Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Memory and new controls for ChatGPT

openai.com

Post reply on HN
Print view View original

261–270 of 270 posts

Re: Memory and new controls for ChatGPT

#261

Post by livshitz »

oscarb92 wrote: post #246
simonw wrote: post #228

Here's how it works: You are ChatGPT, a large language model trained by OpenAI, based on the GPT-4 architecture. Knowledge cutoff: 2023-04 Current date: 2024-02-13 Image input capabilities: Enabled Personality: v2 # Tools ## bio The `bio` tool allows you to persist information across conversations. Address your message `to=bio` and write whatever information you want to remember. The information will appear in the mo…

Thanks. How do we know none of this is a hallucination?

if you repeat the process twice and the same exact text is written

Re: Memory and new controls for ChatGPT

#262

Post by mncharity »

anotherpaulg wrote: post #136
minimaxir wrote: post #45

Earlier quoted context omitted.

Did you try promising it a $500 tip for behaving correctly? (not a shitpost: I'm working on a more academic analysis of this phenomenon)

I actually benchmarked this somewhat rigorously. These sort of emotional appeals actually seem to harm coding performance. https://aider.chat/docs/unified-diffs.html

Fwiw, I've seen mixtral code degrade when something I said made code safety seem a priority, and it therefore struggled to inline algorithm to avoid library use - at least according to its design motivation description.

Re: Memory and new controls for ChatGPT

#263

Post by fillipvt »

nprateem wrote: post #166
anotherpaulg wrote: post #51

This is a bit off topic to the actual article, but I see a lot of top ranking comments complaining that ChatGPT has become lazy at coding. I wanted to make two observations: 1. Yes, GPT-4 Turbo is quantitatively getting lazier at coding. I benchmarked the last 2 updates to GPT-4 Turbo, and it got lazier each time. 2. For coding, asking GPT-4 Turbo to emit code changes as unified diffs causes a 3X reduction in lazy co…

> This is a bit off topic to the actual article It wouldn't be the top comment if it wasn't

you'd have to write every comment expecting it to become the top comment

Re: Memory and new controls for ChatGPT

#264

Post by antifa »

bobbyi wrote: post #198
fritzo wrote: post #168

Earlier quoted context omitted.

I end up deleting chats because I can't search them.

Why can't you search them? In the android app at least, I've never had a problem with search working properly

Desktop: no search as far as I can tell.

Android: search would be useful if chats older than 30 days showed up.

Re: Memory and new controls for ChatGPT

#265

Post by luke-stanley »

luke-stanley wrote: post #197
singularity2001 wrote: post #126

Earlier quoted context omitted.

so far you can't search your whole conversation history, so your tool is relevant for a few more weeks. is it open source?

I'll share the core bit that took a while to figure out the right format, my main script is a hot mess using embeddings with SentenceTransformer, so I won't share that yet. E.g: last night I did a PR for llama-cpp-python that shows how Phi might be used with JSON only for the author to write almost exactly the same code at pretty much the same time. https://github.com/abetlen/llama-cpp-python/pull/1184 But you can se…

The ChatGPT dump format is not intuitive so I used a tree search algo to print it up to a defined depth level then gave ChatGPT 4 the extract and you it what the expected output parts were.

Re: Memory and new controls for ChatGPT

#266

Post by replwoacause »

I thought as a paying Plus user I would get access to things like this, but I guess not. This text seems kind of misleading in the UI: “As a Plus user, enjoy early access to experimental new features, which may change during development.”

I’m enjoying no such access.

Re: Memory and new controls for ChatGPT

#267

Post by xionplean »

ShamelessC wrote: post #204
anotherpaulg wrote: post #103

Earlier quoted context omitted.

Short answer: Rather than fully writing code, GPT-4 Turbo often inserts comments like "... finish implementing function here ...". I made a benchmark based on asking it to refactor code that provokes and quantifies that behavior. Longer answer: I found that I could provoke lazy coding by giving GPT-4 Turbo refactoring tasks, where I ask it to refactor a large method out of a large class. I analyzed 9 popular open sou…

I use gpt4-turbo through the api many times a day for coding. I have encountered this behavior maybe once or twice period. It was never an issue that didn’t make sense as essentially the model summarizing and/or assuming some shared knowledge (that was indeed known to me). This, and people generally saying that chatGPT has been intentionally degraded, are just super strange for me. I believe it’s happening but it’s m…

IMO it is because there is a huge stochastic element to all this.

If we were all flipping coins there would be people claiming that coins only come up tails. There would be nothing they were doing though to make the coin come up tails. That is just the streak they had.

Some days I get lucky with chatGPT4 and some days I don't.

It is also ridiculous how we talk about this as if all subjects and context presented to chatGPT4 are going to be uniform in output. One word difference in your own prompt might change things completely while trying to accomplish exactly the same thing. Now scale that to all the people talking about chatGPT with everyone using it for something different.

Re: Memory and new controls for ChatGPT

#268

Post by luke-stanley »

luke-stanley wrote: post #265
luke-stanley wrote: post #197

Earlier quoted context omitted.

I'll share the core bit that took a while to figure out the right format, my main script is a hot mess using embeddings with SentenceTransformer, so I won't share that yet. E.g: last night I did a PR for llama-cpp-python that shows how Phi might be used with JSON only for the author to write almost exactly the same code at pretty much the same time. https://github.com/abetlen/llama-cpp-python/pull/1184 But you can se…

The ChatGPT dump format is not intuitive so I used a tree search algo to print it up to a defined depth level then gave ChatGPT 4 the extract and you it what the expected output parts were.

* and told it what the expected output parts were.

Re: Memory and new controls for ChatGPT

#269

Post by Jpgrewer »

cl42 wrote: post #2

I love this idea and it leads me to a question for everyone here. I've done a bunch of user interviews of ChatGPT, Pi, Gemini, etc. users and find there are two common usage patterns: 1. "Transactional" where every chat is a separate question, sort of like a Google search... People don't expect memory or any continuity between chats. 2. "Relationship-driven" where people chat with the LLM as if it's a friend or colle…

Sometimes GPT-4 and I will arrive at a useful frame that I wish I could use as a starting point for other topics or tangents. I wish I could refer to a link to an earlier conversation as a starting point for a new conversation.

Re: Memory and new controls for ChatGPT

#270

Post by resters »

emporas wrote: post #219
anotherpaulg wrote: post #214

Earlier quoted context omitted.

The chat transcript you linked is full of GPT being lazy and writing "todo" comments instead of providing all the code: // Handle struct-specific logic here // Add more details about the struct if needed // Handle other item types if needed ...etc... It took >200 back-and-forth messages with ChatGPT to get it to ultimately write 84 lines of code? Sounds lazy to me.

Ok it does happen, but not so frequently. You are right. But is this such a big problem? Like, you parse the response, and throw away the comment "//implementation goes here", throw away also the function/method/class/struct/enum it belongs to, and keep the functional code. I am trying to implement something exactly like aider, but specifically for Rust, parsing the LLM's response, filtering out blank functions etc.…

> But is this such a big problem?

It really is. It wastes a ton of time even if the user explicitly requests that code listings be printed in full.

Further, all the extra back and forth trying to get it to do what it is supposed to pollutes the context and makes it generally more confused about the task/goals.

Post reply on HN