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

Retiring GPT-4o, GPT-4.1, GPT-4.1 mini, and OpenAI o4-mini in ChatGPT

openai.com

Post reply on HN
Print view View original

251–260 of 433 posts

Re: Retiring GPT-4o, GPT-4.1, GPT-4.1 mini, and OpenAI o4-mini in ChatGPT

#251

Post by Hard_Space »

QuadrupleA wrote: post #247

Been unhappy with the GPT5 series, after daily driving 4.x for ages (I chat with them through the API) - very pedantic, goes off on too many side topics, stops following system instructions after a few turns (e.g. "you respond in 1-3 sentences" becomes long bulleted lists and multiple paragraphs very quickly. Much better feel with the Claude 4.5 series, for both chat and coding.

> you respond in 1-3 sentences" becomes long bulleted lists and multiple paragraphs very quickly

This is why my heart sank this morning. I have spent over a year training 4.0 to just about be helpful enough to get me an extra 1-2 hours a day of productivity. From experimentation, I can see no hope of reproducing that with 5x, and even 5x admits as much to me, when I discussed it with them today:

> Prolixity is a side effect of optimization goals, not billing strategy. Newer models are trained to maximize helpfulness, coverage, and safety, which biases toward explanation, hedging, and context expansion. GPT-4 was less aggressively optimized in those directions, so it felt terser by default.

Share and enjoy!

Re: Retiring GPT-4o, GPT-4.1, GPT-4.1 mini, and OpenAI o4-mini in ChatGPT

#252

Post by altmanaltman »

jacquesm wrote: post #129
torben-friis wrote: post #96

Earlier quoted context omitted.

What’s the goal there? Sexting? I’m guessing age is needed to serve certain ads and the like, but what’s the value for customers?

Porn has driven just about every bit of progress on the internet, I don't see why AI would be the exception to that rule.

yeah linus was beating it constantly to porn while developing the linux kernal. its proven fact. every oss project that runs the internet was done the same way, sure.

Re: Retiring GPT-4o, GPT-4.1, GPT-4.1 mini, and OpenAI o4-mini in ChatGPT

#253

Post by dahcryn »

toxic72 wrote: post #112
europeanNyan wrote: post #86

After they pushed the limits on the Thinking models to 3000 per week, I haven't touched anything else. I am really satisfied with their performance and the 200k context windows is quite nice. I've been using Gemini exclusively for the 1 million token context window, but went back to ChatGPT after the raise of the limits and created a Project system for myself which allows me to have much better organization with Proj…

I REALLY struggle with Gemini 3 Pro refusing to perform web searches / getting combative with the current date. Ironically their flash model seems much more likely to opt for web search for info validation. Not sure if others have seen this... I could attribute it to: 1. It's known quantity with the pro models (I recall that the pro/thinking models from most providers were not immediately equipped with web search too…

when I want it to google stuff, I just use the deep research mode. Not as instant, but it googles a lot of stuff then

Re: Retiring GPT-4o, GPT-4.1, GPT-4.1 mini, and OpenAI o4-mini in ChatGPT

#254

Post by QuadrupleA »

dahcryn wrote: post #250
QuadrupleA wrote: post #247

Been unhappy with the GPT5 series, after daily driving 4.x for ages (I chat with them through the API) - very pedantic, goes off on too many side topics, stops following system instructions after a few turns (e.g. "you respond in 1-3 sentences" becomes long bulleted lists and multiple paragraphs very quickly. Much better feel with the Claude 4.5 series, for both chat and coding.

4.1 is great for our stuff at work. It's quite stable (doesn't change personality every month, and one word difference doesn't change the behaviour). IT doesn't think, so it's still reasonably fast. Is there anything as good in the 5 series? likely, but doing the full QA testing again for no added business value, just because the model disappears, is just a hard sell. But the ones we tested were just slower, or tried…

Yeah - agreed, the initial latency is annoying too, even with thinking allegedly turned off. Feels like AI companies are stapling more and more weird routing, summarization, safety layers, etc. that degrade the overall feel of things.

Re: Retiring GPT-4o, GPT-4.1, GPT-4.1 mini, and OpenAI o4-mini in ChatGPT

#255

Post by Der_Einzige »

altmanaltman wrote: post #252
jacquesm wrote: post #129

Earlier quoted context omitted.

Porn has driven just about every bit of progress on the internet, I don't see why AI would be the exception to that rule.

yeah linus was beating it constantly to porn while developing the linux kernal. its proven fact. every oss project that runs the internet was done the same way, sure.

You think RMS isn’t secretly a pervert? Just look at his comments about Epstein that got him cancled.

Unironically if they look disheveled it’s because they are indeed coomers behind closed doors.

Re: Retiring GPT-4o, GPT-4.1, GPT-4.1 mini, and OpenAI o4-mini in ChatGPT

#256

Post by deaux »

qingcharles wrote: post #224
orphea wrote: post #130

Earlier quoted context omitted.

Have you had a chance to compare with Gemini 3?

I switch routinely between Gemini 3 (my main), Claude, GPT, and sometimes Grok. If you came up with 100 random tasks, they would all come out about equal. The issue is some are better at logical issues, some are better at creative writing, etc. If it's something creative I usually drop it in all 4 and combine the best bits of each. (I also use Deep Think on Gemini too, and to me, on programming tasks, it's not really…

This is the only accurate take. Any people who claim that one of the big 3 is all around "bad" or "low quality" compared to the other two, can be ignored. They're close enough in overall "strength" yet different enough in strengths/weakness that it's very much task/domain-specific.

Re: Retiring GPT-4o, GPT-4.1, GPT-4.1 mini, and OpenAI o4-mini in ChatGPT

#257

Post by sandspar »

simonjgreen wrote: post #239
Macha wrote: post #160

Earlier quoted context omitted.

I always thought that the idea that "revealed preferences" are preferences, discounts that people often make decisions they would rather not. It's like the whole idea that if you're on a diet, it's easier to not have junk food in the house to begin with than to have junk food and not eat more than your target amount. Are you saying these people want to put on weight? Or is it just they've been put in a situation that…

Absolutely. Nicotine addiction can meet the criteria for a revealed preference, certainly an observed choice

One example I like to use is schadenfreude. The emotion makes us feel good and bad at the same time: it's pleasurable but in an icky way. So should social media algorithms serve schadenfreude? Should algorithms maximize for pleasure (show it) or for some kind of "higher self" (don't show it). If they maximize for "higher self" then which designer gets to choose what that means?

Re: Retiring GPT-4o, GPT-4.1, GPT-4.1 mini, and OpenAI o4-mini in ChatGPT

#258

Post by Der_Einzige »

IhateAI wrote: post #116
tom1337 wrote: post #68

Would be cool if they'd release the weights for these models so users could now use them locally.

Why would someone want to spend half a million dollars on GPUs and components (if not more) to run one year old models that genuinely aren't useful? You can't self host trillion parameter models unless you own a datacenter lol (or want to just light money on fire).

To do AI research!!!!!!!

Re: Retiring GPT-4o, GPT-4.1, GPT-4.1 mini, and OpenAI o4-mini in ChatGPT

#259

Post by gostsamo »

monksy wrote: post #236
estimator7292 wrote: post #162

Earlier quoted context omitted.

While this is a valid take, I feel compelled to point out Chuck Tingle. The sheer amount and variety of smut books (just books) is vastly larger than anyone wants to realize. We passed the mark decades ago where there is smut available for any and every taste. Like, to the point that even LLMs are going to take a long time to put a dent in the smut market. Humans have been making smut for longer than we've had writin…

I want smut that talks about agent based development and crawdbot to do dirty dirty things. Does that exist yet. I don't think so.

rule 34
Post reply on HN