick.

It sounds unusual until you realize your own company is doing a quieter version of the same thing. The raises, promotions, and interesting projects are quietly flowing to the people who use these tools well. Japan just made the incentive explicit instead of hidden.

The advantage is that the reward is real and it is available now. The employees who move first are getting paid, evaluated higher, and handed the work everyone else will be doing in a year.

Dating apps are losing users, so they are handing the flirting over to AI.

Bumble’s revenue fell to $966 million in 2025, down from $1.07 billion the year before. Match Group, which owns Tinder and Hinge, has posted shrinking subscribers for over a year. About 80 percent of younger users say the apps burned them out. The industry’s fix is AI that writes your profile, drafts your messages, and generates your photos.

So now you can match with a person whose profile was written by AI, exchange messages drafted by AI, using a photo touched up by AI. The date, if it happens, is the first fully human moment in the entire pipeline.

Some people are opting out entirely and going back to meeting in person, which dating coaches say is spiking demand for the one skill the apps quietly deleted: talking to a stranger without a script.