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

GPT-5

openai.com

Post reply on HN
Print view View original

421–430 of 1001 posts

Re: GPT-5

#421

Post by tamimio »

highfrequency wrote: post #331

It is frequently suggested that once one of the AI companies reaches an AGI threshold, they will take off ahead of the rest. It's interesting to note that at least so far, the trend has been the opposite: as time goes on and the models get better, the performance of the different company's gets clustered closer together. Right now GPT-5, Claude Opus, Grok 4, Gemini 2.5 Pro all seem quite good across the board (ie the…

Because AGI is a buzzword to milk more investors' money, it will never happen, and we will only see slight incremental updates or enhancements yet linear after some timr just like literally any tech bubble since dot com to smartphones to blockchain to others.

Re: GPT-5

#423

Post by hodgehog11 »

demirbey05 wrote: post #50

Seems LLMs really hit the wall.

Not really, it's just that our benchmarks are not good at showing how they've improved. Those that regularly try out LLMs can attest to major improvements in reliability over the past year.

Re: GPT-5

#424

Post by mritterhoff »

tamimio wrote: post #421
highfrequency wrote: post #331

It is frequently suggested that once one of the AI companies reaches an AGI threshold, they will take off ahead of the rest. It's interesting to note that at least so far, the trend has been the opposite: as time goes on and the models get better, the performance of the different company's gets clustered closer together. Right now GPT-5, Claude Opus, Grok 4, Gemini 2.5 Pro all seem quite good across the board (ie the…

Because AGI is a buzzword to milk more investors' money, it will never happen, and we will only see slight incremental updates or enhancements yet linear after some timr just like literally any tech bubble since dot com to smartphones to blockchain to others.

You think AGI is impossible? Why?

Re: GPT-5

#425

Post by yahoozoo »

Workaccount2 wrote: post #232
doctoboggan wrote: post #61

Watching the livestream now, the improvement over their current models on the benchmarks is very small. I know they seemed to be trying to temper our expectations leading up to this, but this is much less improvement than I was expecting

Sam said maybe two years ago that they want to avoid "mic drop" releases, and instead want to stick to incremental steps. This is day one, so there is probably another 10-20% in optimizations that can be squeezed out of it in the coming months.

He said that because even then he saw the writing on the wall that LLMs will plateau.

Re: GPT-5

#426

Post by nluken »

jdoe1337halo wrote: post #122

Lmao GPT-5 is still riddled with em dashes. At least we can still identify AI generated text slop for now

The em dash isn't just the present state of AI slop— it's the future!

Re: GPT-5

#427

Post by ricardobayes »

kybernetikos wrote: post #184

ChatGPT5 in this demo: > For an airplane wing (airfoil), the top surface is curved and the bottom is flatter. When the wing moves forward: > * Air over the top has to travel farther in the same amount of time -> it moves faster -> pressure on the top decreases. > * Air underneath moves slower -> pressure underneath is higher > * The presure difference creates an upward force - lift Isn't that explanation of why wings…

To me, it's weird to call it "PhD-level". That, to me, means to be able to take in existing information on a certain very niche area and able to "push the boundary". I might be wrong but to date I've never seen any LLM invent "new science", that makes PhD, really PhD. It also seems very confusing to me that many sources mention "stone age" and "PhD-level" in the same article. Which one is it?

People seem to overcomplicate what LLM's are capable of, but at their core they are just really good word parsers.

Re: GPT-5

#428

Post by hodgehog11 »

modeless wrote: post #257

The reduction in hallucinations seems like potentially the biggest upgrade. If it reduces hallucinations by 75% or more over o3 and GPT-4o as the graphs claim, it will be a giant step forward. The inability to trust answers given by AI is the biggest single hurdle to clear for many applications.

Agreed, this is possibly the biggest takeaway to me. If true, it will make a difference in user experience, and benchmarks like these could become the next major target.

Re: GPT-5

#429

Post by strongpigeon »

highfrequency wrote: post #331

It is frequently suggested that once one of the AI companies reaches an AGI threshold, they will take off ahead of the rest. It's interesting to note that at least so far, the trend has been the opposite: as time goes on and the models get better, the performance of the different company's gets clustered closer together. Right now GPT-5, Claude Opus, Grok 4, Gemini 2.5 Pro all seem quite good across the board (ie the…

I think this is because of an expectation of a snowball effect once a model becomes able to improve itself. See talks about the Singularity.

I personally think it's a pretty reductive model for what intelligence is, but a lot of people seem to strongly believe in it.

Re: GPT-5

#430

Post by belter »

highfrequency wrote: post #331

It is frequently suggested that once one of the AI companies reaches an AGI threshold, they will take off ahead of the rest. It's interesting to note that at least so far, the trend has been the opposite: as time goes on and the models get better, the performance of the different company's gets clustered closer together. Right now GPT-5, Claude Opus, Grok 4, Gemini 2.5 Pro all seem quite good across the board (ie the…

Nobody seems to be on the path to AGI as long as the model of today is as good as the model of tomorrow. And as long as there are "releases". You don't release a new human every few months...LLMs are currently frozen sequence predictors whose static weights stop learning after training.

They lack writable long-term memory beyond a context window. They operate without any grounded perception-action loop to test hypotheses. And they possess no executive layer for goal directed planning or self reflection...

Achieving AGI demands continuous online learning with consolidation.

Post reply on HN