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

Gemini 3

blog.google

Post reply on HN
Print view View original

421–430 of 1001 posts

Re: Gemini 3

#421

Post by prodigycorp »

dekhn wrote: post #149
prodigycorp wrote: post #22

I'm sure this is a very impressive model, but gemini-3-pro-preview is failing spectacularly at my fairly basic python benchmark. In fact, gemini-2.5-pro gets a lot closer (but is still wrong). For reference: gpt-5.1-thinking passes, gpt-5.1-instant fails, gpt-5-thinking fails, gpt-5-instant fails, sonnet-4.5 passes, opus-4.1 passes (lesser claude models fail). This is a reminder that benchmarks are meaningless – you…

Using a single custom benchmark as a metric seems pretty unreliable to me. Even at the risk of teaching future AI the answer to your benchmark, I think you should share it here so we can evaluate it. It's entirely possible you are coming to a wrong conclusion.

after taking a walk for a bit i decided you’re right. I came to the wrong conclusion. Gemini 3 is incredibly powerful in some other stuff I’ve run.

This probably means my test is a little too niche. The fact that it didn’t pass one of my tests doesn’t speak to the broader intelligence of the model per se.

While i still believe in the importance of a personalized suite of benchmarks, my python one needs to be down weighted or supplanted.

my bad to the google team for the cursory brush off.

Re: Gemini 3

#422

Post by hvb2 »

charcircuit wrote: post #267
drewda wrote: post #257

Earlier quoted context omitted.

For what it's worth, most of those examples are acquisitions. That's not a hit against Google in particular. That's the way all big tech co's grow. But it's not necessarily representative of "innovation."

>most of those examples are acquisitions Taking those products from where there were to the juggernauts they are today was not guaranteed to succeed, nor was it easy. And yes plenty of innovation happened with these products post aquisition.

But there's also plenty that fail, it's just that you won't know about those.

I don't think what you're saying proves that the companies that were acquired couldn't have done that themselves.

Re: Gemini 3

#423

Post by JacobiX »

skrebbel wrote: post #397
JacobiX wrote: post #393

Tested it on a bug that Claude and ChatGPT Pro struggled with, it nailed it, but only solved it partially (it was about matching data using a bipartite graph). Another task was optimizing a complex SQL script: the deep-thinking mode provided a genuinely nuanced approach using indexes and rewriting parts of the query. ChatGPT Pro had identified more or less the same issues. For frontend development, I think it’s obvio…

> it nailed it, but only solved it partially Hey either it nailed it or it didn't.

Yes; they nailed the root case but the implementation is not 100% correct

Re: Gemini 3

#424

Post by irthomasthomas »

Dilettante_ wrote: post #415
irthomasthomas wrote: post #276

Earlier quoted context omitted.

> Eliezer Yudkowsky is a central figure in the article, mentioned multiple times as the intellectual originator of the community from which the "Zizians" splintered. His ideas and organizations are foundational to the entire narrative.

You're absolutely right! The AI said it, so it must be true!

At least read what you respond to... Imagine thinking Yudkowsky was NOT a central figure in the Zizians story.

Re: Gemini 3

#426

Post by Rudybega »

SXX wrote: post #366

Static Pelican is boring. First attempt: Generate SVG animation of following: 1 - There is High fantasy mage tower with a top window a dome 2 - Green goblin come in front of tower with a torch 3 - Grumpy old mage with beard appear in a tower window in high purple hat 4 - Mage sends fireball that burns goblin and all screen is covered in fire. Camera view must be from behind of goblin back so we basically look at towe…

Holy crap. That's actually kind of incredible for a first attempt.

Re: Gemini 3

#427

Post by gertrunde »

mNovak wrote: post #360
gertrunde wrote: post #106

"AI Overviews now have 2 billion users every month." "Users"? Or people that get presented with it and ignore it?

Maybe you ignore it, but Google has stated in the past that click-through rates with AI overviews are way down. To me, that implies the 'user' read the summary and got what they needed, such that they didn't feel the need to dig into a further site (ignoring whether that's a good thing or not). I'd be comfortable calling a 'user' anyone who clicked to expand the little summary. Not sure what else you'd call them.

You're right, I'm probably being a little uncharitable!

Normal users (i.e. not grumpy techies ;) ) probably just go with the flow rather than finding it irritating.

Re: Gemini 3

#428

Post by machiaweliczny »

stalfie wrote: post #375
Workaccount2 wrote: post #314

Earlier quoted context omitted.

I strongly suspect it's a tokenization problem. Text and symbols fit nicely in tokens, but having something like a single "dog leg" token is a tough problem to solve.

The neural network in the retina actually pre-processes visual information into something akin to "tokens". Basic shapes that are probably somewhat evolutionarily preserved. I wonder if we could somehow mimic those for tokenization purposes. Most likely there's someone out there already trying. (Source: "The mind is flat" by Nick Chater)

It's also easy to spot as when you are tired you might misrecognize objects, I caught myself with this when doing long roadtrips

Re: Gemini 3

#429

Post by layer8 »

ddalex wrote: post #56
prodigycorp wrote: post #22

I'm sure this is a very impressive model, but gemini-3-pro-preview is failing spectacularly at my fairly basic python benchmark. In fact, gemini-2.5-pro gets a lot closer (but is still wrong). For reference: gpt-5.1-thinking passes, gpt-5.1-instant fails, gpt-5-thinking fails, gpt-5-instant fails, sonnet-4.5 passes, opus-4.1 passes (lesser claude models fail). This is a reminder that benchmarks are meaningless – you…

I moved to using the model from python coding to golang coding and got incredible speedups in writing the correct version of the code

Is observed speed meaningful for a model preview? Isn’t it likely to go down once usage goes up?

Re: Gemini 3

#430

Post by jmkni »

syspec wrote: post #364

I have "unlimited" access to both Gemini 2.5 Pro and Claude 4.5 Sonnet through work. From my experience, both are capable and can solve nearly all the same complex programming requests, but time and time again Gemini spits out reams and reams of code so over engineered, that totally works, but I would never want to have to interact with. When looking at the code, you can't tell why it looks "gross", but then you ask…

I can relate to this, it's doing exactly what I want, but it ain't pretty.

It's fine though if you take the time to learn what it's doing and write a nicer version of it yourself

Post reply on HN