GPT-5 • Hacker News Board

Re: GPT-5

#271

Post by sharkjacobs » Thu, Aug 07, 2025, 5:47 PM UTC

The upgrade from GPT3.5 to GPT4 was like going from a Razr to an iPhone, just a staggering leap forward. Everything since then has been successive iPhone releases (complete with the big product release announcements and front page HN post). A sequence of largely underwhelming and basically unimpressive incremental releases.

Also, when you step back and look at a few of those incremental improvements together, they're actually pretty significant.

But it's hard not to roll your eyes each time they trot out a list of meaningless benchmarks and promise that "it hallucinates even less than before" again

Re: GPT-5

#272

Post by Telemakhos » Thu, Aug 07, 2025, 5:47 PM UTC

bmau5 wrote: post #36

It's very interesting how memetic the language around different models is. Elon seems to have coined "PhD level intelligence in all topics" and now Sam repeated it in his presentation. Despite it not having an actual meaning. I think OpenAI will coin they've achieved AGI first (as they have incentives to based on the rumored contract with MSFT), and then everyone will claim we've achieved it.

As a fairly dumb person with a PhD, I can attest that a degree means perseverance, not intelligence.

Re: GPT-5

#273

Post by wg0 » Thu, Aug 07, 2025, 5:48 PM UTC

When they say "improved in XYZ", what does that mean? "Improved" on synthetic benchmarks is guaranteed to translate to the rest of the problem space? If not that, is there any guarantees of no regressions?

Re: GPT-5

#274

Post by jazzyjackson » Thu, Aug 07, 2025, 5:48 PM UTC

davepeck wrote: post #121

Sam Altman, in the summer update video: > "[GPT-5] can write an entire computer program from scratch, to help you with whatever you'd like. And we think this idea of software on demand is going to be one of the defining characteristics of the GPT-5 era."

"an entire computer program from scratch" != "any entire computer program from scratch"

Re: GPT-5

#275

Post by FerretFred » Thu, Aug 07, 2025, 5:49 PM UTC

Great evaluation by the (UK) BBC Evening News: basically, "it's faster, gives better answers (no detail), has a better query input (text) box, and hallucinates less". Jeez...

Re: GPT-5

#276

Post by swader999 » Thu, Aug 07, 2025, 5:49 PM UTC

quantumwoke wrote: post #236

This health segment is completely wild. Seeing Sam fully co-sign the replacement of medical advice with ChatGPT in such a direct manner would have been unheard of two years ago. Waiting for GPT-6 to include a segment on replacing management consultants.

GPT 9 still won't be able to get through the insurance dance though, maybe ten will.

Re: GPT-5

#277

Post by fidotron » Thu, Aug 07, 2025, 5:49 PM UTC

Going by the system card at: https://openai.com/index/gpt-5-system-card/

> GPT‑5 is a unified system . . .

OK

> . . . with a smart and fast model that answers most questions, a deeper reasoning model for harder problems, and a real-time router that quickly decides which model to use based on conversation type, complexity, tool needs, and explicit intent (for example, if you say “think hard about this” in the prompt).

So that's not really a unified system then, it's just supposed to appear as if it is.

This looks like they're not training the single big model but instead have gone off to develop special sub models and attempt to gloss over them with yet another model. That's what you resort to only when doing the end-to-end training has become too expensive for you.

Re: GPT-5

#278

Post by crowcroft » Thu, Aug 07, 2025, 5:50 PM UTC

I'm drowning in benchmarks and results at this point. Just show me what it can do.

Re: GPT-5

#279

Post by ComplexSystems » Thu, Aug 07, 2025, 5:50 PM UTC

koeng wrote: post #249

I hate the direction that American AI is going, and the model card of OpenAI is especially bad. I am a synthetic biologist, and I use AI a lot for my work. And it constantly denies my questions RIGHT NOW. But of course OpenAI and Anthropic have to implement more - from the GPT5 introduction: "robust safety stack with a multilayered defense system for biology" While that sounds nice and all, in practical terms, they a…

How do you suggest they solve this problem? Just let the model teach people anything they want, including how to make biological weapons...?

Re: GPT-5

#280

Post by primaprashant » Thu, Aug 07, 2025, 5:50 PM UTC

looks like 4 new features for API

- reasoning_effort parameter supports minimal value now in addition to existing low, medium, and high

- new verbosity parameter with possible values of low, medium (default), and high

- unlike hidden thinking tokens, user-visible preamble messages for tool calls are available

- tool calls possible with plaintext instead of JSON