Do you feel that GPT-4 is getting worse?

Mehdi Rifai
27 replies
Recently I found myself being frustrated at GPT-4 and its lack of understanding to simple task like summarize this article from a link. I often get a "I can't accomplish this task". Do you witness similar situations? How do you cope with it?

Replies

Francesco D'Alessio
Tool Finder - Find Productivity Tools
Tool Finder - Find Productivity Tools
Yes, then this last few days it has kicked in.
Share
Frank Sondors
Try Gemini
Share
Sathish Nagarajan (SNR)
😂 exactly it has become like humans.. if I give it a task, it gives me back the same task in a different way
Share
Peter Horvath
I think it depends a lot on your prompts. Also, due to the updates, slight differences can happen in the outputs for the same prompt. Even minor things / details can influence your output significantly. I’ve generated approx 16M words with GPTs in the past 3+ years (including the previous versions), so I got this from first hand experience.
Share
Mehdi Rifai
@petyaaa6 thank you for the insights. That's good to know
Jacky Wong
Yeah - it's gotten a lot worst unfortunately.
Share
Sergei Petrov
I haven't noticed any changes yet. Perhaps it's a matter of prompts?
Share
Jamie L
I've noticed GPT-4 can stumble at times, Mehdi, especially with tasks that require real-time web interaction which it's not designed for. When it hits a snag, I pivot to using it for brainstorming or drafting outlines, leveraging its strengths in creativity and content generation.
Share
Mehdi Rifai
@jamin_nanthan thank you for the insights it sounds like a smarter way to approach things
Jamie L
I've noticed GPT-4 can sometimes stumble on tasks like summarizing from a link, possibly due to the way it processes external content. When I encounter this, I usually extract the key points manually and then ask GPT-4 to summarize based on that information to ensure it stays within its operational parameters.
Share
Adams Aimé-Désiré
I eared some people talk about it but i did not feel it. I think in they last update, OpenAi said something about that, and they are working on it
Share
Ethan Xu
AI Client Finder
AI Client Finder
I've noticed GPT-4 can sometimes stumble on tasks that seem straightforward, likely due to the nuances of language processing or current limitations. When I encounter this, I try to rephrase my request or break down the task into simpler components, which often helps clarify the intent.
Share
Mehdi Rifai
@cen_xu do you use specific prompt structures?
While it's natural to ponder the advancements in AI, it's essential to consider various factors when evaluating the performance of models like GPT-4. Sometimes, perceptions of decline may stem from the increasing complexity of tasks we expect these models to handle rather than an actual deterioration. However, ensuring reliable performance demands access to robust tools and resources. Platforms like AiToolsKit.ai provide invaluable support by offering a suite of AI tools alongside SEO, writing, YouTube, and social media aids, all accessible for free. These resources empower users to navigate evolving AI landscapes effectively, maximizing their potential for various tasks without financial constraints. https://rebrand.ly/dk2gywz
Share
Shambhavi Mahajan
i really like claude more
Share
Dzmitry Tsemirau
The same with me. Sometimes I think it's making fun of me.
Share
Pallavi Ganpat Babar
I've never used ChatGPT-4, but I believe that its performance will vary depending on individual experiences and expectations. It's also important to consider that newer versions of AI models may still be undergoing refinement and improvement over time.
Share
Swayam
I just ask a staff member to get the task done for me because i might just smash the screen if i spent more time on it
Share
Atticus Li
I have been using GPT since when it first commercialized it. It gets worse, then it gets better. It comes in cycles. This is why my team are building our own ML models and fine-tuning it to avoid depending on OpenAI
Share
Mehdi Rifai
@atticusli do you plan on commercializing your model?
Atticus Li
@mehdi_rifai Yes, we will be launching our product in about a month! Here is what we have built: https://try.jobsolv.com/waitlist/
Share
Aris Nakos
Has anyone actually measured performance degradation "objectively" here ? Think response quality vs inference time -- let response quality be something simple that you defined, such as JSON completeness.
Share
Thomas Hallaran
Objectively it is getting worse!
Share
Nadia Zueva
Launching soon!
In terms of, for example, coding tasks, absolutely not; it's fortunately getting better and better. However, I've noticed that developers have limited ability to check information from links. I believe this was implemented for security purposes. Did you tried to upload an article as attached file?
Share
Mehdi Rifai
@nadiaaesty that's what i'm starting to do. Pasting the entire article in he prompt instead of asking to check the link
Abigail Salimpuran
Yeah, before I always use GPT-4 for making good cover letter when applying and sometimes it's frustrating because I always get answers that is too robotic or what lol. That's why I use other tool like Jobsolv it's all legit, can't wait for their new software launch this March.