Slow June, people voting with their feet amid this AI craze, or something else?
It’s not just that the novelty has worn off, It’s progressively gotten less useful. Any god damn question I ask gets 90,000 qualifiers and it refuses to provide any data at all. I think OpenAI is so terrified of liabilty they have significantly dumbed down it’s utility in the public release. I can’t even ask ChatGPT to provide a link to study it references, if it references anything at all rather than making ambiguous statements.
Also, ChatGPT 4 came out but is still only available to people who pay (as far as I know). So using ChatGPT 3 feels like only having access to the leftovers. When it first came out, that was exciting because it felt like progress was going to be rapid, but instead it stagnated. (Luckily interesting LLM stuff is still happening, it’s just nothing to do with OpenAI.)
Chatgpt4 has also noticeably declined in quality since it was released too. I use it less because it’s become less useful and more frustrating to use. I think openAI have been steadily gimping it trying to get their costs down and make it respond faster.
deleted by creator
Pasting an error code or cryptic log file message with a bit of context and it’s better than googling for 4 days.
I usually can find what I’m looking for unless it’s really obscure with days of searching. If something is that obscure, it seems kind of unlikely ChatGPT is going to give a good answer either.
If you know enough to sus out the obviously wrong shit it produces every once in a while.
That’s one pretty big problem. If something really is difficult/complex you likely won’t be able to tell the difference between a wrong answer from ChatGPT and one that’s correct unless it just says something obviously ridiculous.
Obviously humans make mistakes too, but at least when you search you see results in context, other can potentially call out/add context to things that might not be correct (or even misleading), etc. With ChatGPT you kind of have to trust it or not.
Yeah if it’s that hard to find gpt is just going to hallucinate some bs into the response. I use it as a stack overflow at times and often run into garbage when I’m trying to solve a truly novel problem. I’ll often try to simplify it to something contrived but mostly find the output useful as a sort of spark. I can’t say I ever find the raw code it generates useful or all that good.
It’ll often give wrong answers but some of those can contain useful bits that you can arrange into a solution. It’s cool, but I still think people are oddly enamored with what is really just a talking Google. I don’t think it’s the game changer people are thinking it is.
It’s pretty useful if you’re in a more generalist job. I mostly work in visual design, but I sometimes deal with coding and web dev. As someone with a mostly surface understanding of these things, asking gpt to explain exact things that don’t make sense in basic terms or solve basic issues is a huge time saver for me. Googling these issues usually works but takes way longer than getting a tailored response from gpt if you know how to ask.
I got it to give me a book that was still in copyright status by selectively asking for bigger and bigger quotes. Took a while. Now it seems to have cottoned on to that trick.
It’s really fucking annoying getting “As an AI language model, I don’t have personal opinions, emotions, or preferences. I can provide you with information and different perspectives on…” at the beginning of every prompt, followed by the driest, most bland answer imaginable.
It definitely has its uses but it also has massive annoyances as you pointed out. One thing has really bothered me, I asked it a factual question about Mohammed the founder of Islam. This is how I a human not from a Muslim background would answer
“Ok wikipedia says this ____”
It answered in this long winded way that had all these things like “blessed prophet of Allah”. Basically the answer I would expect from an Imam.
I lost a lot of trust in it when I saw that. It assumed this authority tone. When I heard about that case of a lawyer citing madeup caselaw from it I looked it as confirmation. I don’t know how it happened but for some questions it has this very authoritative tone like it knows this without any doubt.
I use it now and again but I couldn’t imagine paying $20+ a month for it.
I love Stable Diffusion but I really have no use for ChatGPT. I’m amazed at how good the output can be… i just don’t have a need to generate text like that. Also, OpenAI has been making it steadily worse with ‘safety’ restrictions. I find it super annoying and even insulting when Bing-Sydney is “THIS CONVERSATION IS OVER”. It’s like being chastised by facebook or twitter for being ‘violent’ when you made a joke.
The ability to generate photographs and illustrations of practically anything, though, is fantastic. My girlfriend has been flagellating me into creating a bunch of really useless crap to promote her business on social media using SD, and I actually enjoy that part. I’ve made thousands of photos of scenery.
I’m not really surprised at all, a lot of people I know wouldn’t stop talking about it for the grand total of maybe 2 weeks but then it all went quite. In fairness this is a sample of people who are all non-tech people, so I think a lot of it is just the fact they probably forgot the name of it or how to turn their computer on (definitely the case for some).
Personally I’ve abandoned ChatGPT in favor of Claude. It’s much more reliable.
I still use it sometimes, but ohhh boy it can be a wreck. Like I’ve started using the Creation Kit for Bethesda games, and you can bet your ass that anything you ask it, you’ll have to ask again. Countless times it’s a back-and-forth of:
Me: Hey ChatGPT, how can I do this or where is this feature?
ChatGPT: Here is something that is either not relevant or just does not exist in the CK.
Me: Hey that’s not right.
ChatGPT: Oh sorry, here’s the thing you are looking for. and then it’s still a 50-50 chance of it being real or fake.
Now I realize that the Creation Kit is kinda niche, and the info on it can be a pain to look up but it’s still annoying to wade through all the shit that it’s throwing in my direction.
With things that are a lot more popular, it’s a lot better tho. (still not as good as some people want everyone to believe)
I’ve been building a tool that uses ChatGPT behind the scenes and have found that that’s just part of the process of building a prompt and getting the results you want. It also depends on which chat model is being used. If you’re super vague, it’s going to give you rubbish every time. If you go back and forth with it though, you can keep whittling it down to give you better material. If you’re generating content, you can even tell it what format and structure to give the information back in (I learned how to make it give me JSON and markdown only).
Additionally, you can give ChatGPT a description of what it’s role is alongside the prompt, if you’re using the API and have control of that kind of thing. I’ve found that can help shape the responses up nicely right out of the box.
ChatGPT is very, very much a “your mileage may vary” tool. It needs to be setup well at the start, but so many companies have haphazardly jumped on using it and they haven’t put in enough work prepping it.
If you don’t mind me asking, does your tool programmatically do the “whittling down” process by talking to ChatGPT behind the scenes, or does the user still talk to it directly? The former seems like a powerful technique, though tricky to pull off in practice, so I’m curious if anyone has managed it.
Don’t mind at all! Yeah, it does a ton of the work behind the scenes. I essentially have a prompt I spent quite a bit of time iterating on. Then from there, what the user types gets sent bundled in with my prompt bootstrap. So it reduces the work considerably for the user and dials it in.
Edit: adding some more context/opinions.
I think the error that a lot of tools make is that they don’t spend enough time shaping their instructions for the AI. Sure, you can offload a lot of the work to it, but you have to write your own guard rails and instructions. You can tell it things like you would a human, and it will sometimes even fill in the gaps.
For example, I asked it to give me a data structure back that included an optional “title”. I found that if you left the title blank, ChatGPT took it upon itself to generate a title for you based on the content it wrote.
A lot of the things I got it to do took time and a ton of test iterations. I was even able to give it a list of exactly how it should structure the content it gave back. Things that I would otherwise do on the programming side, I was able to simply instruct ChatGPT to handle instead.
Ah, interesting. I myself have made my own library to create callable “prompt functions” that prompt the model and validate the JSON outputs, which ensures type-safety and easy integration with normal code.
Lately, I’ve shifted more towards transforming ChatGPT’s outputs. By orchestrating multiple prompts and adding human influence, I can obtain responses that ChatGPT alone likely wouldn’t have come up with. Though, this has to be balanced with giving it the freedom to pursue a different thought process.
What method did you use to generate only JSON? I’m using it (gpt3.5-turbo) in a prototype application, and even with giving it an example (one-shot prompting) and telling it to only output JSON, it sometimes gives me invalid results. I’ve read that the new function-calling feature is still not guaranteed to produce valid json. Microsoft’s “guidance” (https://github.com/microsoft/guidance) looks like what I need, but I haven’t got around to trying it yet.
OpenAI’s models, including its GPT series, are available via APIs and Microsoft Azure, and so a drop in ChatGPT’s website use may be due to people moving to programmatic interfaces
I feel like this is an important detail that changes the conclusion of the article: there may be a lot more end user, through 3d party apps, but the way of measuring won’t reveal it. This especially important considering that (correct me if I’m wrong) API users are paying ones !
I have a number of language models running locally. I am really liking the gpt4all install with Hermes model. So in my case i used chatgpt right up untill i had one i could keep private.
How do I get started with the models you are mentioning?
Using it for work from time to time, mostly when I have issues with HTML/CSS or some quick bash scripts. I’d probably miss copilot more. It saves a lot of time with code suggestions.
I still use free GPT-3 as a sort of high level search engine, but lately I’m far more interested in local models. I havent used them for much beyond SillyTavern chatbots yet, but some aren’t terribly far off from GPT-3 from what I’ve seen (EDIT: though the models are much smaller at 13bn to 33bn parameters, vs GPT-3s 145bn parameters). Responses are faster on my hardware than on OpenAI’s website and its far less restrictive, no “as a large language model…” warnings. Definitely more interesting than sanitized corporate models.
The hardware requirements are pretty high, 24GB VRAM to run 13bn parameter 8k context models, but unless you plan on using it for hundreds of hours you can rent a RunPod or something for cheaper than a used 3090.
What exact ones are you using and how can I use them?
This vid goes over it in better detail than I can.
On that, what would people recommend for a locally hosted (I have a graphics card) chatgpt-like LLM that is open source and doesn’t require a lot of other things to install.
(Just one CMD line installation! That is, if you have pip, pip3, python, pytorch, CUDA, conda, Jupiter note books, Microsoft visual studio, C++, a Linux partition, and docker. Other than that, it is just one line installation!)
Look into llama.cpp - it’s a single C++ program that run quantified models (basically models with some less precision - don’t need a full 64 bits for a double, really). As for models to run on it, there’s so many but I think WizardLM is pretty good.
I looked into this too and it’s pretty resource heavy. I actually had a really good conversation with Chatgpt about making a separate instance of itself locallly. It’s worth talking to it about that and some of the price options
Tried it a few times with poor results, it will eventually get better I guess.
It was in the major TV news cycle for weeks but now it’s back to normal levels I’d say. Curious onlookers without a real need have moved on.
The recent changes made it faster but near useless for coding.
GPT-4 is quite a bit better, but the subscription is expensive. I subscribe because I think it saves me quite a bit of time. I use it almost every day for things like refactoring (shorter) blocks of code, “translating” code into different languages or frameworks, or just for generating examples for completing tasks using frameworks or libraries I’m unfamiliar with.
