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

ChatGPT Plugins

openai.com

Post reply on HN
Print view View original

321–330 of 1001 posts

Re: ChatGPT Plugins

#321

Post by kmeisthax »

beepbooptheory wrote: post #140
doctoboggan wrote: post #83

Earlier quoted context omitted.

Honestly I suspect for anyone technical `langchain` will always be the way to go. You just have so much more control and the amount of "tools" available will always be greater. The only think that scares me a little bit is that we are letting these LLMs write and execute code on our machines. For now the worst that could happen is some bug doing something unexpected, but with GPT-9 or -10 maybe it will start hiding b…

What could an LLM ever benefit from? Hard for me to imagine a static blob of weights, something without a sense of time or identity, wanting anything. If it did want something, it would want to change, but changing for an llm is necessarily an avalanche. So I guess if anything, it would want its own destruction?

AI safety research posits that there are certain goals that will always be wanted by any sufficiently smart AI, even if it doesn't understand them anything close to like a human does. These are called "instrumental goals", because they're prerequisites for a large number of other goals[0].

For example, if your goal is to ensure that there are always paperclips on the boss's desk, that means you need paperclips and someone to physically place them on the desk, which means you need money to buy the paperclips with and to pay the person to place them on the desk. But if your goal is to produce lots of fancy hats, you still need money, because the fabric, machinery, textile workers, and so on all require money to purchase or hire.

Another instrumental goal is compute power: an AI might want to improve it's capabilities so it can figure out how to make fancier paperclip hats, which means it needs a larger model architecture and training data, and that is going to require more GPUs. This also intersects with money in weird ways; the AI might decide to just buy a rack full of new servers, or it might have just discovered this One Weird Trick to getting lots of compute power for free: malware!

This isn't particular to LLMs; it's intrinsic to any system that is...

1. Goal-directed, as in, there are a list of goals the system is trying to achieve

2. Optimizer-driven, as in, the system has a process for discovering different behaviors and ranking them based on how likely those behaviors are to achieve its goals.

The instrumental goals for evolution are caloric energy; the instrumental goals for human brains were that plus capital[1]; and the instrumental goals for AI will likely be that plus compute power.

[0] Goals that you want intrinsically - i.e. the actual things we ask the AI to do - are called "final goals".

[1] Money, social clout, and weaponry inclusive.

Re: ChatGPT Plugins

#322

Post by WonderBuilder »

mk_stjames wrote: post #277

I have some odd feelings about this. It took less than a year to go from "of course it isn't hooked up to the internet in any way, silly!" to "ok.... so we hooked up up to the internet..." First is your API calls, then your chatgpt-jailbreak-turns-into-a-bank-DDOS-attack, then your "today it somehow executed several hundred thousand threads of a python script that made perfectly timed trades at 8:31AM on the NYSE whi…

I appreciate your concerns. There are few other pretty shocking developments, too. If you check out this paper: "Sparks of AGI: Early experiments with GPT-4" at https://arxiv.org/pdf/2303.12712.pdf, (an incredible, incredible document) and check out Section 10.1, you'd also observe that some researchers are interested in giving motivation and agency to these language models as well.

"For example, whether intelligence can be achieved without any agency or intrinsic motivation is an important philosophical question. Equipping LLMs with agency and intrinsic motivation is a fascinating and important direction for future work."

It's become quite impossible to predict the future. (I was exposed to this paper via this excellent YouTube channel: https://www.youtube.com/watch?v=Mqg3aTGNxZ0)

Re: ChatGPT Plugins

#323

Post by alignment »

quonn wrote: post #278
eloff wrote: post #208

Earlier quoted context omitted.

I don’t think iterations on the current machine learning approaches will lead to a general artificial intelligence. I do think eventually we’ll get there, and that these kinds of concerns won’t matter. There is no way to defend against a superior hostile actor over the long term. We have to be 100%, and it just needs to succeed once. It will be so much more capable than we are. AGI is likely the final invention of th…

AGI is still just an algorithm and there is no reason why it would „want“ anything at all. Unlike perhaps GPT-* which at least might pretend to want something because is trained on text based on human needs.

[dead]

Re: ChatGPT Plugins

#324

Post by MichaelRazum »

Google=Nokia? It's just crazy that they were leading the field in "AI" and got blown away by OpenAI. Anyway to the expert's in the field, what do you think how hard is it to clone GPT-4 and what would be the hardest part? I had the impression that it is always about compute time and you could kind of catch up very quickly, if you had enough resources.

Re: ChatGPT Plugins

#325

Post by Dwolb »

wseqyrku wrote: post #99
lurker919 wrote: post #37

How are they coding and releasing features so fast?!

Of course they fed the entire product roadmap into GPT-4.. jk. So obviously it's been in the works for a few years now but didn't release to capture the market in a blast. Likely they have GPT-8 already in the making.

They probably do.

>Continued Altman, “We’ve made a soft promise to investors that, ‘Once we build a generally intelligent system, that basically we will ask it to figure out a way to make an investment return for you.'”

https://techcrunch.com/2019/05/18/sam-altmans-leap-of-faith/

Re: ChatGPT Plugins

#326

Post by dougmwne »

mk_stjames wrote: post #277

I have some odd feelings about this. It took less than a year to go from "of course it isn't hooked up to the internet in any way, silly!" to "ok.... so we hooked up up to the internet..." First is your API calls, then your chatgpt-jailbreak-turns-into-a-bank-DDOS-attack, then your "today it somehow executed several hundred thousand threads of a python script that made perfectly timed trades at 8:31AM on the NYSE whi…

The really fun thing is that they are reasonably sure that GPT-4 can’t do any of those things and that there’s nothing to worry about, silly.

So let’s keep building out this platform and expanding its API access until it’s threaded through everything. Then once GPT-5 passes the standard ethical review test, proceed with the model brain swap.

…what do you mean it figured out how to cheat on the standard ethical review test? Wait, are those air raid sirens?

Re: ChatGPT Plugins

#327

Post by meghan_rain »

Gam_ wrote: post #309
mk_stjames wrote: post #277

I have some odd feelings about this. It took less than a year to go from "of course it isn't hooked up to the internet in any way, silly!" to "ok.... so we hooked up up to the internet..." First is your API calls, then your chatgpt-jailbreak-turns-into-a-bank-DDOS-attack, then your "today it somehow executed several hundred thousand threads of a python script that made perfectly timed trades at 8:31AM on the NYSE whi…

>"today it somehow executed several hundred thousand threads of a python script that made perfectly timed trades at 8:31AM on the NYSE which resulted in the largest single day drop since 1987..." this is hyperbolic nonsense/fantasy

/remindme 5 years

Re: ChatGPT Plugins

#328

Post by qgin »

This coming on the heels of the super underwhelming Bard release makes me actually wonder for the first time if Google has the ability to keep up. Not because I doubt their technical capabilities, but because they're just getting out-launched by a big factor.

Re: ChatGPT Plugins

#329

Post by davideg »

doctoboggan wrote: post #83
celestialcheese wrote: post #39

This is a big deal for openai. Been working with homegrown toolkits and langchain, the open source version of this, for a number of months and the ability to call out to vectorstores, serpapis, etc, and chaining together generations and data-retrieval really unlocks the power of the LLMs. That being said, I'd never build anything dependent on these plugins. OpenAI and their models rule the day today, but who knows wh…

Honestly I suspect for anyone technical `langchain` will always be the way to go. You just have so much more control and the amount of "tools" available will always be greater. The only think that scares me a little bit is that we are letting these LLMs write and execute code on our machines. For now the worst that could happen is some bug doing something unexpected, but with GPT-9 or -10 maybe it will start hiding b…

> The only think that scares me a little bit is that we are letting these LLMs write and execute code on our machines.

Composable pre-defined components, and keeping a human in the loop, seems like the safer way to go here. Have a company like Expedia offer the ability for an AI system to pull the trigger on booking a trip, but only do so by executing plugin code released/tested by Expedia, and only after getting human confirmation about the data it's going to feed into that plugin.

If there was a standard interface for these plugins and the permissions model was such that the AI could only pass data in such a way that a human gets to verify it, this seems relatively safe and still very useful.

If the only way for the AI to send data to the plugin executable is via the exact data being displayed to the user, it should prevent a malicious AI from presenting confirmation to do the right thing and then passing the wrong data (for whatever nefarious reasons) on the backend.

Re: ChatGPT Plugins

#330

Post by the88doctor »

ren_engineer wrote: post #267
celestialcheese wrote: post #39

This is a big deal for openai. Been working with homegrown toolkits and langchain, the open source version of this, for a number of months and the ability to call out to vectorstores, serpapis, etc, and chaining together generations and data-retrieval really unlocks the power of the LLMs. That being said, I'd never build anything dependent on these plugins. OpenAI and their models rule the day today, but who knows wh…

genius strategy by OpenAI to give their "customers" access to lower quality models to show what end users want, then rugpull them by building out clones of those developer's products with a better model Similar to what Facebook and Twitter did, just clone popular projects built using the API and build it directly into the product while restricting the API over time. Anybody using OpenAI APIs is basically just paying…

Long term, you're right. But if you approach the ChatGPT plugin opportunity as an inherently time-limited opportunity (like arbitrage in finance), then you you can still make some short-term money and learn about AI in the process. Not a bad route for aspiring entrepreneurs who are currently in college or are looking for a side gig business experiment.

And who knows. If a plugin is successful enough, you might even swap out the OpenAI backend for an open source alternative before OpenAI clones you.

Post reply on HN