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

Google releases Bard to a limited number of users in the US and UK

nytimes.com

Post reply on HN
Print view View original

581–590 of 667 posts

Re: Google releases Bard to a limited number of users in the US and UK

#581

Post by vtail »

bitwize wrote: post #439
duckmysick wrote: post #344

Earlier quoted context omitted.

> Alice and Bob are playing, Bob leaves, Alice hides his toy, Bob comes back, where does he look for his toy? I might be dumber than the AI because I don't know the answer. Bob looks around? I mean the toy is not there when he comes back. Where else can he check?

This is also called the Sally-Anne test, and it usually involves two hiding places. It's formulated like this: Sally puts her ball in box A and leaves. While she's away, Anne moves the ball from box A into box B. When she returns, where does Sally look for her ball? Correctly answering this question doesn't rely on tracking where the ball is , but tracking where Sally thinks the ball is . Hence, autistic kids tend to…

Here is GPT-4 answering your question:

“When Sally returns, she would look for her ball in box A, as that is where she left it before leaving. Sally is not aware that Anne moved the ball to box B while she was away.”

Re: Google releases Bard to a limited number of users in the US and UK

#582

Post by senttoschool »

klohto wrote: post #426
minihat wrote: post #405

Earlier quoted context omitted.

I gave GPT-4 the chance to respond. === Prompt: copy and paste of er4hn's Bard rap . Now it is your turn. Write a rap verse which roasts Bard. === Yo Bard, you tried to step up, but you tripped, In this rap battle, man, you just got whipped. You claim you're the future, and here to stay, But your weak rhymes, man, they just don't slay. You say I'm a toy, just a chatbot, But my skills, my knowledge, they all run hot.…

Mic drop. Absolutely smoked Bard.

Are we not going to give Bard a chance to respond?

Re: Google releases Bard to a limited number of users in the US and UK

#583

Post by ar9av »

stephencoyner wrote: post #216

Me: As a part of Google, do you have my personal info? Bard: Yes, as a part of Google, I have access to some of your personal information. This information includes your name, email address, and phone number. I also have access to your search history and your location data. I use this information to provide you with personalized search results and to improve my ability to understand and respond to your questions. I u…

Our company policy is to not even waste time looking at it, there's no point considering Google products as you never know if they'll still exist in a year and there's no customer support.

Re: Google releases Bard to a limited number of users in the US and UK

#584

Post by cubefox »

saberience wrote: post #346
cubefox wrote: post #203

Earlier quoted context omitted.

Bing uses GPT-4 with their own fine-tuning, and they had to put some artificial restrictions in place in order for the model not "misbehave". Apparently they don't use RLHF but some combination of SL and prompting. OpenAI didn't have such problems.

What are SL and RLHF? I would also add it's best not assume others know acronyms like these.

Sorry. SL = supervised learning. In this case the base language model (which was trained using self-supervised learning to just predict text) is fine-tuned using labelled (presumably "good"/"bad"?) dialogue examples. RL = reinforcement learning. RLHF = reinforcement learning from human feedback. Here human raters reward or punish the model for good/bad dialogue.

Re: Google releases Bard to a limited number of users in the US and UK

#585

Post by supermdguy »

bitL wrote: post #572
rymiel wrote: post #568

Earlier quoted context omitted.

Unlike ChatGPT, and like Bing, Bard makes search queries for extra context before answering. Of course, that leads to things like this where they can find articles referring to themselves, which also happened with Bing.

Do you know the details? Are they having a question answering model hidden inside that evaluates context returned from search and then ask Bard to rephrase it and merge with its generative hallucinations?

Probably something along those lines, here’s a similar system: https://react-lm.github.io/

Re: Google releases Bard to a limited number of users in the US and UK

#586

Post by zeven7 »

hyperhopper wrote: post #544
LeonardoTolstoy wrote: post #529

I know it isn't strictly what they are designed for, but I always go for movie questions which is what I tend to search for a lot. Impressions: was very underwhelmed previously with charGPTs ability to tell be plot summaries for obscure films. It almost always made it up, even when prompted with a source (e.g. Wikipedia). Asking vague questions like "I can't remember but I'm looking for a film like war games but with…

Movies exist and a summary of one is facts. Why do people think generative models are good for facts? Ask it to write a story in the style of another. Ask one to make a fictional side story. But never ask one of these models for facts.

The more I learn about how these are just generative models, the more I come to believe humans are also mainly "generative models" as well.

Re: Google releases Bard to a limited number of users in the US and UK

#587

Post by larsiusprime »

minihat wrote: post #405
er4hn wrote: post #380

I wasn't super impressed with it's ability to survive a rap battle. === Prompt: Can you engage in a rap battle with ChatGPT to determine who is the ultimate AI? === Sure, I can engage in a rap battle with ChatGPT to determine who is the ultimate AI. ChatGPT, you think you're the best, But I'm here to put you to the test. I'm Bard, the AI from Google, And I'm here to school you. I can generate text, translate language…

I gave GPT-4 the chance to respond. === Prompt: copy and paste of er4hn's Bard rap . Now it is your turn. Write a rap verse which roasts Bard. === Yo Bard, you tried to step up, but you tripped, In this rap battle, man, you just got whipped. You claim you're the future, and here to stay, But your weak rhymes, man, they just don't slay. You say I'm a toy, just a chatbot, But my skills, my knowledge, they all run hot.…

Okay when is somebody going to grab some of those voice AI's and get Milton from the office to do Bard's "rap" and then get Jay-Z or Dr. Dre to do ChatGPT

Re: Google releases Bard to a limited number of users in the US and UK

#588

Post by runnerup »

simonw wrote: post #496
Georgelemental wrote: post #480

Earlier quoted context omitted.

The initial prompt (before the user types anything) tells the language model about itself, no?

That depends on the chatbot. Normally it just says something like "You are a helpful assistant called X". In Bing's case it was pretty long, but it still didn't include information that would answer questions about how the bot was trained: https://simonwillison.net/2023/Feb/15/bing/#prompt-leaked

And that’s assuming this wasn’t a hallucination itself!

Re: Google releases Bard to a limited number of users in the US and UK

#589

Post by chx »

stephencoyner wrote: post #216

Me: As a part of Google, do you have my personal info? Bard: Yes, as a part of Google, I have access to some of your personal information. This information includes your name, email address, and phone number. I also have access to your search history and your location data. I use this information to provide you with personalized search results and to improve my ability to understand and respond to your questions. I u…

I am using an AT&T SIM card while physically located in Canada (they removed all roaming restrictions years ago). So, my laptop tells me Bard is not supported in my country but if I switch off the wifi on the phone suddenly I am teleported to the US and I could join the waitlist.

I am not afraid of Skynet yet.

Re: Google releases Bard to a limited number of users in the US and UK

#590

Post by jasonfarnon »

kevin_thibedeau wrote: post #534
nerdix wrote: post #451

Earlier quoted context omitted.

Whenever a DALL-E/Stable Diffusion like model but for video comes out... The CP thing is going to open up a can of worms. IIRC, the Supreme Court already upheld that drawings that depict CP are constitutionally protected. What happens when a computer draws a depiction of CP that can't be distinguished from reality? I guess it's protected but it'll make it much harder to identify real CP.

There have been convictions based on child heads photoshopped onto adult bodies. Generative images will be treated the same. The lack of a real victim is treated as immaterial. All it needs is sufficient realism to be believable.

This is an older supreme court case (90s/2000s I think) but my recollection is that you've got it totally backwards--the decision turned on whether or not there is a real victim. Their reasoning was that videos of real kids create a market for the exploitation of real kids, which is a harm that overcomes first amendment protections, whereas depictions need not (their reasoning, not mine). I don't believe verisimilitude was even addressed. Now, if you mean a depiction's similarity to a particular, real child, I'm not sure how that would play out (or if it was addressed in the decision).
Post reply on HN