Geek Stuff category archive

Cocksure of itself? At the Psychology Today website, Alain Samson warns us that “(b)oth LLMs themselves and users tend to overestimate the correctness of LLMs’ answers.”

Gullible? It’s willing to buy that bridge in Brooklyn.

To be trusted implicitly? At the Psychology Today website, Cornelia C. Walther says no way. Here’s a bit (emphasis added):

In three pre-registered experiments involving more than 1,300 participants, something striking showed: When people had access to an AI assistant, they consulted it on more than half of all tasks—and their accuracy mirrored the AI’s almost perfectly. When the AI was right, they were right. When it was wrong, so were they. Most crucially, they weren’t checking. They were simply adopting the AI’s answers, bypassing both instinct and analysis entirely.

Secure? Security maven Bruce Schneier notes that

Unlike traditional computing systems that strictly separate executable code from user data, LLMs process all input—whether it is a system command, a user’s email, or a retrieved document—as a single, undifferentiated sequence of tokens. There is no architectural boundary to enforce a distinction between trusted instructions and untrusted data. Consequently, a malicious instruction embedded in a seemingly harmless document is processed with the same authority as a system command.

Click to read the whole article.

Corrosive? At the Psychology Today website, Cornelia C. Walther reports that “(r)esearch found that AI improved efficiency while eroding underlying expertise and agency.”

To put it another way, relying on AI to do our thinking for us may make us dumber.

Promoting passivity? At Psychology Today Blogs, John Nosta posits that the risk of AI “isn’t machine thought, but emergent passivity in us.”

Manipulative? You bet your sweet bippy.

Good for doing homework? Not if you actually, like, you know, want to learn stuff.

My most excellent hosting provider has informed me that maintenance is scheduled for my server “between February 09, 2026, 6:30 AM GMT. and February 10, 2026, 1:30 PM GMT,” and that some hopefully minimal down time may result. Sporadic bloggery is a possibility.

Mageia v. 9 with the Plasma desktop. The wallpaper is from my collection.

Programmed to prevaricate? At Psychology Today Blogs, John Nosta explains why your AI bot will come up with an answer even when there isn’t one.

At Psychology Today Blogs, Mitchell B. Liester reminds us that “social” media isn’t. He notes that (emphasis added)

Family estrangement has reached epidemic proportions. . . .

What’s causing this destruction? Causes include substance abuse, violence, and personality conflicts, but a newer and increasingly powerful force is social media algorithms designed to increase engagement by promoting divisive content.

In these algorithmic times, methinks the entire article is worthy of your attention.