AI sucks!

[bill718]

Phone calls routed to the wrong departments, AI programs posting payments to the wrong accounts in business ledgers, secure emails forwarded to the wrong addresses, Insurance companies automatically billed for medical services with the wrong dates of service - The list of incompetence is almost endless!

Good news for me. My medical billing company is racking up lots of billable hours fixing these mini disasters for clinics, but AI is just 1 reason healthcare costs are so high in America. Maybe one day this AI stuff will work like it's supposed to, but right now it's (As the Shakespeare play said) A Comedy of Errors.

Why doesn't AI work very well and what can we do about it?

Despite billions of dollars invested in ever-larger models, problems with AI hallucinations persist. So what's going wrong, and is there a better path forward? Professor Artur d'Avila Garcez of City St George's, University of London, explores.

www.bcs.org

 

[citygator]

According to AI (Gemini):

The question of whether AI "sucks" is a bit like asking if a hammer "sucks"—it depends entirely on whether you're trying to drive a nail or perform heart surgery with it.

Right now, we are in a "messy middle" phase of the technology. Whether it’s helpful or frustrating usually comes down to three main friction points:

1. The "Confident Hallucination" Problem​

AI is a statistical prediction engine, not a database of facts. Because it’s designed to be helpful and fluent, it will sometimes invent a very convincing lie rather than admit it doesn't know something. If you need 100% objective truth (like legal citations or complex math) without double-checking, AI can definitely feel like it "sucks."

2. The Creative "Uncanny Valley"​

While AI can generate art, music, and text in seconds, it often lacks the "soul" or intentionality of human creation.

• The Pro: It’s great for overcoming a blank page or automating tedious tasks.
• The Con: It can result in a "slop" of generic, repetitive content that makes the internet feel less human.

3. Expectations vs. Reality​

We often treat AI like an all-knowing oracle, but it’s actually more like a very fast, slightly distracted intern.

• Where it excels: Summarizing long documents, coding boilerplate, brainstorming, and translating languages.
• Where it fails: Nuanced emotional intelligence, long-term planning, and physical world reasoning.

---

The Verdict?

If you use it as a collaborator to handle the "grunt work" of thinking, it’s a superpower. If you use it as a replacement for critical thinking, it will eventually let you down.

What’s been your biggest frustration with it so far—is it the accuracy, or just how "robotic" the output feels?

 

[TroglocratsRdumb]

AI is faster mistakes and faster misinformation
AI is faster left wing lies and propaganda.

 

[1srelluc]

I asked a question about production timeline of an AK I bought today and Copilot came up with more info than I figured. The sources were OK too.

That's a good thing since the Arsenal (Las Vegas) importer won't tell you shit.








The Arsenal SLR‑107FR was first imported into the United States in 2013, when Arsenal Inc. announced the arrival of the rifle in response to customer demand AmmoLand Shooting Sports News. This marked its return to the U.S. market after a period of absence, as the company had previously produced and sold it in earlier years.

The SLR‑107FR is a 7.62×39 mm semi‑automatic rifle built with a Bulgarian‑made 1 mm stamped receiver and a cold hammer‑forged, chrome‑lined barrel, imported through Arsenal Inc. of Las Vegas, Nevada Arsenal, Inc.+1. It was reintroduced in multiple configurations in 2013 and again in 2015 when Arsenal announced a new shipment The Firearm Blog.

Because the SLR‑107FR is a licensed import and subject to U.S. import regulations, the exact date of importation for a specific rifle can vary depending on the batch and shipment. However, the earliest confirmed U.S. importation was in 2013, with subsequent shipments in later years AmmoLand Shooting Sports News+1.

 

[The Sage of Main Street]

Phone calls routed to the wrong departments, AI programs posting payments to the wrong accounts in business ledgers, secure emails forwarded to the wrong addresses, Insurance companies automatically billed for medical services with the wrong dates of service - The list of incompetence is almost endless!

Good news for me. My medical billing company is racking up lots of billable hours fixing these mini disasters for clinics, but AI is just 1 reason healthcare costs are so high in America. Maybe one day this AI stuff will work like it's supposed to, but right now it's (As the Shakespeare play said) A Comedy of Errors.

Why doesn't AI work very well and what can we do about it?

Despite billions of dollars invested in ever-larger models, problems with AI hallucinations persist. So what's going wrong, and is there a better path forward? Professor Artur d'Avila Garcez of City St George's, University of London, explores.

www.bcs.org

Beware of Geeks Bearing Gifts