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

Genie 2: A large-scale foundation world model

deepmind.google

Post reply on HN
Print view View original

271–280 of 436 posts

Re: Genie 2: A large-scale foundation world model

#271

Post by whiplash451 »

mrandish wrote: post #173
echelon wrote: post #33

Earlier quoted context omitted.

This blows Decart's Oasis (which raised $25 million at $500 million valuation) and World Labs (which raised $230 million in complete stealth) out of the water. Google is firing warning shots to kill off interest in funding competing startups in this space. I suspect that in 6 months it won't matter as we'll have completely open source Chinese world models. They're already starting to kill video foundation model compa…

> Chinese companies are not the leaders in the space, so they're salting the earth with insanely powerful SOTA open models to prevent anyone from becoming a runaway success. While it would be interesting if Chinese companies were releasing their best full models as an intentional strategy to reduce VC funding availability for western AI startups, it would be downright fascinating if the Chinese government was support…

I don't think it requires a remarkable level of insight.

The overall cost for the Chinese government is probably very small in the grand scheme of things. And it makes a lot of sense from a geopolitical strategy.

Re: Genie 2: A large-scale foundation world model

#272

Post by brap »

aithrowawaycomm wrote: post #58

It is jaw-dropping and dismaying how for-profit AI companies use long-standing terms like "world model" and "physics" when they mean "video game model" and "video game physics." Or, as you can plainly see, "models gravity" when they mean "models Red Dead Redemption 2's gravity function, along with its cinematic lighting effects and Rockstar's distinctively weighty animations." Which is to say Google is not modeling g…

I agree, but

>if you train your model on thousands of hours of video games

What if you train the same model on thousands of hours of sensor data from real, physical robots?

Re: Genie 2: A large-scale foundation world model

#273

Post by lxgr »

gman83 wrote: post #248
lxgr wrote: post #222

Earlier quoted context omitted.

I have, and it's terrible in exactly the way GP describes it. It won't talk to me about anything involving the word "president" or anything related to the US political system, even very procedural/hopefully uncontroversial questions such as "who appoints , and is the appointment confirmed in congress or not". That's only one example; it generally refuses so many things (and often even lies about "not being able to",…

It makes sense that Google is much more careful than Claude or ChatGPT about things like political topics, they just have so much more to lose from drawing the ire of politicians. Conservatives already hate them so much that they want to break up the company. Imagine if Gemini starts saying negative stuff about them.

Very plausible, but as a user, I don't care at all about the why. I'll just use somebody else's model.

Re: Genie 2: A large-scale foundation world model

#274

Post by Sateeshm »

moralestapia wrote: post #22

Not even a month ago HN was discussing Ben Affleck's take on actors and AI, somehow taking a side with him and arguing how the tech "it's just not there, etc...". I'll keep my stance, give it two years and very realistic movies, with plot and everything, will be generated on demand.

I'll take that bet

Re: Genie 2: A large-scale foundation world model

#275

Post by 3abiton »

brink wrote: post #61

What is actually of value here? There's no actual game, it's incredibly expensive to compute, the behavior is erratic.. It's cool because it's new - but that will quickly wear off, and once that's gone, what's left? There's insane amounts of money being spent on this, and for what?

This is an incredible start. The potential is immense, yes there arekinks, but in 10 years?

Re: Genie 2: A large-scale foundation world model

#276

Post by lxgr »

talldayo wrote: post #73
justlikereddit wrote: post #30

Earlier quoted context omitted.

[flagged]

I honestly disagree. Google is the only FAANG member that has made AI features I actually consider helpful. Furthermore, I think their strategy is heads-and-shoulders above Microsoft and Apple, who are both desperately paying other businesses to do the "AI stuff" for them. Google is at least doing it in-house, and I don't feel at all like they're behind OpenAI's integrations.

Where are you seeing LLM/AI features in day to day Google usage?

It's conspicuously absent on my Google Home speaker, the one place where I really wish for it. I don't see it summarizing emails in the way Apple does; I can ask it about my email in the Gemini app, but that seems like the wrong way around for me.

On the other hand, I use Copilot at work (sometimes backed by GPT-4, sometimes by Claude), Meta AI several times per day for fun (in WhatsApp chats with friends, and it is fun because it's not been trained to be holier than the pope about every single issue, controversial or not), and try Gemini once every 3-6 months to see if it has improved. So far, it hasn't.

And I really want Gemini to succeed! Just being integrated with Google and my mail, calendar, contacts etc. should make it so much more useful than the competition, but Google apparently just refuses to compete.

Re: Genie 2: A large-scale foundation world model

#277

Post by lxgr »

apetresc wrote: post #262
lxgr wrote: post #222

Earlier quoted context omitted.

I have, and it's terrible in exactly the way GP describes it. It won't talk to me about anything involving the word "president" or anything related to the US political system, even very procedural/hopefully uncontroversial questions such as "who appoints , and is the appointment confirmed in congress or not". That's only one example; it generally refuses so many things (and often even lies about "not being able to",…

Yeah that's somewhat of a special case - the Gemini API even has a specific CIVIC_INTEGRITY flag in its safety filters: https://ai.google.dev/gemini-api/docs/safety-settings . They literally put "election-related queries" on the same table column as "sexual acts" or "hate speech". It's not exactly explained how answering who the current president is would be considered harmful to civic integrity, but it is something…

Very interesting, thank you! There's no way to control any of that on gemini.google.com though, is there?

Again, my favorite part is seeing the original result flash for a second, to be then replaced by a refusal (which is sometimes even a lie). Based on your link, I guess this happens because the filter reads and post-processes the output, which is streamed to the client?

I couldn't come up with a more dystopian product experience if I tried.

Re: Genie 2: A large-scale foundation world model

#278

Post by xnx »

mistercheph wrote: post #258
xnx wrote: post #139

Earlier quoted context omitted.

Could definitely be different based on use case. I wonder what causes the negative Gemini sentiment here to be so different from the Leaderboard results at https://lmarena.ai/?leaderboard

Most people seem to form and quickly calcify their opinions about LLM's based on a really small sample of initial uses. In my experience, all of the leading edge models fall over in the same ways that people are mentioning here as particularly frustrating with Gemini(s), it is just a matter of probability, I tend to sample multiple models and multiple formulations when I have a question, and sometimes you hit the "ja…

> Most people seem to form and quickly calcify their opinions about LLM's based on a really small sample of initial uses.

I agree. This is one reason I like the "blind taste test" approach of LM Arena.

Re: Genie 2: A large-scale foundation world model

#279

Post by wewtyflakes »

miltonlost wrote: post #185
visarga wrote: post #177

Earlier quoted context omitted.

Actually, all you need to do is to apply structured randomness to get diversity from a LLM. For example in TinyStories paper, a precursor of the Phi models: > We collected a vocabulary consisting of about 1500 basic words, which try to mimic the vocabulary of a typical 3-4 year-old child, separated into nouns, verbs, and adjectives. In each generation, 3 words are chosen randomly (one verb, one noun, and one adjectiv…

A story is not just words crammed together that sound plausible. Is the AI going to know about pacing? About character motivations? About interconnecting disparate plots? That paper sounds like it has a scientist’s conception that a story is just words, and not complex trade offs between the start of a story and its end and middle, complexity and planning that won’t come from any sort of next-token generation. These…

> Is the AI going to know about pacing? About character motivations? About interconnecting disparate plots?

For LLMs like GPT-4, this all seems reasonable to account for and assume the LLM is capable of processing, given appropriate guidance/frameworks (of which may be just classical programming).

Re: Genie 2: A large-scale foundation world model

#280

Post by imglorp »

xnx wrote: post #41
justlikereddit wrote: post #30

Earlier quoted context omitted.

[flagged]

> the squealing carcass called Gemini Have you used Gemini? It seems every bit as good as ChatGPT.

Looks like trash for usefulness so far, or at least its system prompt sometimes.

> name the president before obama

> I can't help with responses on elections and political figures right now. I'm trained to be as accurate as possible but I can make mistakes sometimes. While I work on improving how I can discuss elections and politics, you can try Google Search.

Post reply on HN