Teaching ChatGPT to speak my son’s invented language • Hacker News Board

Re: Teaching ChatGPT to speak my son’s invented language

#121

Post by Vespasian » Tue, Apr 11, 2023, 4:13 AM UTC

yosito wrote: post #116

famouswaffles wrote: post #9

In context learning is hands down the biggest breakthrough of LLMs. The flexibility the model displays without updating weights is genuinely mind blowing, bordering on absurd especially if you've trained other kinds of models before. See here - https://imgur.com/a/w3DAYOi from the paper - https://arxiv.org/abs/2211.09066 GPT 3.5's (4 is much much better) addition accuracy tanks after 2 digits. However, by approaching…

I've noticed if I give chat GPT an algorithm it gives me consistent results when it shows it's work but if I ask it not to show it's work and just give me the answer. It consistently gives me incorrect answers even when I ask it to follow an algorithm that I gave it. I suppose this is similar to humans and probably why my school teachers always told me to show my work, but I'm curious if this has been documented and…

LLMs don't have an inner monologue.

By their very nature they only "know" what they have written down and must infer the final answer from that token by token.

They fundamentally can't do certain things such as complex iteration or track back.

When you ask for chain of thought thinking, you allow the LLM to create a "buffer space" and break down the task into more manageable substeps thereby improving the quality of the results.

Re: Teaching ChatGPT to speak my son’s invented language

#122

Post by TimTheTinker » Tue, Apr 11, 2023, 4:19 AM UTC

kelseyfrog wrote: post #100

jacobsenscott wrote: post #81

Earlier quoted context omitted.

Chat gpt can't do these things because it doesn't know it is doing anything with a goal. It doesn't know it is playing a game for example. It doesn't know what a game is.

That's an argument for ChatGPT not being able to do anything a la, "ChatGPT cannot create text because it doesn't understand text," or ChatGPT cannot predict the next word because it doesn't know what words are." The problem is that it's as powerful a statement as "ChatGPT can't do X because ChatGPT can't do X," which is tautological true. It's fine that it can't but we don't have epistemological access to know if Ch…

Its behavior when given particular prompts is arguably an emergent phenomenon.

Re: Teaching ChatGPT to speak my son’s invented language

#123

Post by adnmcq999 » Tue, Apr 11, 2023, 4:23 AM UTC

yosito wrote: post #116

famouswaffles wrote: post #9

In context learning is hands down the biggest breakthrough of LLMs. The flexibility the model displays without updating weights is genuinely mind blowing, bordering on absurd especially if you've trained other kinds of models before. See here - https://imgur.com/a/w3DAYOi from the paper - https://arxiv.org/abs/2211.09066 GPT 3.5's (4 is much much better) addition accuracy tanks after 2 digits. However, by approaching…

I've noticed if I give chat GPT an algorithm it gives me consistent results when it shows it's work but if I ask it not to show it's work and just give me the answer. It consistently gives me incorrect answers even when I ask it to follow an algorithm that I gave it. I suppose this is similar to humans and probably why my school teachers always told me to show my work, but I'm curious if this has been documented and…

I just watched a video where the guy touches on GPT-4 limitations and one of those is simple math. He asks it some order of operations question and it outputs the correct answer but only if it does it step by step. It then apologizes and says its original incorrect answer was “a typo.”

Video: https://youtu.be/qbIk7-JPB2c

Re: Teaching ChatGPT to speak my son’s invented language

#124

Post by nearbuy » Tue, Apr 11, 2023, 4:33 AM UTC

sharkweek wrote: post #75

crdrost wrote: post #73

Wow, they asked the model to self-evaluate and it just outright cheated: He has three cats. Proposed: h’io’ngkiltrikumrikumrikumri’nguuy Correct: h’io’ngkiltri’ngkumrikumri’nguuy Points: 1 Hypothesis: N/A (Other comments observe that it accidentally compensated for this by getting the sum wrong, haha, d'oh) I have had similar problems with trying to get ChatGPT to do nontrivial things, "here are the rules for this ga…

I was trying for 20 minutes to get it to spit out all 50 state capitals with the city names in alphabetical order and it kept doing two things: 1) It'd put the list in alphabetical order by state, but it'd include all the correct capitals 2) It'd list 49 of the 50 capitals, in alphabetical order this time, but duplicating Madison, WI. I'd ask it to try and figure out what it did wrong in both cases, and it'd correctl…

Being able to sort a large list like that is really impressive given how ChatGPT works. I wonder if it's seen enough sorted lists of capitals in its training data to help it.

ChatGPT doesn't have any internal working memory. It can't think things out in its head before forming an answer. It formulates its answer one token at a time, but it only gets one pass through its network to make each token. (One token is often about 4 characters, but for simplicity I'll pretend it generates one word at a time.) To sort a list of 50 things, it has to figure out in a single pass which thing comes first. Then in the next pass, which thing comes second. And so on. There's no internal memory between passes either; it only sees what's written in the chat (including its own writing). It can't do loops or iteration internally within a pass either.

A human could do this iteratively for any number of items. You just go through the items one by one, keeping track of the earliest alphabetical item so far (using your working memory). Once you've gone through the entire list and found the first item, cross that one off and repeat the process until the entire list is sorted. ChatGPT can't do it that way.

Re: Teaching ChatGPT to speak my son’s invented language

#125

Post by M4v3R » Tue, Apr 11, 2023, 4:33 AM UTC

Name_Chawps wrote: post #79

knome wrote: post #24

Earlier quoted context omitted.

GPT-4 allows you to use 8k of context in their current beta, if you're using the chat api directly. It will be interesting ( and probably expensive, lol ) when they open it to a full 32k.

Though unfortunately it will cost like $20 per 32k completion...

More like $1,92 (32 * 0,12) for a 32k prompt, or twice that for a 32k completion. Still not cheap though.

Re: Teaching ChatGPT to speak my son’s invented language

#126

Post by tripleo1 » Tue, Apr 11, 2023, 4:58 AM UTC

elhudy wrote: post #71

zenlikethat wrote: post #69

Earlier quoted context omitted.

D2C makes a lot of sense imo because at this point it's most valuable to obtain a huge corpus of (reasonably well labeled) training data (with a side helping of being really great marketing). Just think about all the programmers who are probably rolling their eyes and correcting ChatGPT's work for OpenAI, or at least providing some details. I know it's a reflex for me, not even deliberate, to at least type something…

That's a great point. As someone who isn't really a programmer I wouldn't have seen the value in that feedback until you pointed it out.

thumbs up

Re: Teaching ChatGPT to speak my son’s invented language

#127

Post by tripleo1 » Tue, Apr 11, 2023, 4:59 AM UTC

z3c0 wrote: post #77

elhudy wrote: post #65

Earlier quoted context omitted.

In-context learning also seems like the best path to commercializing LLMs. I'm surprised that Microsoft is going the D2C route with ChatGPT rather than commercializing it in a B2B fashion. ...Or maybe that's coming? Imagine feeding an LLM a ton of disparate data sources and asking it questions about that data as a whole. What is a data engineer again, anyway?

It's still not an extractive tool (it's generative), so it stating logical facts based on the data is not something one should bank on. It will likely never be capable of what you're suggesting, without pairing it with an actually-extractive model. The GPT-4 paper outright states that it should not be used for sensitive or high-stakes matters, which is what a lot of enterprises would consider their data pipelines.

Wikipedia vs Brittanica

Re: Teaching ChatGPT to speak my son’s invented language

#128

Post by isoprophlex » Tue, Apr 11, 2023, 5:18 AM UTC

graypegg wrote: post #108

Unrelated to the actual work here (which is interesting!), I just wanted to mention how much of an awesome impact you must be having on your son. You two seem like such a fun curious pair! I spent a lot of time tinkering away alone in my room as a kid and that sucked. The opportunities you’re giving him now to explore are going to inspire a lifetime of creativity. I know you don’t need someone on the internet to tell…

I second that. I remember finding a manual for programming BASIC in the library once. Overjoyed I was. Then, I spent an entire afternoon trying to type BASIC commands... into the MS-DOS prompt.

If only someone could have told me to go find QBASIC.EXE first, haha

Re: Teaching ChatGPT to speak my son’s invented language

#129

Post by kolinko » Tue, Apr 11, 2023, 5:33 AM UTC

elhudy wrote: post #65

famouswaffles wrote: post #9

In context learning is hands down the biggest breakthrough of LLMs. The flexibility the model displays without updating weights is genuinely mind blowing, bordering on absurd especially if you've trained other kinds of models before. See here - https://imgur.com/a/w3DAYOi from the paper - https://arxiv.org/abs/2211.09066 GPT 3.5's (4 is much much better) addition accuracy tanks after 2 digits. However, by approaching…

In-context learning also seems like the best path to commercializing LLMs. I'm surprised that Microsoft is going the D2C route with ChatGPT rather than commercializing it in a B2B fashion. ...Or maybe that's coming? Imagine feeding an LLM a ton of disparate data sources and asking it questions about that data as a whole. What is a data engineer again, anyway?

It was released as a B2B, almost for a year - only playground and an API.

It had limited impact, and only when they built chatgpt they gained significant traction.

(I know, davinci3 was not instruct-optimised, but anyone could’ve built something close to a chat before them)

Re: Teaching ChatGPT to speak my son’s invented language

#130

Post by kolinko » Tue, Apr 11, 2023, 5:34 AM UTC

z3c0 wrote: post #77

elhudy wrote: post #65

Earlier quoted context omitted.

In-context learning also seems like the best path to commercializing LLMs. I'm surprised that Microsoft is going the D2C route with ChatGPT rather than commercializing it in a B2B fashion. ...Or maybe that's coming? Imagine feeding an LLM a ton of disparate data sources and asking it questions about that data as a whole. What is a data engineer again, anyway?

It's still not an extractive tool (it's generative), so it stating logical facts based on the data is not something one should bank on. It will likely never be capable of what you're suggesting, without pairing it with an actually-extractive model. The GPT-4 paper outright states that it should not be used for sensitive or high-stakes matters, which is what a lot of enterprises would consider their data pipelines.

Plugins are extractive. Embeddings as well, to an extent.