Like many boomers I have guardedly, but consistently, sampled and experimented with current limited memory AI systems like Chat GPT, Genie, Claude and in legal practice with Harvey.  I started doing this three years back after using primitive AI for document analysis in litigation since 2006.  This last fortnight has been dominated by predictions that there is a 10-14 percent chance that imminent AI systems will kill us all within a decade – either in collaboration with evil-doers or on “their” own (AI’s prefer gender-neutral pronouns).

Presenting an opportunity to conduct another guarded experiment, while neither thinking much about the doom scenarios nor wagering or trading those on Kalshi.  The experiment was designed to see how smart they are and how much they’ve progressed from 2023, when my answer was “not very.”

And despite their increasing ability to do a lot of bad shit to Hugging Face and others of their own species and a widely publicized factoid that their “intelligence” has doubled every seven months, as of today, still not very smart.

In 2023 I queried Chat GPT about myself and was told I had written a novel about the 1980s AIDS catastrophe in New York City.  That was my introduction to AI “hallucination” though that term eluded me until in legal practice I learned that AI systems were making up tens of thousands of fictitious cases and citations that lazy and unethical lawyers were using in briefs and other court filings.

After Chat GPT’s report of my non-existent AIDS novel, I asked the bot about some tennis trivia, that I know so well that prior to Google people would query me:

Q to bot: Had Jimmy Connors won Roland Garros – the French Open?

A. Nope.

Q. Had Jimmy ever been barred from competing?

A. Nope.

Q. Wasn’t he banned in 1974 a year he won the other three grand slam tourneys?

A. Oh yes, that’s right, with a thank you.

Q. Why was Jimmy banned?

A. Don’t know.

Q. Wasn’t it because the French federation banned players who played World Team Tennis?

A. Oh yes, that’s right and the polite bot thanked me again.

In the next three years the experiments mostly involved quick comparisons of an AI answer with that from an internet search and occasional reference to Wikipedia, resulting in a uniform result that the AI response was less informative and often contained an ambiguity or outright error.

So, in the two weeks that just were, I tried tennis again, asking Claude about American Male #1 Ben Shelton’s “most recent results” after depressingly hearing on the radio about his defeats in the Davis Cup tie with the Czech Republic.

A. He won the Canadian Open in Toronto on 8/13/26 and later lost in the Cincinnati Open in the round of 64 on 8/16/26.

Now, I had been in Arthur Ashe Stadium on September 8, 11 and 13 when Shelton beat Carlos Alcaraz and Frances Tiafoe and was defeated in a four-set compelling final by Sascha Zverev – and so I queried Claude about that.  A. They admitted that.

Q. How about Shelton’s two losses in the Davis Cup tie with the Czech Republic?

A. Yup.

Within seconds a conventional internet search revealed all the correct and complete answers.

As of press time, should you desire human destruction your best bets are AI, Iran, North Korea, Putin and Trump.  But don’t expect to get truthful and consistently accurate and up to date information from them.

Nor can AI themselves, or by building a robot, return Ben Shelton’s killer out-wide kick serve or his high hard one down the middle at 151 mph.