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

Gemini AI

deepmind.google

Post reply on HN
Print view View original

281–290 of 1001 posts

Re: Gemini AI

#281

Post by Racing0461 »

scarmig wrote: post #275
Racing0461 wrote: post #243

How do we know the model wans't pretrained on the evaluations to get higher scores? In general but especially for profit seeking corporations, this measure might become a target and become artificial.

Most engineers and researchers at big tech companies wouldn't intentionally do that. The bigger problem is that public evals leak into the training data. You can try to cleanse your training data, but at some point it's inevitable.

Yeah, i not saying it was intentional (misleading shareholders would be the worse crime here). Having these things in the training data without knowing due to how vast the dataset is is the issue.

Re: Gemini AI

#282

Post by verdverm »

ilaksh wrote: post #58
albertzeyer wrote: post #42

So, better than GPT4 according to the benchmarks? Looks very interesting. Technical paper: https://goo.gle/GeminiPaper Some details: - 32k context length - efficient attention mechanisms (for e.g. multi-query attention (Shazeer, 2019)) - audio input via Universal Speech Model (USM) (Zhang et al., 2023) features - no audio output? (Figure 2) - visual encoding of Gemini models is inspired by our own foundational work o…

That's for Ultra right? Which is an amazing accomplishment, but it sounds like I won't be able to access it for months. If I'm lucky.

There was a waiting period for ChatGPT4 as well, particularly direct API access, and the WebUI had (has?) a paywall

Re: Gemini AI

#283

Post by infoseek12 »

tikkun wrote: post #60

One observation: Sundar's comments in the main video seem like he's trying to communicate "we've been doing this ai stuff since you (other AI companies) were little babies" - to me this comes off kind of badly, like it's trying too hard to emphasize how long they've been doing AI (which is a weird look when the currently publicly available SOTA model is made by OpenAI, not Google). A better look would simply be to sh…

> "we've been doing this ai stuff since you (other AI companies) were little babies"

Well in fairness he has a point, they are starting to look like a legacy tech company.

Re: Gemini AI

#284

Post by xnx »

It doesn't feel like a coincidence that this announcement is almost exactly one year after the release of ChatGPT.

Re: Gemini AI

#285

Post by kernal »

MadSudaca wrote: post #28

It's truly astounding to me that Google, a juggernaut with decades under its belt on all things AI, is only now catching up to OpenAI which is on all camps a fraction of its size.

How many other companies can you say that have possibly passed GPT-4?

Re: Gemini AI

#286

Post by ghaff »

This is hilarious for anyone who knows the area:

"The best way to get from Lake of the Clouds Hut to Madison Springs Hut in the White Mountains is to hike along the Mt. Washington Auto Road. The distance is 3.7 miles and it should take about 16 minutes."

What it looks like it's doing is actually giving you the driving directions from the nearest road point to one hut to the nearest road point to the other hut.

An earlier version actually did give hiking directions but they were hilariously wrong even when you tried to correct it.

That said, I did ask a couple historical tech questions and they seemed better than previously--and it even pushed back on the first one I asked because it wanted me to be more specific. Which was very reasonable; it wasn't really a trick question but it's one you could take in multiple directions.

Re: Gemini AI

#287

Post by xen2xen1 »

vilunov wrote: post #75
jodrellblank wrote: post #51

There's some dissonance in the the way this will swamp out searches for the web-alternative Gemini protocol by the biggest tech company in the world proudly boasting how responsible and careful they are being to improving things "for everyone, everywhere in the world".

It's probably just an unfortunate coincidence. After all, Gemini is a zodiac sign first and foremost, you'd have to specify what exactly you want anyway.

Wasn't Gemini part of Greek Mythology way, way before? Aren't you losing maybe thousands of years here?

Re: Gemini AI

#288

Post by summerlight »

It looks like they tried to push it out ASAP? Gemini Ultra is the largest model and it usually takes several months to train such, especially if you want to enable more efficient inference which seems to be one of its goals. My guess is that the Ultra model very likely finished its training pretty recently so it didn't have a much time to validate or further fine-tune. Don't know the contexts though...

Re: Gemini AI

#289

Post by xfalcox »

aaronharnly wrote: post #183

I did some side-by-side comparisons of simple tasks (e.g. "Write a WCAG-compliant alternative text describing this image") with Bard vs GPT-4V. Bard's output was significantly worse. I did my testing with some internal images so I can't share, but will try to compile some side-by-side from public images.

I'm researching using LLMs for alt-text suggestion for forum users, can you share your finding so far?

Outside of GPT-4V I had good first results with https://github.com/THUDM/CogVLM

Re: Gemini AI

#290

Post by deeviant »

jillesvangurp wrote: post #168
logicchains wrote: post #94

Interesting that they're announcing Ultra many months in advance of the actual public release. Isn't that just giving OpenAI a timeline for when they need to release GPT5? Google aren't going to gain much market share from a model competitive with GPT4 if GPT5 is already available.

I don't think there are a lot of surprises on either side about what's coming next. Most of this is really about pacifying shareholders (on Google's side) who are no doubt starting to wonder if they are going to fight back at all. With either OpenAI and Google, or even Microsoft, the mid term issue is as much going to be about usability and deeper integration than it is about model fidelity. Chat gpt 4 turbo is prett…

> Chat gpt 4 turbo is pretty nice but the UI/UX is clumsy.

Strong disagree. ChatGPT is the only UI/UX that I find usable (great, no. Better than the rest, by miles). Bing and Google are damned clunky.

It's also easy to make you own interface with via the API.

Post reply on HN