You have to be good at prompt engineering when it comes to single letter problems.
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
ChatGPT Plus Users Can Keep Using GPT-4o After Complaints About GPT-5
- Thread starter MacRumors
- Start date
- Sort by reaction score
You have to be good at prompt engineering when it comes to single letter problems.
The hype of "AI" overshadows and distorts the actual usefulness and capabilities of the underlying models. Sam Altman isn't delusional about the actual capabilities of AI, he just knows that he has to continue to make this all sound like it is headed toward some inflated form of magic in order to continue to extract money from investors. The term everywhere people keep parroting is "superintelligence" like AGI is just going to self-manifest/evolve from one of models any minute now, which is very unlikely.
Apple decided 10 years ago they were not going to follow the business/enterprise space into the cloud, and so while macOS and iOS are all over the workplace these days, everyones data and operations live at Microsoft, Google or Amazon. The whole industry is wandering a little aimlessly trying to find the "killer app" that will build consumer trust and adoption of AI. A platform like Copilot or Gemini at least has the tentpoles of business, enterprise and cloud customers that Microsoft/Google can sell the real world practical applications of AI, that don't necessarily sell the "magic" to everyday consumers, while they build and experiment. So it is no wonder the only features of Apple Intelligence to materialize have been not much more than toys. Apple probably should have never put its foot in AI at this stage. They should have decided that creating their own platform was not the best way to invest in AI and left that to the players already out there like OpenAI, Google, Anthropic etc., while adding hardware and software support to make it easier for developers to integrate and build solutions with whichever of those tools they want to bring in. They could have then waited and seen where it made sense to make their own plays.
Bingo. People need to understand every LLMs' output is based on training from prior works. It is 100% impossible for it to deliver/discover novel concepts or ideas on it's own. Sure it can put words together in an entirely new/unique way, and it can pass information on to other models that are good at other analytical things and get that data back, but if you are not doing your own sanity checking and directing it with thoughts/understanding from your own head, everything it is spitting out is influenced by some other work, quite possibly fiction.
Thank you.
Damn, with so many models, I can understand why they wanted to streamline and simplify the offer to just GPT-5. And for me it’s working better than any other previous model but I guess that YMMV.
What I’ve noticed tho, and maybe this is just me, is that all the noise that the release of GPT-3 and GPT-4 got at the moment, has not been replicated with GPT-5. Maybe it isn’t perceived as that big of a leap? Or maybe there’s more competition now…
AI is still very much in the hot garbage stage. AI is like Windows 95, if Clippy ate a bucket of paint chips and ivermectin, followed by a Clorox chaser. Watching people get sucked into that dumpster fire is hilarious.
Calling LLMs “AI” is hilarious. Fancy next word prediction isn’t even close to “intelligence.”
Lol. 40 percent is generous. It’ll probably be more like 75 percent. These agents can’t even add the correct items to an Amazon checkout cart, let alone handle corporate data. What a joke.
Oh some of the changes are beyond annoying. I just want it to answer a question, keep it factual and concise because its a dam computer, but they keep programming it to sound more human like with pauses in response, umms and ahhs and throw praise at me with every answer like somehow I was looking for that. Its a narcissists dream partner the way they code it out of the box. I have to keep telling it to take a neutral tone and keep answers concise. Even my custom models I create in ChatGPT are "improved" without me wanting or asking for it.
To be fair, you just described the skill level of a large chunk of middle managers.
I‘ve missed out on the AI hype for most of 2022-2024.
I‘ve fairly recently, like two months ago, gotten a subscription for ChatGPT, after realizing that it is actually pretty useful for certain tasks, as long as you are critical of its outputs and double-check its answers.
That being said, I agree with people here that what goes for AI these days feels very much overhyped.
In my profession (Nuclear Medicine Physician and Radiologist in Germany), I‘ve been told for the better part of the last ten years that I will soon be replaced by AI. But even though some amazing progress has been made in the field, here we are in 2025 and last time I checked, I still spend ten hours a day in the office, churning out reports. And I don‘t see that change anytime soon.
As for ChatGPT, I actually like version 5 better than the previous ones, because in my case it appears to have been toned down a bit.
I‘ve fairly recently, like two months ago, gotten a subscription for ChatGPT, after realizing that it is actually pretty useful for certain tasks, as long as you are critical of its outputs and double-check its answers.
That being said, I agree with people here that what goes for AI these days feels very much overhyped.
In my profession (Nuclear Medicine Physician and Radiologist in Germany), I‘ve been told for the better part of the last ten years that I will soon be replaced by AI. But even though some amazing progress has been made in the field, here we are in 2025 and last time I checked, I still spend ten hours a day in the office, churning out reports. And I don‘t see that change anytime soon.
As for ChatGPT, I actually like version 5 better than the previous ones, because in my case it appears to have been toned down a bit.
Hey it’s “a PhD in your pocket!” Only Altman didn’t mention that stood for “pretty huge dumbass.”
This is an excellent idea though I assume you meant "should not have opinions on subjects".
No, "have opinions" is good. I've had to tell ChatGPT similar things. I don't know about ChatGPT 5 yet, but in the past it has a tendency to try to balance tone and both-sides controversial issues.
Indeed. I too find GPT-5 better. More to the point, concise and substantially less theatrical than GPT-4. For reasoning tasks it seems that Claude 4 Opus is even better, but Anthropic ask you to provide a phone number to open an account in them. Fortunately, it is available through other platforms.