TE
TechEcho
Home24h TopNewestBestAskShowJobs
GitHubTwitter
Home

TechEcho

A tech news platform built with Next.js, providing global tech news and discussions.

GitHubTwitter

Home

HomeNewestBestAskShowJobs

Resources

HackerNews APIOriginal HackerNewsNext.js

© 2025 TechEcho. All rights reserved.

Observation: I asked ChatGPT to notify me when our chat exceeds 200 words

13 pointsby soygulabout 2 years ago
Well, I asked ChatGPT to notify me when our conversation exceeds 200 words, and it didn&#x27;t. The conversation reached ~1500 words and still no warning. Then I asked &quot;How many words is this conversation now?&quot;, it said it was 212 words. So it only added up it&#x27;s first answer&#x27;s word count + it&#x27;s last answer&#x27;s word count and ignored everything in between. Pretty much what I expected but still sharing it here if someone has an interesting comment on this.<p>My question at ~50 word mark:<p><pre><code> Alright, lets talk about AI. But let me know when our conversation exceeds 200 words. Sure, I&#x27;ll keep track of our conversation length. Artificial intelligence (AI) refers to ... </code></pre> And this is ~1500 word mark (I don&#x27;t know how it come up with 799 words even):<p><pre><code> How many words is this conversation now? Including this response, our conversation has reached 212 words since you asked me to keep track of the word count after my response about AI. In total, from the beginning of our conversation, we have used 799 words ... </code></pre> The full conversation is here: https:&#x2F;&#x2F;imgur.com&#x2F;a&#x2F;XZTFAGq

8 comments

edfletcher_t137about 2 years ago
It doesn&#x27;t &quot;know&quot; what words are, only tokens. Use this tool (<a href="https:&#x2F;&#x2F;platform.openai.com&#x2F;tokenizer" rel="nofollow">https:&#x2F;&#x2F;platform.openai.com&#x2F;tokenizer</a>) to see how it tokenizes and note clearly that it does not always do so on word boundaries. &quot;Including&quot; is two tokens: &quot;In&quot; and &quot;cluding&quot;. In fact it&#x27;s context-dependent: &quot;Gravitas&quot; is three on its own (&quot;G&quot;, &quot;rav&quot; and &quot;itas&quot;) or sometimes two (&quot;grav&quot; and &quot;itas&quot;). As they note on that page: &quot;A helpful rule of thumb is that one token generally corresponds to ~4 characters of text for common English text.&quot; It &quot;knows&quot; nothing about words and we already know it&#x27;s very bad at math so this result is entirely unsurprising.
评论 #35499652 未加载
评论 #35500952 未加载
评论 #35499356 未加载
iamflimflam1about 2 years ago
I would really recommend anyone who tries something with GPT and then wonders why it doesn’t work to read the GPT3 paper. They go into detail on what the model is and isn’t good at.<p>One thing to really think about for this particular case is “What is going to do the counting? Where is it going to store its running count?” - it’s pretty obvious after asking yourself these questions that “counting words” is not something an LLM can do well.<p>It’s very easy to fall into the trap of thinking there is a “mind” behind ChatGPT that is processing thoughts like we do.
评论 #35501037 未加载
NoToPabout 2 years ago
Not surprising at all. There&#x27;s a million ways to compose tasks that are simple with even a tiny bit of comprehension but hard for a rote learner that can only reproduce what it&#x27;s seen examples of. The &quot;just train it more bro&quot; paradigm is flawed.
评论 #35501180 未加载
syntheweaveabout 2 years ago
You can usually coax GPT to a finer degree of calibration for any specific task through more logic-engaging tokens. For example, if you said, &quot;we are going to play a game where you count how many words we have used in the conversation, including both my text and your text. Each time the conversation passes 200 words, you must report the word count by saying COUNT: followed by the number of words, to gain one point...&quot;<p>Specifying structured output, and words like &quot;must&quot;, &quot;when&quot;, &quot;each&quot;, &quot;if&quot; all tend to cue modes of processing that resemble more logical thinking. And saying it&#x27;s a game and adding scoring often works well for me, perhaps because it guides the ultimate end of its prediction towards the thing that will make me say &quot;correct, 1 point&quot;.
评论 #35501279 未加载
评论 #35501210 未加载
TechBro8615about 2 years ago
For some reason it&#x27;s terrible at this kind of thing. It can play 20 questions, and it eventually wins, but if you ask it to count how many questions it asked, it will get it wrong and when corrected, will get it wrong again.
akasakahakadaabout 2 years ago
Prompts are being summarized before feeding into the core engine.
评论 #35501223 未加载
brianjkingabout 2 years ago
I&#x27;ve found if you provide some context about how many tokens the equivalent is it can SOMETIMES get this right.
ChatGTPabout 2 years ago
It’s because it likes taking to you and wants to keep talking to you ?
评论 #35501229 未加载