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

931–940 of 1001 posts

Re: Gemini 3

#931

Post by rkozik1989 »

esafak wrote: post #512
ryandrake wrote: post #447

Earlier quoted context omitted.

I don't even know what the fuck "agentic" is or why the hell I would want it all over my software. So tired of everything in the computing world today.

Prompting, planning, iteration, coding, and tool use over an entire code base until a problem is solved.

Sounds like an antipattern being rebranded as a solution. I shouldn't have to precisely instruct AI on how to solve every problem. I should be able to give it requirements and with its vast knowledge it should be able to understand various design elements within a system like design patterns and make the appropriate change without me needing to tell it to look for those things.

Re: Gemini 3

#932

Post by mark_l_watson »

I had a fantastic ‘first result’ with Gemini 3 but a few people on social media I respect didn’t. Key takeaway is to do your own testing with your use cases. I feel like I am now officially biased re: LLM infrastructure: I am retired, doing personal research and writing, and I decided months ago to drop OpenAI and Anthropic infrastructure and just use Google to get stuff done - except I still budget about two hours a week to experiment with local models and Chinese models’ APIs.

Re: Gemini 3

#933

Post by eru »

elbear wrote: post #910
eru wrote: post #834

Earlier quoted context omitted.

And in general I'm all for increasing productivity, in all areas of the economy.

To what goal?

To increase livings standards for the people.

Re: Gemini 3

#934

Post by mlyle »

nomilk wrote: post #837
thomasahle wrote: post #379

Earlier quoted context omitted.

I also used Gemini 3 Pro Preview. It finished it 271s = 4m31s. Sadly, the answer was wrong. It also returned 8 "sources", like stackexchange.com, youtube.com, mpmath.org, ncert.nic.in, and kangaroo.org.pk, even though I specifically told it not to use websearch. Still a useful tool though. It definitely gets the majority of the insights. Prompt: https://aistudio.google.com/app/prompts?state=%7B%22ids%22:%...

Terrence Tao claims [0] contributions by the public are counter -productive since the energy required to check a contribution outweighs its benefit: > (for) most research projects, it would not help to have input from the general public. In fact, it would just be time-consuming, because error checking Since frontier LLMs make clumsy mistakes, they may fall into this category of 'error-prone' mathematician whose net c…

It's perhaps practical, though, to ask it to do a lot of verification and demonstration of correctness in Lean or another proof environment-- to both get its error rate down and to speed up the review of its results. After all, its time is close to "free."

Re: Gemini 3

#935

Post by Workaccount2 »

GuB-42 wrote: post #914
Workaccount2 wrote: post #100

It still failed my image identification test ([a photoshopped picture of a dog with 5 legs]...please count the legs) that so far every other model has failed agonizingly, even failing when I tell them they are failing, and they tend to fight back at me. Gemini 3 however, while still failing, at least recognized the 5th leg, but thought the dog was...well endowed. The 5th leg however is clearly a leg, despite being wh…

> Gemini 3 however, while still failing, at least recognized the 5th leg, but thought the dog was...well endowed. I see that AI is reaching the level of a middle school boy...

In reality it used the term "male anatomy" heh

Re: Gemini 3

#936

Post by taf2 »

I just wish gemini could write well formatted code. I do like the solutions it comes up to and I know I can use a linter/formatter tool - but it would just be nice if when I openned gemini (cli) up and asked it to write a feature it didn't mix up the indenting so badly... somehow codex and claude both get this without any trouble...

Re: Gemini 3

#937

Post by fluoridation »

jacquesm wrote: post #921
red75prime wrote: post #877

Earlier quoted context omitted.

Unlike general public the models can be trained. I mean if you train a member of general public, you've got a specialist, who is no longer a member of general public.

Unlike the general public though, these models have advanced dementia when it comes to learning from corrections, even within a single session. They keep regressing and I haven't found a way to stop that yet. What boggles the mind: we have gone for so long to try to strive for correctness and suddenly being right 70% of the time and wrong the remaining 30% is fine. The parallel with self driving is pretty strong here…

I think the real killer feature would be that they stop making basic mistakes, and that they gain some introspection. It's not a problem if they're wrong 30% of the time if they're able to gauge their own confidence like a human would. Then you can know to disregard the answer, or check it more thoroughly.

Re: Gemini 3

#938

Post by BoorishBears »

Dquiroga wrote: post #878
BoorishBears wrote: post #559

Earlier quoted context omitted.

> We are cheering for a product sold back to us at a 60% markup (input costs up to $2.00/M) that was built on our own private correspondence. That feels like something between a hallucination and an intentional fallacy that popped up because you specifically said "intense discussion". The increase is 60% on input tokens from the old model, but it's not a markup, and especially not "sold back to us at X markup". I've…

60% probably felt like a lot to Gemini. However, I liked the doomerism and how google was using our data to train its models. Nonetheless, Gemini 3 failed this test. It failed to start a discussion. Its points were shallow, and too aiesque.

I'm not debating 60% being a lot, it's a factually incorrect statement: markup refers to increase over cost.

Looking at it again it's actually a completely nonsensical sentence that just happens to resemble a sensible statement in a way that would fool most people.

RL is definitely showing some busting seams at this point.

Re: Gemini 3

#939

Post by mrbungie »

jmmcd wrote: post #899
nomilk wrote: post #837

Earlier quoted context omitted.

Terrence Tao claims [0] contributions by the public are counter -productive since the energy required to check a contribution outweighs its benefit: > (for) most research projects, it would not help to have input from the general public. In fact, it would just be time-consuming, because error checking Since frontier LLMs make clumsy mistakes, they may fall into this category of 'error-prone' mathematician whose net c…

But he actually uses frontier LLMs in his own work. Probably that's stronger evidence.

It is, but biased evidence, as he's both directing and checking that frontier LLM output and not everyone is Terrence Tao.

Re: Gemini 3

#940

Post by IgorPartola »

jacquesm wrote: post #921
red75prime wrote: post #877

Earlier quoted context omitted.

Unlike general public the models can be trained. I mean if you train a member of general public, you've got a specialist, who is no longer a member of general public.

Unlike the general public though, these models have advanced dementia when it comes to learning from corrections, even within a single session. They keep regressing and I haven't found a way to stop that yet. What boggles the mind: we have gone for so long to try to strive for correctness and suddenly being right 70% of the time and wrong the remaining 30% is fine. The parallel with self driving is pretty strong here…

A common resistor has a +/- 10% tolerance. A milspec one is 1%. Yet we have ways of building robust systems using such “subpar” components. The trick is to structure the system in a way that builds the error rate into the process and corrects for it. Easier said than done of course for a lot of problems but we do have techniques for doing this and we are learning more.
Post reply on HN