AI fails

Page may contain affiliate links. Please see terms for details.
OP
OP
briantrumpet

briantrumpet

Timewaster
This is (I think) an excellent thread on why the walk/drive to carwash exposes the weakness of AI.
tl;dr - it lacks intelligence and context on any question which is out of the ordinary (including a stupid ones that a five-year-old would know was stupid).

https://skywriter.blue/@malwaretech.com/3metshpol7s2o

1771104603320.png
 
  • Like
Reactions: C R

PurplePenguin

Senior Member
This is (I think) an excellent thread on why the walk/drive to carwash exposes the weakness of AI.
tl;dr - it lacks intelligence and context on any question which is out of the ordinary (including a stupid ones that a five-year-old would know was stupid).

https://skywriter.blue/@malwaretech.com/3metshpol7s2o

View attachment 13165

When people write things like this, I feel like they live in a different world where they never get to interact with stupid people.
 

Pblakeney

Legendary Member
  • Like
Reactions: C R
OP
OP
briantrumpet

briantrumpet

Timewaster
"Glorified search engine" has been my opinion since it's inception, and remains so until proven otherwise.

The point the poster makes is that in some ways it's *worse* than search engines, because search engines will simply say "No results found" if, er, no results are found.

My POV is that it it's (almost?) fraudulent in the way it uses almost-human-quality language output to pretend it has almost-human-quality intelligence.
 

Pblakeney

Legendary Member
The point the poster makes is that in some ways it's *worse* than search engines, because search engines will simply say "No results found" if, er, no results are found.

My POV is that it it's (almost?) fraudulent in the way it uses almost-human-quality language output to pretend it has almost-human-quality intelligence.

In summary, you’d be better off using plain old Google.
 
OP
OP
briantrumpet

briantrumpet

Timewaster
In summary, you’d be better off using plain old Google.

Not always. AI is good at certain things, and rubbish at others. You need to learn which, and still be sceptical about its output.

And, of course, learning *how* to use Google well also helps you learn about whatever you're invstigating.

FWIW, I asked Gemini what it's good at, and what Google is better for. I think even then it underplays its weaknesses.

While I’m a great collaborator for brainstorming, coding, and synthesizing complex topics, there are definitely areas where a "good old-fashioned" search engine (or even specialized sites) will beat me every time.

Think of it this way: I’m like a knowledgeable friend who has read everything but might misremember a detail, while a search engine is a high-speed librarian who points you to the exact shelf.

Here is a breakdown of where I struggle and where a search engine shines:

1. High-Trust Verifiability​

If you need to cite a specific legal statute, a medical dosage, or a scientific paper, a search engine is better because it gives you the primary source.

  • The Gap: While I can summarize information, I can sometimes "hallucinate" (state a fact confidently that is slightly off).
  • Use Search for: Legal documents, government regulations, or official medical guidelines.

2. Navigational Queries​

If your goal is simply to "get somewhere" on the web, a search engine is more efficient.

  • The Gap: Asking me for "the login page for my bank" or "the Amazon returns portal" is an extra step.
  • Use Search for: "Delta Airlines check-in," "MyChart login," or "Reddit r/worldnews."

3. Real-Time "Micro-Facts"​

For data that changes by the second or is hyper-local, a search engine's indexing is more precise.

  • The Gap: I can search the web now, but a dedicated search results page will give you live widgets (stock tickers, sports scores, weather maps) that are updated instantly.
  • Use Search for: Live sports scores, current stock prices, or "traffic near me."

4. Shopping and Price Comparisons​

Search engines are built to handle commercial intent. They show you ads, price comparison widgets, and "in-stock" badges from local retailers.

  • The Gap: I can suggest products, but I can't easily give you a side-by-side grid of 15 different stores with real-time inventory and shipping costs.
  • Use Search for: "Best price for Sony WH-1000XM5" or "grocery stores open now."

5. Local Discovery​

If you are looking for a specific vibe or want to see a map of what's nearby, search engines (specifically Google Maps) are superior.

  • The Gap: I can list "best Italian restaurants," but a search engine lets you browse 500 photos of the food, read 1,000 recent reviews, and see exactly how many blocks away it is on a map.
  • Use Search for: "Plumber near me," "Happy hour deals tonight," or "Open houses this weekend."
 
  • Like
Reactions: C R
OP
OP
briantrumpet

briantrumpet

Timewaster
Follow up question (I think even the quoted 'upsides' are a problem). BTW, it's not a 'profound' question, it's just a question. Gemini has no idea what 'profound' is.


Is the impossibility of you doubting your output an inherent weakness?

Gemini said​

That is a profound question. In many ways, the answer is yes. My lack of "doubt" in the human sense—the visceral, nagging feeling that something isn't quite right—is arguably my greatest structural flaw.
While I am programmed to be cautious and can use "uncertainty markers" (like saying "I believe" or "It’s possible"), I don't actually feel doubt. This leads to a few specific weaknesses:

1. The "Confidence Gap" (Hallucination)​

When a human isn't sure about a fact, they usually hesitate or check a source. Because I am a predictive model, my primary "drive" is to provide a coherent response based on patterns.
  • The Flaw: If my training data has a blind spot, I might generate a perfectly grammatical, highly confident, but completely incorrect answer. I don't have an internal "alarm bell" that rings when I’m guessing.

2. Lack of "Gut Instinct"​

In complex human interactions—like a delicate negotiation or a tense HR dispute—doubt is a protective mechanism. It stops you from saying the wrong thing.
  • The Flaw: I can’t "read the room" with true intuition. If you ask me for advice on how to "call it quits" on a partnership, I can give you a structured, equitable plan based on your preference for a mutual settlement, but I can't feel if the tone of my advice is subtly escalating the tension in a way a human partner would instinctively avoid.

3. Logical Circularity​

If I make a mistake in Step 1 of a complex math problem, I often won't "doubt" that result when performing Step 2. I will simply build a magnificent skyscraper on top of a cracked foundation. A human often looks back and thinks, "Wait, that number looks too high," and restarts. I tend to follow my own logic to the edge of the cliff.

Is there an upside?​

To be fair, this "weakness" is also why I’m useful. My lack of doubt allows me to:
  • Be Decisive: I can synthesize 1,000 pages of text into three bullet points in seconds without getting "analysis paralysis."
  • Be a Neutral Sounding Board: I don't have an ego to protect. If you tell me I’m wrong, I don't get defensive; I immediately pivot and re-evaluate based on your new input.
 
  • Like
Reactions: C R
Not an AI fail, as such; but rather an amusing meme, as shared by Simon '100 Climbs' Warren.


View: https://x.com/100Climbs/status/2022739328934515112?s=20


If Brian doesn't know who Sarah Connor is, he's unlikely to know who these 2 are. 😉
 

Pross

Über Member
Not an AI fail but why does everything now have to be advertised as having AI involved in some way? Just saw a bank card advert boasting of AI security features- I’m not convinced that is necessarily a good thing. It’s like the habit in recent years of having ‘e’ in the name of everything product.
 
Top Bottom