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

Gemini 3 Deep Think

blog.google

Post reply on HN
Print view View original

601–610 of 722 posts

Re: Gemini 3 Deep Think

#601

Post by te7447 »

pixl97 wrote: post #346
verdverm wrote: post #87

Earlier quoted context omitted.

I suspect the non-spikey part is the more interesting comparison Why is it so easy for me to open the car door, get in, close the door, buckle up. You can do this in the dark and without looking. There are an infinite number of little things like this you think zero about, take near zero energy, yet which are extremely hard for Ai

>Why is it so easy for me to open the car door Because this part of your brain has been optimized for hundreds of millions of years. It's been around a long ass time and takes an amazingly low amount of energy to do these things. On the other hand the 'thinking' part of your brain, that is your higher intelligence is very new to evolution. It's expensive to run. It's problematic when giving birth. It's really slow wi…

> There's a term for this, but I can't think of it at the moment.

Moravec's paradox: https://epoch.ai/gradient-updates/moravec-s-paradox

Re: Gemini 3 Deep Think

#602

Post by usefulposter »

red75prime wrote: post #524
toraway wrote: post #515

Earlier quoted context omitted.

Huh? AI labs are routinely spending millions to billions to various 3rd party contractors specializing in creating/labeling/verifying specialized content for pre/post-training. This would just be one more checkbox buried in hundreds of pages of requests, and compared to plenty of other ethical grey areas like copyright laundering with actual legal implications, leaking that someone was asked to create a few dozen pel…

How do you think who's in on that? Not only pelicans, I mean, the whole thing. CEOs, top researchers, select mathematicians, congressmen? Does China participate in maintaining the bubble? I, myself, prefer the universal approximation theorem and empirical finding that stochastic gradient descent is good enough (and "no 'magic' in the brain", of course).

Well, since we're all talking about sourcing training material to "benchmaxx" for social proof, and not litigating the whole "AI bubble" debate, just the entire cottage industry of data curation firms:

https://scale.com/data-engine

https://www.appen.com/llm-training-data

https://www.cogitotech.com/generative-ai/

https://www.telusdigital.com/solutions/data-for-ai-training/...

https://www.nexdata.ai/industries/generative-ai

---

P.S. Google Comms would have been consulted re putting a pelican in the I/O keynote :-)

https://x.com/simonw/status/1924909405906338033

Re: Gemini 3 Deep Think

#603

Post by ruszki »

cubefox wrote: post #542
janalsncm wrote: post #363

Earlier quoted context omitted.

The term “model” is one of those super overloaded terms. Depending on the conversation it can mean: - a product (most accurate here imo) - a specific set of weights in a neural net - a general architecture or family of architectures (BERT models) So while you could argue this is a “model” in the broadest sense of the term, it’s probably more descriptive to call it a product. Similarly we call LLMs “language” models e…

I'm pretty sure only the second is properly called a model, and "BERT models" are simply models with the BERT architecture.

It depends on time. 5 years ago it was quite well defined that it’s the last one, maybe the second one in some context. Especially when distinction was important, it was always the last one. In our case it was. We trained models to have weights. We even stored models and weights separately, because models change slower than weights. You could choose a model and a set of weights, and run them. You could change weights any time.

Then marketing, and huge amount of capital came.

Re: Gemini 3 Deep Think

#604

Post by fatherwavelet »

sdeiley wrote: post #587
HardCodedBias wrote: post #152

Always the same with Google. Gemini has been way behind from the start. They use the firehose of money from search to make it as close to free as possible so that they have some adoption numbers. They use the firehose from search to pay for tons of researchers to hand hold academics so that their non-economic models and non-economic test-time-compute can solve isolated problems. It's all so tiresome. Try making model…

I'm sorry but this is an insane take. Flash is leading its category by far. Absolutely destroys sonnet, 5.2 etc in both perf and cost. Pro still leads in visual intelligence. The company that most locks away their gold is Anthropic IMO and for good reason, as Opus 4.6 is expensive AF

I think we highly underestimate the amount of "human bots" basically.

Unthinking people programmed by their social media feed who don't notice the OpenAI influence campaign.

With no social media, it seems obvious to me there was a massive PR campaign by OpenAI after their "code red" to try to convince people Gemini is not all that great.

Yea, Gemini sucks, don't use it lol. Leave those resources to fools like myself.

Re: Gemini 3 Deep Think

#605

Post by simonw »

enraged_camel wrote: post #168
simonw wrote: post #39

The pelican riding a bicycle is excellent . I think it's the best I've seen. https://simonwillison.net/2026/Feb/12/gemini-3-deep-think/

Is there a list of these for each model, that you've catalogued somewhere?

At the moment that's mostly my tag page here but I really need to formalize it: https://simonwillison.net/tags/pelican-riding-a-bicycle/

Re: Gemini 3 Deep Think

#606

Post by cubefox »

ruszki wrote: post #603
cubefox wrote: post #542

Earlier quoted context omitted.

I'm pretty sure only the second is properly called a model, and "BERT models" are simply models with the BERT architecture.

It depends on time. 5 years ago it was quite well defined that it’s the last one, maybe the second one in some context. Especially when distinction was important, it was always the last one. In our case it was. We trained models to have weights. We even stored models and weights separately, because models change slower than weights. You could choose a model and a set of weights, and run them. You could change weights…

It seems unlikely "model" was ever equivalent in meaning to "architecture". Otherwise there would be just one "CNN model" or just one "transformer model" insofar there is a single architecture involved.

Re: Gemini 3 Deep Think

#607

Post by mohsen1 »

aliljet wrote: post #186

The problem here is that it looks like this is released with almost no real access. How are people using this without submitting to a $250/mo subscription?

I have some very difficult to debug bugs that Opus 4.6 is failing at. Planning to pay $250 to see if it can solve those.

Re: Gemini 3 Deep Think

#608

Post by Foobar8568 »

pell wrote: post #497
thefounder wrote: post #495

Earlier quoted context omitted.

Am I the only one that can’t find Gemini useful except if you want something cheap? I don’t get what was the whole code red about or all that PR. To me I see no reason to use Gemini instead of of GPT and Anthropic combo. I should add that I’ve tried it as chat bot, coding through copilot and also as part of a multi model prompt generation. Gemini was always the worst by a big margin. I see some people saying it is sm…

I find the quality is not consistent at all and of all the LLMs I use Gemini is the one most likely to just verge off and ignore my instructions.

Same, as far as I am concerned, Gemini is optimized for benchmarks.

I mean last week it insisted suddenly on two consecutive prompts that my code was in python. It was in rust.

Re: Gemini 3 Deep Think

#609

Post by ebonnafoux »

recursive wrote: post #50
throwup238 wrote: post #47

Earlier quoted context omitted.

For every combination of animal and vehicle? Very unlikely. The beauty of this benchmark is that it takes all of two seconds to come up with your own unique one. A seahorse on a unicycle. A platypus flying a glider. A man’o’war piloting a Portuguese man of war. Whatever you want.

No, not every combination. The question is about the specific combination of a pelican on a bicycle. It might be easy to come up with another test, but we're looking at the results from a particular one here.

You can easily make a RLAIF loop.

- Take a list of n animals * m vehicule

- Ask a LLM to generate SVG for this n*m options

- Generate png from the svg

- Ask a Model with vision to grade the result

- Change your weight accordingly

No need to human to draw the dataset, no need of human to evaluate.

Re: Gemini 3 Deep Think

#610

Post by virgildotcodes »

xnx wrote: post #15

Google is absolutely running away with it. The greatest trick they ever pulled was letting people think they were behind.

These benchmarks are super impressive. That said, Gemini 3 Pro benchmarked well on coding tasks, and yet I found it abysmal. A distant third behind Codex and Claude.

Tool calling failures, hallucinations, bad code output. It felt like using a coding model from a year ago.

Even just as a general use model, somehow ChatGPT has a smoother integration with web search (than google!!), knowing when to use it, and not needing me to prompt it directly multiple times to search.

Not sure what happened there. They have all the ingredients in theory but they've really fallen behind on actual usability.

Their image models are kicking ass though.

Post reply on HN