ChatGPT is a blurry JPEG of the web • Hacker News Board

Re: ChatGPT is a blurry JPEG of the web

#221

Post by dgb23 » Thu, Feb 09, 2023, 8:57 PM UTC

majormajor wrote: post #192

secabeen wrote: post #32

This is a decent summary. I've been thinking about how ChatGPT by it's very nature destroys context and source reputation. When I search for something on the Internet, I get a link to the original content, which I can then evaluate based on my knowledge and the reputation of the original source. Wikipedia is the same, with a big emphasis on citation. ChatGPT and other LLMs destroy that context and knowledge, giving m…

I would love to know their plan for having new facts propagate into these models. My idle speculation makes me think this is a hard problem. If ChatGPT kills Search it also kills the websites that get surfaced by search that were relying on money from search-directed users. So stores are fine, but "informational" websites are probably in for another cull. Paywall premium publications are probably still fine - the peo…

I almost never find these kind of websites with search, except I already know there might be one on a specific topic.

The way I find them is from forums, chat, links from other such sites. All goes into my RSS reader.

I use search with !wiki or !mdn etc. most of the time.

Re: ChatGPT is a blurry JPEG of the web

#222

Post by avereveard » Thu, Feb 09, 2023, 9:01 PM UTC

bbor wrote: post #211

avereveard wrote: post #72

Earlier quoted context omitted.

Want niche use case? I feed chatgpt the whole warstuff one page rule miniature combat system, and I can ask it to generate thematic units. I generated a wide variety of content from LOTR to ships. it even created new trait to complement the existing one when needed by special units, along with explanation of the mechanics of the trait. It doesn't quite understand positioning, but it will simulate round of combat betw…

You’re by far the most interesting person in this thread - I’ve been coding up a Warhammer 40k (symbolic) AI for a while, but I bet chatgpt could blow it out of the water…

well warstuff rules are simple and I gues they fit nicely in gpt memory the ruleset is here and is magnitudes simpler than warhammer https://onepagerules.com/portfolio/one-off-games/ - I just copy pasted the whole ruleset into a message prepended by "Consider a wargaming system built with these rules:"

this is a sample prompt afterward:

> create a set of units using these rules, themed around warhammer 40k

this is what chatgpt came up with, it's impressive as I just set a theme, and it dig up appropriate original special rules that fit within the ruleset:

Sure, here are three examples of units themed around Warhammer 40k that you can use in your wargaming system:

    Space Marines: Quality 4, Special Rules: Shooter, Tough, Cost: 50pts

    The Space Marines are a heavily armored and well-trained fighting force, with a Quality value of 4 reflecting their combat abilities and morale. They have the Shooter special rule, allowing them to fire ranged weapons, and the Tough special rule, making them more resistant to damage. Their point cost is 50, calculated by multiplying Quality 4 by 10 and adding the point cost of the special rules.

    Chaos Space Marines: Quality 3, Special Rules: Shooter, Daemonic, Cost: 40pts

    The Chaos Space Marines are a corrupted version of the Space Marines, with a lower Quality value of 3 but the same Shooter and Daemonic special rules. Daemonic units are immune to morale tests and can only be killed by a roll of 6 on the melee attack. Their point cost is 40, calculated by multiplying Quality 3 by 10 and adding the point cost of the special rules.

    Necron Warriors: Quality 2, Special Rules: Shooter, Reanimation, Cost: 30pts

    The Necron Warriors are an undead race of robots, with a Quality value of 2 reflecting their lower morale and combat abilities. They have the Shooter special rule, allowing them to fire ranged weapons, and the Reanimation special rule, allowing them to automatically recover from being stunned instead of taking the Recover action. Their point cost is 30, calculated by multiplying Quality 2 by 10 and adding the point cost of the special rules.

Re: ChatGPT is a blurry JPEG of the web

#223

Post by pegasus » Thu, Feb 09, 2023, 9:03 PM UTC

thundergolfer wrote: post #45

> ChatGPT is so good at this form of interpolation that people find it entertaining: they’ve discovered a “blur” tool for paragraphs instead of photos, and are having a blast playing with it. “‘blur’ tool for paragraphs” is such a good way of describing the most prominent and remarkable skill of ChatGPT. It is fun, but so obviously trades off against what makes paragraphs great. It is apt that this essay against Chat…

What he misses in this analogy is that part of what produces the "blur" is the superimposing of many relevan paragraphs found on the web into one. This mechanism can be very useful, because it could average out errors and give one a less one-sided perspective on a particular issue. It doesn't always work like this, but hopefully it will more and more. Also, even more useful is to do a cluster analysis of the existant perspectives and give a representative synthesis of each of these, along with a weight representing their popularity. So there's a lot of room for improvement, but the potential in my opinion is there.

Re: ChatGPT is a blurry JPEG of the web

#224

Post by bbor » Thu, Feb 09, 2023, 9:04 PM UTC

omnicognate wrote: post #86

bbor wrote: post #14

Ugh I’m beginning to think I’m going to spend the next 6-12 months commenting “no, large language models aren’t supposed to somehow know everything in the world. No, that’s not what they’re designed for. Yes, hooking one up to our long-standing record-of-everything-in-the-world (google’s knowledge graph) is going to be powerful.” It’s getting to point where I need to consider stop going on HN. This is like when my fa…

> Yes, hooking one up to our long-standing record-of-everything-in-the-world (google’s knowledge graph) is going to be powerful. This hasn't happened yet and, while I may just lack imagination, despite having a fairly solid understanding of how the latest round of AI works I can't see how it can be done successfully. Until it is in fact done and my lack of imagination is demonstrated, your "going to be powerful" is a…

In my view it’s very simple, which is what makes it so exciting. Summarizing a design doc that I imagine Microsoft and google are both spending millions of dollars of manhours/day working on their versions of:

1. User enters query.

2. LLM augments query if necessary, adding extra terms or clauses.

3. Normal search pipeline returns ranked links, just like it does now.

4. LLM reads the content of the first 100 links and decides which are the best based on your stated preferences and past behavior, uses that to augment the ranking a bit.

5. LLM generates various summaries depending on the type of query, such as laying out a few common answers to a controversial political question or giving a summary of a technical Wikipedia article tailored to your expertise level in that field.

6. Finally, for a tiny subset of queries, maybe the user wants to converse with the AI in a chat-like format, where it cites all of its claims with direct links.

It’s gonna be awesome :)

Re: ChatGPT is a blurry JPEG of the web

#225

Post by galleywest200 » Thu, Feb 09, 2023, 9:06 PM UTC

zetazzed wrote: post #17

Damn, I hate to plug products on HN, but I'd say that the New Yorker is the one subscription I've loved maintaining throughout my life. First got it right out of college and appreciate it 20 years later. Everyone is publishing think pieces about ChatGPT - yawn. But only the New Yorker said, hmm, how about if we get frickin' Ted Chiang to write a think piece? (It is predictably very well written.)

For anybody who does not want to pay, your taxes likely already pay for a subscription you can use from your local library on the Libby app. King and Snohomish counties in Washington provide unlimited digital copies of The New Yorker and The Economist as an example.

Re: ChatGPT is a blurry JPEG of the web

#226

Post by pegasus » Thu, Feb 09, 2023, 9:06 PM UTC

nneonneo wrote: post #84

This is very well written, and probably one of my favorite takes on the whole ChatGPT thing. This sentence in particular: > Indeed, a useful criterion for gauging a large-language model’s quality might be the willingness of a company to use the text that it generates as training material for a new model. It seems obvious that future GPTs should not be trained on the current GPT's output, just as future DALL-Es should…

That part made the least sense for me. Since a more advanced version of a LLM would be better at extracting the truth of things from the given data, what could it possibly gain from ingesting the output of a less precise version of itself? It couldn't ever add anything useful, almost by definition.

Re: ChatGPT is a blurry JPEG of the web

#227

Post by causalmodels » Thu, Feb 09, 2023, 9:12 PM UTC

boh wrote: post #142

Does anyone have any idea how ChatGPT will actually make money? As novel as it is to use with all the "potential" applications, the possible revenue streams don't seem to prop up the recent investments into OpenAI. We've already been through enough hype cycles in the past ten years to realize "potential" use-cases or user counts don't necessarily produce a sustainable business model. Nor does a new innovative thing n…

Well OpenAI's explicit goal is AGI. While this is farfetched, it would clearly make them the most valuable company in the history of the world and the ability to monetize something like AGI would essentially be unlimited.

OpenAi isn't really interested in making money from selling products. So instead they have worked out a deal with MS where they give them the best new tools they can come up with in exchange for tons of cash and compute. They are basically a research lab of which MS owns a significant, but not controlling, share. ChatGPT itself is probably not going to make any money. It is a very cool tech demo that functions as an ad for MS products. The money making part comes in when MS starts to bundle OpenAI models into MS products which then get sold to businesses. And if there is one thing MS is very good at, it is selling software to businesses.

Re: ChatGPT is a blurry JPEG of the web

#228

Post by foruhar » Thu, Feb 09, 2023, 9:12 PM UTC

andrewflnr wrote: post #125

nneonneo wrote: post #84

This is very well written, and probably one of my favorite takes on the whole ChatGPT thing. This sentence in particular: > Indeed, a useful criterion for gauging a large-language model’s quality might be the willingness of a company to use the text that it generates as training material for a new model. It seems obvious that future GPTs should not be trained on the current GPT's output, just as future DALL-Es should…

I think what makes AlphaZero's recursion work is the objective evaluation provided by the game rules. Language models have no access to any such thing. I wouldn't even count user-based metrics of "was this result satisfactory": that still doesn't measure truth. I generally respect the heck out of Chiang but I think it's silly to expect anyone to be happy feeding a language model's output back into it, unless that out…

There’s a version of this where the output is mediated by humans. Currently chatgpt has a thumbs up/down UI next to each response. This feedback could serve as a signal for which generated output may be useful for future ingestion. Perhaps OpenAI is already doing this with our thumb signals.

Re: ChatGPT is a blurry JPEG of the web

#229

Post by fooker » Thu, Feb 09, 2023, 9:13 PM UTC

andrewflnr wrote: post #125

nneonneo wrote: post #84

This is very well written, and probably one of my favorite takes on the whole ChatGPT thing. This sentence in particular: > Indeed, a useful criterion for gauging a large-language model’s quality might be the willingness of a company to use the text that it generates as training material for a new model. It seems obvious that future GPTs should not be trained on the current GPT's output, just as future DALL-Es should…

I think what makes AlphaZero's recursion work is the objective evaluation provided by the game rules. Language models have no access to any such thing. I wouldn't even count user-based metrics of "was this result satisfactory": that still doesn't measure truth. I generally respect the heck out of Chiang but I think it's silly to expect anyone to be happy feeding a language model's output back into it, unless that out…

> Language models have no access to any such thing.

And this is exactly why MS is in such a hurry to integrate it into Bing. The feedback loop can be closed by analyzing user interaction. See Nadella’s recent interview about this.

Re: ChatGPT is a blurry JPEG of the web

#230

Post by bbor » Thu, Feb 09, 2023, 9:14 PM UTC

CPLX wrote: post #107

bbor wrote: post #74

Earlier quoted context omitted.

Hmm. I’d say definitely yes. I mean we are on the internet right now, presumably across many km of distance. Do you disagree?

I have mixed feelings about it. I do believe with certainty that there are many people, millions, whose life has been made substantially worse by the invention of ubiquitous computing devices. Probably the minority. But I’d say his questions (ie “What’s in it for me?”) are excellent ones to pose in the face of new technology.

I think this is anti-status quo bias, which for once is unjustified. Sure the world has many systemic problems, but to say that the MASSIVE efficiency gains the internet have brought us are for a minority seems like an insane claim. Just because capitalists have been claiming all the new excess value doesn’t mean it doesn’t exist.

To illustrate, consider: spreadsheets, video chat, shared documents, automated calendars, maps, autopilot assistants, CGI, I could go on… the world is not quite good enough that I’m ready to say that we don’t need more efficiency. Not everyone has a home on mars quite yet.