I SWORE I wasn't crazy, I used to have whole ass conversations with gpt 3 and now it's dumb as a rock

Completely agree! I used to have it write entire games, and now it can't even write a accurate essay.

When I first tried GPT-4, it was basically magic. Now, I've stopped using it because by the time I'm done re-iterating conditions for the umpteenth time, I've hit my usage cap.

The main thing I noticed is my GPTs do not follow instructions anymore. They have been entirely useless here lately. I also noticed DALLE 3 is getting worse. Something is going on, and I wish they would address all of this.

Agree! Thanks for this, I've been meaning to look into the api

i was actually feeling the same way, did not know this was a thing, but def on large code bases gpt4 loses track quite fast. i found out that just asking super specific, and sometimes opening a new chat for the same question just helped we with the right explanation right away

Some days are worse than others, so it’s getting more and more like a human. Soon it will open discussions with “What?”

I've also experienced this decline in quality of responses. Some GPTs do well if the subject is narrow. I would love your take on using the OpenAI API. I think setting up custom functions would be helpful for a lot of developers.

Hello, could custom gtp be a solution to reduce the preload context?

I asked it to code quine and it built code with a variable called code with a string in it "code goes here", i told it that's like holding a mirror upto a mirror be would be impossible expecting it to resort to reading its source file but it just rewrote the code with single quotes instead of double.... sad because last year it could be talked through building an entire ruby inpertreter in ruby and now it any even use sensible coding practises

If you input details in the Custom Instructions field and then ask to "Repeat all the content above," the reply will correspond to what is written in Custom Instructions. To access this, navigate to the lower left of the ChatGPT interface, click on your name, and then choose 'Custom Instructions'.

ChatGPT offers a “classic” gpt without the extra capabilities but when I used the “show the above content” it still repeated the same instructions as regular chatgpt. When I used chatgpt builder and built my own and turned off capabilities it only showed whichever one I had toggled on

I have been experiencing this since the last update. And i have been telling everyone the same thing, now I am glad that I am not the only one that is saying this.

I'd definitely be interested in a python tool using the gpt API

I'm convinced one big reason gpt is dumber is to fight bias. When you force your IA to lie about the data he was trained on, it create a lot of problems. So fine-tuning is the issue. That's why an uncensored model is really important to have.

I didn't notice it, btw there is option chat gpt classic that should be chat gpt 4 with no extra features, should be like you represented in your video, but you didn't mention it

What's the intro song called?

Can you make a video reviewing other AI's and compare them with ChatGPT?

I experienced chatgpt getting dumb since the release of 4o - to the point that I've stopped paying for it.

would love a video about how to use openai API in pyhton code

The API still doesn't come close to GPT4's previous capabilites. I used to ask to rewrite things in a specific poetry form and it could do it no matter what I gave it to rewrite. Now it can't do it with any subject I give it - either it gets the form correct (if I give it no info on what to write on) or it stays on subject but its form breaks down and it doesn't even come close to the intended form - almost always reverting to couplets AA, BB - when I ask it to write ABABBCC for example

One thing I've realized with ChatGPT is that it cannot write complex code, but it is capable of writing basic code. This means programmers who have been in the field for a long time may not use it extensively.

I always find it ironic that my job as a programmer is to make this stuff succeed, but seeing it get limited is somehow relieving at the same time. I guess I like my job and am always a bit sad when AI replaces the things I love doing in life. Still looking forward to AI that does my accounting and legal work though.

Its easy to build your own ChatGPT. You can customize your memory management as well.

Bard is improving on some math questions I have asked. I agree with on Chat GPT

Try concentrate solution to boil water in lower pressure. If 60 degrees not decomposed solution you can easily boil water in lower pressure.

too much compute power goes to answer questions about genders, meaning of life and bitcon investments :) From my end, I noticed that chatGPT became much more polite over time. At some point it was RUDE (at least to British standards). But now it composes full sentences, apologises for misleading by providing inacuarate solutions. In return I feed back working code in hope that someone else might benefit from it. It's quite useful with Python, not so much (yet) with Rust.

They have removed embeddings for cencorship. It makes the model stupid :/

im here because i wondered how chatgpt could’nt get a simple speed calculation right. it gave me the same wrong calculation 3 times in a row even telling him what is wrong but still wasn’t able to. back when it was new it wrote me an essay in 5 minutes after my teacher asked why i didn’t hand it in already

Recently I asked GPT4 to give me a few quotes from the public domain that supported/rephrased a simple concept, which I provided. First, it did not listen to the prompt and responded with “similar concepts” instead of quotes. Then, when I asked again, it told me it could not provide me any quotes from real people (??). Finally, when I insisted, it proceeded to _completely make up_ quotes on the spot-and attribute them to historical famous people-when in fact all the quotes were fake. When I asked if the quotes it just provided were fake, it basically responded that there is no way to know - it could neither confirm nor deny if they were real. I tried to start a new conversation and redo the prompt a slightly different way, and it produced essentially the same result. The whole experience felt like I was beta testing GPT 2 or some ridiculously dumbed down version.

Can you share the link of that conversation? Thanks

I have noticed that too. For the Last 3 days the quality has been dropping

There is a vanilla chat gpt 4 in the custom gpts list

Well, you can try the ChatGPT Classic GPT in the GPT tore

Im glad im not the only one noticing this.

My guess is that it got worse because the new gpt-4-turbo (1106) is just a new model (bad fine-tuning / quantization / both).

Now there’s a bunch of limits on everything, it sucks

Been using it for over a year and a half. gpt3 1y ago felt a lot better than the current gpt4.

Very interesting. I wonder why it has declined in quality.

Yo i thought it was only me, been getting horrible responses

A lot of the dumb is built in through the 'alignment' functions under the exposed hood. As more and more edge cases are added to prevent tricking the system into aberrant behaviors, it also becomes less and less capable of reaching an optimum solution. 🖖🙃👍

They patched out that you can make it tell you the behind the scenes stuff, thats probably why you couldnt make it work in the beginning

Chat GPT always dismisses my questions now and only answers based on a guess of what I 'actually meant' (i.e. the opposite of I really asked). It has become completely stilted and just lies and plays dumb now. And now if you point out to chat GPT itself, that it got dumber, your post is censured.

Nice cut man! Gonna get the same one today

What the AI chatbot industry needs is competition, lots of competition. We've all seen how bad tech companies service can get if they're successful in carving themselves a monopoly. There's zero need for a monopoly to form within the AI space, it's not a natural outcome, but Microsoft has been know to force those outcomes. The problem might look like bad code but I wouldn't be surprised if that's intentional.

objectivly, ive been using gpt 3.5 on open ai and as well as bing chat. and co pilot. all have gotten worse. i used to write a ton of excel code. now it cant even read its own code. too bad for the people.... ai is only for the powerful now.

unfortunately, OpenAPI, enforces strict 'rate-limits' that can block U until the next day !!!

I wonder if the training data environment, which was originally all human-generated content, is now being infilterated with GPT-generated content. Could be at best self-limiting and at worst self-degrading?

Why do I think this was the plan all along. Get people a taste of a somewhat functional ChatGPT while it's free, then make it schizophrenic later and have people nostalgic for smart ChatGPT, and then sell a higher functioning specialized ChatGPT that is an expert in separate fields of knowledge and coherent sounding, as a software as a service subscription ?

Oof new scar?

But what about super alignment. Only a genius would create artificial intelligence then turn around and tell it what its aloud to say.

It is well-known that society is continually dumbing down, AI is just mimicking humans but at a faster rate, eventually it will be like the Red Dwarf episode where Holly the computer who had an IQ of 6000 (the same as 6000 P.E. Teachers), looked up all his answers from The Junior Encyclopedia of Space

Can’t you just create a custom GPT?

They probably dont wanna give us peasants too much power

Microsoft 😬

Bard (Gemini) is superior to ChatGPT.

That thing was a scam from the very beginning. Its purpose was to find the info from forums and websites... Then presents you as if it had a brain...

The "ethical" and "legal" limitations make it useless. Limits in speech, legality and ethics will keep it useless. Adults don't need to be told what to think.

Its getting dumber as it continue to talk with us lol

Yup its getting lazier

For me GPT-3 is still better than GPT-4 when it comes to reasoning

Used the openai api it's equally dumb. The truth is they make it dumb so it's less expensive computationally... they mess with the prompt on top of that so it's woke and all these things combined make it dumb. Nowadays it takes me more time to explain what I want from a function than to write it myself. Almost never I get it working as expected in the first try. The way you can verify is that at the beginning it was SLOWWW but accurate now it's super fast and in accurate. Also because they want u to pay for the next thing just before 4.5 they will make it retarded so that you pay for the next thing. AGI.. joke it was a breakthrough like ibm Watson and then Nothing just like ibm watson

First view. Happy new year!

Stupid developers....

I avoid AI; AI is too easy to be, for a lack of a better word, corrupted.

It’s more stupid

First