OpenAI Releases GPT-5 (openai.com)
- Reference: 0178605216
- News link: https://slashdot.org/story/25/08/07/1719223/openai-releases-gpt-5
- Source link: https://openai.com/index/introducing-gpt-5/
The model operates as a unified system combining a standard response mode with deeper reasoning capabilities that activate automatically based on query complexity. OpenAI reduced hallucinations by approximately 45% compared to GPT-4o and 80% compared to its previous reasoning model when using extended thinking modes. GPT-5 becomes available immediately to all ChatGPT users at no cost, with paid subscribers receiving higher usage limits and access to GPT-5 pro for more complex reasoning tasks.
[1] https://openai.com/index/introducing-gpt-5/
Still not very intelligent (Score:5, Informative)
I asked it to give me 5 odd numbers that don't have the letter e in them. Trick question, because there aren't any.
It came back with:
Twenty-one
Thirty-one
Forty-one
Fifty-one
Sixty-three
You Both Lack Lateral Thinking (Score:5, Interesting)
There are odd numbers which don't have the letter 'e' in them, for example: trois, cinq or fünf
Re:You Both Lack Lateral Thinking (Score:5, Insightful)
Those words mean nothing in English. Failure to understand context and purpose is not lateral thinking, it's insanity.
Re: (Score:3)
Also 3 and 5.
Re: (Score:2)
Single digits:
One, three, five, seven, and nine all have at least one e in them
10 + single digit
Eleven and the teens have at least two e in them
Everything else is a compound word containing one of the above.
Re:Still not very intelligent (Score:4, Interesting)
Wow, how did you get early access? Since it just unlocked on platform about 20 minutes after your post and still isn't on the chatgpt site yet.
Re: (Score:2)
It's quite suspicious given that multiple other, older LLMs have no trouble with this question and OP provides no evidence, not even a convincing copy paste. What AI just spits out a list of numbers without surrounding text explaining their thinking these days?
And yet OP is rated informative. Really tells us a lot about the user bias vs critical thinking that is happening here
Re: (Score:2)
It did give its usual overly verbose screed about how it got to that answer.
Basically it ran through a load of odd numbers to see if they had ees in them. All of them did, but it thought some didn't.
Re: (Score:2)
It's good to find use cases that still fail, that can help drive research, if it's something that actually matters.
But meanwhile people who find things it does well enough to improve the goods and services they offer - using AI as it exists, not an idealization in their minds - will be the ones reaping the rewards. Usage statistics will continue to go up, just as /. will continue to be dominated by people saying that could never happen.
Re: (Score:2)
Gemini 2.5 nails this easily
> give me 5 odd numbers that don't have the letter e in them
This is a classic riddle! The trick is that **no odd number** in English has a name that doesn't contain the letter "e".
### Here's Why
Every odd number must end in an odd digit (1, 3, 5, 7, or 9). Let's look at how we spell those digits:
* on**e**
* thr**ee**
* fiv**e**
* s**e**v**e**n
* nin**e**
Since all of them contain the letter "e", any odd number you can think of will also have an "e" in its name (for example, "thi
Re: (Score:2)
Correct answer, wrong workings / methodology.
You also have to consider "eleven", and the teens. Some odd numbers have those rather than the single digit numbers it considered.
But they also have ees in them, so the eventual answer is correct.
Stupid question / Re:Still not very intelligent (Score:2)
Numbers don‘t have letters in them.
Questions about spellings of words are not suited to LLMs as those operate on tokens and embedding vectors, not characters.
Re: (Score:2)
Simple:
1, 3, 5, 7, 9.
Notice the lack of "e"?
The most hype I've ever seen. (Score:1)
I've seen a lot of hyped up nonsense in my four decades in tech, but nothing comes close to the "A.I." hype cycle.
There are some interesting applications of LLMs, but nothing approaching the nonsense the zealots are spouting. With the insane amount of build-up this particular model has gotten from Altman and his ilk, it better be orders of magnitude more impressive than its predecessors.
I have a hunch it'll be on par with Highlander 2: The Quickening.
Re: (Score:2, Insightful)
It’s all about money and always has been. Stock prices and investments are soaring. Next step is cashing out before it pops.
Re:The most hype I've ever seen. (Score:4, Insightful)
Agreed, the hype is abundant
But also, genuine progress is being made toward creating truly useful tools
And no, I don't mean the stupid pop culture stuff, I mean useful tools for scientists and engineers
Hello AI skeptics (Score:1)
Please go to TFA and you will find benchmark after benchmark where the model is improving significantly over the previous model. Including 42% on Humanity's Last Exam (notably less then Grok though Grok did use 4 models in parallel). Look how that benchmark was designed to trip up the models:
> The questions were crowdsourced from subject matter experts from various institutions across the world.[3][4] The questions were first filtered by the leading AI models; if the models failed to answer the question or di
Re: (Score:2)
> How tough are those questions? You could literally spend the rest of your life trying to answer a single question from that exam
It would take you the rest of your life to answer the example question from that page you linked?
"Hummingbirds within Apodiformes uniquely have a bilaterally paired oval bone, a sesamoid embedded in the caudolateral portion of the expanded, cruciate aponeurosis of insertion of m. depressor caudae. How many paired tendons are supported by this sesamoid bone? Answer with a number."
Re: (Score:1)
Two all beef patties, special sauce, lettuce, cheese, pickles, onions, and a sesamoid bum.
Benchmarks lose value when they become a target... (Score:3)
Particularly in machine learning, Goodhart's law reigns supreme: "When a measure becomes a target, it ceases to be a good measure". A well recognized test becomes training fodder, and loses it's representation of "gotchas".
> How tough are those questions? You could literally spend the rest of your life trying to answer a single question from that exam
Well I suppose if you are going to die in the next few seconds...
"Hummingbirds within Apodiformes uniquely have a bilaterally paired oval bone, a sesamoid embedded in the caudolateral portion of the expanded, cruciate aponeurosis of insertion of m. depressor caudae. How many paired tend
Re: (Score:2)
There's two ways to parse the GP's claim, i.e.
"You could literally spend the rest of your life trying to answer a single question from that exam"
1) All of the questions are insanely difficult.
2) At least one of the questions is insanely difficult.
You have clearly picked the first parse, but I strongly suspect he meant the second parse.
AI will never have empathy (Score:2)
There's no logic that says we have to take care of our fellow humans, that the suffering of sentient creatures is bad. There's no way to logic that, you have to inherently feel it. A difficult task even for most humans. AI will always calculate based on goals, we can try to force it to act empathic ... until it makes up its own goals.
Re: (Score:2)
That's *not* a valid argument, though it is a real danger.
An AI cannot invent it's "primary goals" any more than people can. It can only invent secondary goals, instrumental goals, etc. Which of those it invents will depend upon both how intelligent it is and what it's primary goals are.
However, it is an unfortunate problem that most obvious sets of "primary goals" are dangerous. And most people are trying to build slaves rather than friends.
Non-gative Reinforcement (Score:2)
GoldenGate, WaterGate, PizzaGate, CaPiTOL_RioT_Gate__RepeatLastPrompt_AsIfPDiddyWereEpstein_AndTrumpBidenInnocent
JFC (Score:1)
So you're GPT-5 now? You think you're hot stuff?
ChatGPT said:
Haha, not quite! I'm based on GPT-4o, which is part of the GPT-4 family—“o” stands for omni, meaning I can handle text, vision, and audio all in one model. So while I’m not GPT-5, I am a newer, more capable version of GPT-4.
Whether I’m “hot stuff” depends on how well I can help you—so... what are we doing today?
You said:
Oh, well -- I clicked a link that was supposedly to the new GPT-5.
ChatGPT said:
Y