He should do a follow-up on GPT-4.
BTW, on a different note, would people consider Gwern the best writer in the rationalist community or are Yud and Slate Star Codex considered better?
BTW, on a different note, would people consider Gwern the best writer in the rationalist community or are Yud and Slate Star Codex considered better?
A lot of the criticism (almost all of it) you read on the quality of various language models is people trying out chat gpt without doing that. And while it's alright, it indeed has many flaws. Which a lot of people are quick to point out before they give up.
Relative to that gpt-4 is trained on more data, more languages, less inclined to hallucinate (it still does it sometimes but a lot less), and if you know how to access this, it knows how to use tools via plugins.
After you figure out how to access gpt-4 properly, it mostly boils down to knowing what to ask and actually thinking of asking it to begin with. And then you need to follow up with more questions to refine it, etc. Shit in, shit out principle basically. Using it properly is actually a lot of work and it only makes sense if your need is big enough. It's like every other tool really. Having the tool doesn't make you magically better until you learn how to use it properly. Asking naive questions to which you already know the answers is not a productive way to use it, or learn how to use it. It gets better when you step outside your comfort zone and ask it the things you don't know that you need to know. The more specific your request, the more helpful it gets.
A big limitation is actually the UX. Chat is very accessible but not necessarily very user friendly. In a way it was a happy accident for openai that it works so well. But there are tools and extensions that provide you a better experience. For example with code gpt (configured to use gpt-4), you can get it to critique code selections, suggest improvements, documentation, etc. All you need to do is ask it to. With gpt for docs, you can select bits of text and ask it to improve it, critique it, expand on it, suggest counter arguments, additional arguments, supporting facts, translate it, simplify it, etc. A good writer will be able to ask better questions and get better results. The more text you select, the slower it gets. Use it to brainstorm, explore topics, refine, etc.
Agreed, but I don't think this goes broad or deep enough, particularly with respect to the power of iteration and recursion in conversation: the basics of reflection over prior action. The key here is to realize that there's a natural sensemaking process that goes from action to reflection to repetition and that sensemaking process is a collaboration between computer agents and human agents whether it remains implicit or explicit in design and whether the computer agent is programmed to recognize and resolve conflict and consensus in not just subjective but objective fact-or-fiction. Right now, we aren't even at the latter, let alone the former. That will come with time and acceptance of the veracity problem that is writ large in the AI UX currently.
> it mostly boils down to knowing what to ask and actually thinking of asking it to begin with.
Indeed, recognition vs. recall is as old as GUI vs. CLI. However, the key is to realize that some folks will be faster through recall because they've actually built up the knowledge to recall vs the symbolism to recognize.
My argument is that, for me, there's little value in chat because that's not how I prefer to interact with any human or computer agent via writing. It's too slow. I much prefer the batch mode where there is larger latency between question and answer, but that latency provides an emergent cadence to the conversation that is more natural, providing room for listening-and-thought-before-response. This may sound old-fashioned like punch cards, but conversation has been missing from many technical things for years and we've paid the price for that silence.
Cohercing them in specific output is becoming harder and harder, and postprocessing to cut the fat tedious to maintain as unreliable, plus who wants that on their pipeline.
So yeah they are fine for having a chat about test passage but as authoring tools are heavy,slow,and require lot of manually moving strings back and forth.
(Nevermind that you pay the token to generate that "as an ai language model I can" warning)
Same with Midjourney. I’m not a discord user, but it is so painful to use compared to the automatic1111+extensions that scratch the surface of what a powerful UI can be a given generative AI use case.
My suggestion: Create a developer account on platform.openai.com and use their 'playground'. You pay per 1k tokens; I've been using it fairly regularly, and with GPT-4 it usually ends up being $0.10-$0.30 per day. You can't use plugins however.
In terms of popular impact, SSC obviously wins. However among "insiders" there are always implicitly lauded favorites. Is it about the general capacity to convey complex topics succinctly, general "immersion" in the writing and ability to consistently foster a compelling mind-space via the imagery and ideas handled in the text, etc?
This is just me being a curious outsider. Interestingly I often peer into insular communities and find that there's often an unsaid tapestry of implicit assumptions and customs. All that I've discovered so far is that SSC is the darling favorite, Yud is liked but also made fun of for his behavior and more blunt approach, and Gwern is liked but not as notable, though imo he has the best website.
Is the best way to peer into these ideas via a prolonged investigative type outside observation, or should one feel only a worthy potential to fit in to the common character of the crowd if they can take a short peek at the general shape of the culture and immediately "get it"?
Yudkowsky obviously thought about AI things so much, and he has many interesting ideas and he is the highest status poster on lesswrong dot com, but he admits he is not a great writer and I think not all his ideas are so great. For example I like Paul Christiano's ideas more (https://news.ycombinator.com/item?id=35635345) although I have to credit Yudkowsky bc without such debate I am sure Christiano's ideas wouldn't have been put as clearly or publicly. Here's one of Yudkowsky's own tweets (March 24 2023) where he self-deprecatingly roasts his own bad writing by showing how GPT-4 writes it in a clearer way https://twitter.com/ESYudkowsky/status/1639425421761712129
Slate Star Codex I don't like as much as either one and I don't think it has such new or interesting ideas, sorry it's only my opinion.
People are in the denial stage right now.