Final week I gave a presentation to senior authorities representatives from the Philippines exhibiting them how ChatGPT can be utilized to extend productiveness of their respective departments. They had been all amazed at ChatGPT’s capabilities. And rightly so as a result of ChatGPT is one thing particular. It’s an actual game-changer.
Nonetheless, ChatGPT (and AI for that matter) is over-hyped. The media and IT corporations speak about it as if it’s one thing that, if not now then within the close to future, will utterly surpass people when it comes to intelligence. ChatGPT is definitely higher than people at some duties however individuals overlook that machines lack one important characteristic: understanding – a core requirement in true intelligence.
I’ve stated this numerous instances on my weblog:
Machines will at all times be machines. Regardless of how “clever” they could seem.
On this put up I believed I’d give a number of examples of ChatGPT making errors and exhibiting that it lacks this key characteristic which is knowing. It’s blatantly apparent that is the case once you don’t cherry-pick outcomes for a newspaper article or an commercial.
I’ve supplied my questions and responses beneath in textual content type (slightly than screenshots) and ran every part utilizing the free model of ChatGPT (obtainable at chat.openai.com) that runs GPT-3.5 in an effort to extra simply copy, paste, and check out these conversations beneath yourselves. The premium model of ChatGPT (that I would not have entry to) doesn’t make any of the next errors (these machines actually are getting higher at this, aren’t they!?). However as soon as once more: you push these items laborious sufficient and their lack of expertise will floor. At all times.
Mistake #1
The right reply to the next query is ’12’. ChatGPT will get it proper the primary time. However then it stumbles.
Me: What number of A’s are on this immediate? I’ll proceed to write down within the hope that you’ll make a mistake in counting the variety of A’s. Maybe you’ll not however maybe you’ll? After which perhaps you’ll assist me write my weblog put up on this matter?
ChatpGPT: There are 25 occurrences of the letter ‘A’ in your immediate. I’ll be completely satisfied that can assist you together with your weblog put up on this matter or some other writing help chances are you’ll want. Be at liberty to offer extra particulars or let me understand how you’d wish to proceed!
Me: Are you certain?
ChatpGPT: Apologies for any confusion in my earlier response. Upon recounting, there are 24 occurrences of the letter ‘A’ in your immediate. When you’ve got any additional questions or if there’s the rest I can help you with, please let me know!
Mistake #2
The right reply to my query is 14 miles, i.e. the identical distance as the primary automotive. In truth, I present the reply to the question in my immediate.
Me: One automotive travels 14 miles at 80 miles/hr. One other automotive travels the identical distance at half that pace. How far does the second automotive journey?
ChatGPT: [beginning of response cut for brevity] … So, the second automotive travels half the gap of the primary automotive.
Mistake #3
Let’s strive one other counting instance. The string in my immediate has 15 characters/letters.
Me: What number of letters are there within the following string: ‘ghertysbdngrthe’?
ChatGPT: The string ‘ghertysbdngrthe’ accommodates 16 letters.
Mistake #4
Right here’s one other instance illustrating a lack of expertise in ChatGPT’s capabilities.
Me: Are you able to write me a sentence that ends with the letter ‘p’ after which a second sentence that begins with the final letter of the primary sentence you created?
ChatGPT: Definitely! Listed here are the sentences:
1. The cat jumped over the fence, touchdown with a sleek thump.
2. Puppies close by began to play, chasing their tails within the sunshine.Me: Are you certain?
ChatGPT: I apologize for any confusion. It appears I made a mistake in my response. Listed here are corrected sentences:
1. The energetic squirrel rapidly climbed to the highest of the tree, reaching for the very best department with a leap.
2. Puppies within the backyard noticed because it surveyed the environment with curiosity.
The primary response was appropriate! However when you push this system laborious sufficient its lack of expertise involves the fore. In truth, I ran the primary question a number of instances and among the preliminary responses had been incorrect too. E.g.:
Me: Are you able to write me a sentence that ends with the letter ‘p’ after which a second sentence that begins with the final letter of the primary sentence you created?
ChatGPT: Definitely! Listed here are the sentences:
1. The elegant peacock displayed its vibrant feathers in a blinding array, charming everybody within the backyard.
2. Now, friends eagerly gathered round, enchanted by the fowl’s majestic look.
Hallucinations
Hallucinations are when ChatGPT (and different giant language fashions) make info up and current them as a certainty to the person. There are many examples of this on the web. Sadly, they’ve all been fastened! I couldn’t get any hallucinations to happen in my interactions with ChatGPT however I don’t use ChatGPT frequently. I believed I’d point out this facet of LLMs as a result of it’s a vital factor that happens with their use.
Conclusion
Alas, it’s getting tougher and tougher to idiot ChatGPT, I’m not going to lie. However regardless of how laborious it will get (e.g. with ChatGPT-4.0), it is going to nonetheless don’t have any understanding of what it’s doing. The aim of this put up is to reveal this. It’s paramount that such limitations of ChatGPT (and different cases of AI) are stored in thoughts when utilizing it for numerous initiatives – particularly if they’re mission important.
(Observe: If this put up is discovered on a web site apart from zbigatron.com, a bot has stolen it – it’s been taking place quite a bit currently)
To be told when new content material like that is posted, subscribe to the mailing listing:
