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

GPT-4o

openai.com

Post reply on HN
Print view View original

281–290 of 1001 posts

Re: GPT-4o

#281

Post by freedomben »

onemiketwelve wrote: post #98
Jimmc414 wrote: post #59

Big questions are (1) when is this going to be rolled out to paid users? (2) what is the remaining benefit of being a paid user if this is rolled out to free users? (3) Biggest concern is will this degrade the paid experience since GPT-4 interactions are already rate limited. Does OpenAI have the hardware to handle this? Edit: according to @gdb this is coming in "weeks" https://twitter.com/gdb/status/1790074041614717…

thanks, I was confused because the top of the page says to try now when you cannot in fact try it at all

I'm a ChatGPT Plus Subscriber and I just refreshed the page and it offered me the new model. I'm guessing they're rolling it out gradually but hopefull it won't take too long.

Edit: It's also now available to me in the Android App

Re: GPT-4o

#282

Post by quenix »

dom96 wrote: post #163

I can't help but feel a bit let down. The demos felt pretty cherry picked and still had issues with the voice getting cut off frequently (especially in the first demo). I've already played with the vision API, so that doesn't seem all that new. But I agree it is impressive. That said, watching back a Windows Vista speech recognition demo[1] I'm starting to wonder if this stuff won't have the same fate in a few years.…

I think the voice was getting cut off because it heard the crowd reaction and paused (basically it's a feature, not a bug).

Re: GPT-4o

#283

Post by mtam »

GPT-4o is very fast but seems to generate some very random ASCII Art compared to GPT-4 when text in the art is involved.

Re: GPT-4o

#284

Post by benlivengood »

seydor wrote: post #233

I would still prefer the features in text form, in the chat GUI. Right now chatGPT doesnt seem to have options to lengthen parts of the text response, to change it etc. Perplexity and gemini do seem to get the gui right. Voice chat is fun for demos but won't catch much, just like all the predecessors. Perhaps an advanced version of this could be used as a student tutor however

I am guessing text chat will be improved in all multimodal models because they have a broader base of data for pre-training. Benchmarks seem to show 4o slightly exceeding 4 (despite being a smaller model, or at least more parallelizable)

Re: GPT-4o

#285

Post by ta-run »

This looks too good to be true? What's the catch?

Also, wasn't expecting the perf to improve by 2x

Re: GPT-4o

#286

Post by 0xbadc0de5 »

As a paid user, it would have been nice to see something that differentiates that investment from the free tier.

The tech demos are cool and all - but I'm primarily interested in the correctness and speed of ChatGPT and how well it aligns with my intentions.

Re: GPT-4o

#287

Post by ta-run »

Jensson wrote: post #24

The most impressive part is that the voice uses the right feelings and tonal language during the presentation. I'm not sure how much of that was that they had tested this over and over, but it is really hard to get that right so if they didn't fake it in some way I'd say that is revolutionary.

Crazy that interruption also seems to work pretty smoothly

Re: GPT-4o

#290

Post by razodactyl »

I think this is a great example of the bootstrapping that was enabled when they pipelined the previous models together.

We do this all the time in ML. You can generate a very powerful dataset using these means and further iterate with the end model.

What this tells me now is that the runway to GPT5 will be laid out with this new architecture.

It was a bit cold in Australia today. Did you Americans stop pumping out GPU heat temporarily with the new model release? Heh

Post reply on HN