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

141–150 of 1001 posts

Re: Gemini 3

#141

Post by iamdelirium »

HardCodedBias wrote: post #68
golfer wrote: post #29

Supposedly this is the model card. Very impressive results. https://pbs.twimg.com/media/G6CFG6jXAAA1p0I?format=jpg&name=... Also, the full document: https://archive.org/details/gemini-3-pro-model-card/page/n3/...

If you believe another thread the benchmarks are comparing Gemini-3 (probably thinking) to GPT-5.1 without thinking. The person also claims that with thinking on the gap narrows considerably. We'll probably have 3rd party benchmarks in a couple of days.

This is easily shown that the numbers are for GPT 5.1 thinking high.

Just go to the leaderboard website and see for yourself: https://arcprize.org/leaderboard

Re: Gemini 3

#142

Post by NullCascade »

I'm not a mathematician but I think we underestimate how useful pure mathematics can be to tell whether we are approaching AGI.

Can the mathematicians here try ask it to invent new novel math related to [Insert your field of specialization] and see if it comes up with something new and useful?

Try lowering the temperature, use SymPy etc.

Re: Gemini 3

#143

Post by stevesimmons »

A nice Easter egg in the Gemini 3 docs [1]:

    If you are transferring a conversation trace from another model, ... to bypass strict validation in these specific scenarios, populate the field with this specific dummy string:

    "thoughtSignature": "context_engineering_is_the_way_to_go"
[1] https://ai.google.dev/gemini-api/docs/gemini-3?thinking=high...

Re: Gemini 3

#144

Post by paganel »

63stack wrote: post #94
recitedropper wrote: post #89

[flagged]

I noticed this as well, you are already downvoted into gray

That used to be the case even before when Alphabet/Apple/Meta were negatively commented upon, I used to blame it in many of the users here (and who also happen to work for those companies) not wanting to see their total comps go down, but this right here I think that can squarely be blamed on AI-bots.

Re: Gemini 3

#145

Post by inkysigma »

rvz wrote: post #109

I expect almost no-one to read the Gemini 3 model card. But here is a damning excerpt from the early leaked model card from [0]: > The training dataset also includes: publicly available datasets that are readily downloadable; data obtained by crawlers; licensed data obtained via commercial licensing agreements; user data (i.e., data collected from users of Google products and services to train AI models, along with u…

Isn't Gmail covered under the Workspace privacy policy which forbids using that for training data. So I'm guessing that's excluded by the "in accordance" clause.

Re: Gemini 3

#146

Post by blinding-streak »

bilekas wrote: post #120

> The Gemini app surpasses 650 million users per month, more than 70% of our Cloud customers use our AI, 13 million developers have built with our generative models, and that is just a snippet of the impact we’re seeing Not to be a negative nelly, but these numbers are definitely inflated due to Google literally pushing their AI into everything they can, much like M$. Can't even search google without getting an AI re…

Gemini app != Google search.

You're implying they're lying?

Re: Gemini 3

#147

Post by ComplexSystems »

ofa0e wrote: post #126
thefourthchime wrote: post #103

Earlier quoted context omitted.

I like to ask "Make a pacman game in a single html page". No model has ever gotten a decent game in one shot. My attempt with Gemini3 was no better than 2.5.

Your benchmarks should not involve IP.

Why? This seems like a reasonable task to benchmark on.

Re: Gemini 3

#148

Post by jstummbillig »

srameshc wrote: post #112

I think I am in this AI fatigue phase. I am past all hype with models, tools and agents and back to problem and solution approach, sometimes code gen with AI , sometimes think and ask for a piece of code. But not offloading to AI and buying all the bs, waiting it to do magic with my codebase.

I think it's fun to see what is not even considered magic anymore today.

Re: Gemini 3

#149

Post by dekhn »

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.

Post reply on HN