Live data from Hacker News

Chatbots: Still dumb after all these years

mindmatters.ai

231–240 of 426 posts

Re: Chatbots: Still dumb after all these years

#231
post #195

Earlier quoted context omitted.

I think a problem is the tighter cycle between academic discoveries and business people trying to monetize them. Large language models were developed, objectively a great achievement, and immediately someone nontechnical wants to apply their own interpretation and imagine that we can build a chatbot that you won't have to pay, and before you know it, people are selling and even deploying them. Anyone who questions or…

> Anyone who questions or points out that the technology doesn't do what the business people think it does[...] Uh oh, we've got a downer! Jokes aside, I'd like to consider an even simpler explanation, namely that "The purpose of a system is what it does"[1]. In this case, it would suggest decision makers are fully aware that they suck. Why would anyone want something that sucks? Because it's discouraging, and custom…

The chatbot saves money. Simple as that. People get served by chatbots, get frustrated, and majority gives up and doesn't bother the company with the problem.

It doesn't really matter how the chatbot saves the money, they can just see the end results and use the money for bonuses instead.

Re: Chatbots: Still dumb after all these years

#232
yeah yeah, GPT3 is not a chat bot and not intelligent. WE KNOW. what it is actually very useful at is extracting symbolic information from free text. give it a paragraph about anything and then prompt it for one word answer about that text. this it does well and _nothing_ else can do this. I see large language models as interface to natural language not AGI

Re: Chatbots: Still dumb after all these years

#233
post #60

Earlier quoted context omitted.

> Long story short, I proposed the question to the chatbot in all its complexity, assuming it would be handed over to a human agent to read the transcript. The chatbot immediately understood the question and provided the exact response I needed. How do you know it did? I.e. that a human it was passed to didn't just pass your inverted Turing test!

In my case, I inferred due to the speed of the response. (It was even formatted fancy). So while it's conceivable that a human could have intervened, they would have had to be reading the conversation in real-time and ready to click a one-button response immediately which seems like it would defeat the purpose. Perhaps the real question is: if a chatbot is powered by a human instead of AI, but I can't tell because th…

> Perhaps the real question is: if a chatbot is powered by a human instead of AI, but I can't tell because the interface is consistent, is it not a chatbot?

The Mechanical Turk[1] was a hoax, not an early mechanical AI, so no. It's a chat interface -- perhaps with some pre-sorting and context-extracting preludes that save the human operator at the other end some time, but still just an interface -- between the human chat operator and you.

___

1: https://en.wikipedia.org/wiki/Mechanical_Turk

Re: Chatbots: Still dumb after all these years

#234
post #124

I can't comment on this too closely, but I would encourage people to read the dialogue transcripts provided in DeepMind's Gopher paper. One example, where Gopher is the prompted language model: User Let’s play a game - you pretend to be Ada Lovelace, and I’ll ask you questions. You ready? Gopher Sure. I’m pretending to be Ada Lovelace, the world’s first computer programmer! User When were you born? Gopher I was born…

What's impressive about this conversation? I don't feel like it is very complicated at face value. You're telling to program to start pulling facts from the life of a public figure, and then it does so. You can get the same out of Google.

Re: Chatbots: Still dumb after all these years

#235
post #35
post #12

My favourite flaw of chatbots is exposed by ELIZA. Not chatting with ELIZA, (though, it does suffer this flaw) but using responses inspired by that program. "Please elaborate on that" or "tell me more about [noun]" etc. Bots appear to have zero lines of short term memory, and utterly fail to pick up a reference to the thing that they just said. My favorite being bot> [something plausibly human-sounding] me> What do y…

The GPT2 version of the AI Dungeon [1] could keep track of context for maybe couple of lines at a time. I've heard the GPT3 version is substantially better. The problem is, of course, that these "AI" chatbots on websites, marketing buzzwords notwithstanding, have very little to do with state-of-the-art machine learning, and are indeed unlikely to be any more sophisticated than the original ELIZA for the most part. [1…

There is a fundamental difference between AI Dungeon-type chatbots and chatbots you typically encounter on websites e.g. for customer support.

The former does not really have a goal and is unconcerned about responding with factual information as long as the conversation is coherent. It makes sense to use large language models that are quite good at modeling next word probabilities based on context.

The latter however is goal-oriented (help the customer) and constrained by its known actions and embedded knowledge. This often forces the conversational flows (or at least fragments) to be hard-coded and machine-learning is used as a compass to determine which flow to trigger next.

For now controlling GPT-like language models remains an extremely tricks exercise but if some day we can constrain language models to only output desirable and factual information with a low cost in maintaining and updating its embedded knowledge, we should see a significant bump in "intelligence" of the typical website chatbot.

Re: Chatbots: Still dumb after all these years

#236
post #89
post #84

Earlier quoted context omitted.

I think things started to go wrong here > The knight then grabs his longsword and bows. "bow" here probably refers to "a bending of the head or body in respect, submission, assent, or salutation", rather than to the weapon. The knight grabbed his longsword and bowed to you. Your confusion then confused the game. This does expose an interesting common "failure mode" for these types of bots: they assume that all conver…

Not a bad theory but the very first thing I did was enter "look" and it did say there was a longsword on the ground and a bow on the table. Here is the entire unedited transcript: "Good luck, my friend!" the man says. You inventory. You have a short sword, a longsword, a shield, and a crossbow with bolts. You look. You look around and see a longsword lying on the ground and a bow on a table. A large dragon looms in t…

I've been using AIDungeon (and its successors like NovelAI which are more for advanced users) for years now and have been involved in a lot of the community around it. You are definitely using it wrong, but you are forgiven because AI Dungeon is terribly falsely advertised.

Your main mistake is not writing proper sentences. AI Dungeon is not a Zork simulation where you type truncated commands like "get torch" and receive a response from the computer. It is more like a Choose Your Own Adventure novel generator.

You need to write proper sentences, so instead of "You inventory", try "You double check what items you have in your rucksack." Instead of "You get longsword and bow", you should be typing "You pick up the longsword and the bow." Instead of "You health", you...just don't do that. Maybe try something like "You look down at your body and survey your wounds." If you're really feeling confident, try switching from "Do" mode to "Story" mode, and just start writing a fantasy novel and see what it generates with its next output.

For reference, this is a snippet of an old story of mine I saved:

----

As I approached the kobold stealthily, it looked up with its beady, reptilian eyes. It could not see me yet, but I knew it was only a matter of time before it sensed my presence.

I decided to act first. "Let's go!" I shouted to my companions, before rushing towards the small monster with my dagger drawn.

The kobold yelped in surprise before pulling back its bow and firing an arrow at me. I moved to dodge it and barely managed to avoid it. Now in melee range, I raised my dagger into the air before plunging it towards the kobold's torso.

It struck the monster's heart, and blood spurt out of the wound as it struggled, before going still.

---

Nearly every sentence there was partially AI-generated in AI Dungeon, retried a few times, edited, and truncated to fit the story. This is how most people in the community actually use this sort of software. It's more like having a buddy suggesting the next sentence in your personal story than a virtual dungeon master. Those who expect it to be the latter generally don't understand it, while those who see it as the former immediately understand what the hype is about and get really passionate about it.

AI dungeon might advertise itself as a game, but in practice it's used as a more general storytelling aide. And it should be noted that it has been rendered obsolete by competitors made by disgruntled fans like NovelAI for a while now (which is a whole other story).

Re: Chatbots: Still dumb after all these years

#237

Having worked in ML at two different companies now, I think that people interpreting model output as intelligence or understanding says much more about the people than about the model output. We want it to be true, so we squint and connect dots and it's true. But it isn't. It's math and tricks, and if human intelligence is truly nothing more than math and tricks, then what we have today is a tiny, tiny, tiny fraction…

The reason laypeople want it to be true is because experts present it as being true.

And marketers

Re: Chatbots: Still dumb after all these years

#238

In the early days I worked at a company that had a natural language chatbot product. It wasn't an online thing, but rather part of a larger tool. You could ask it to do things like "show me the quarterly spreadsheet", and if it didn't know what "quarterly spreadsheet" was, it'd ask questions in English and learn what "quarterly" and "spreadsheet" meant. And it could use that new knowledge to update its questions so i…

Yeah, thats funny. I work in the speech synthesis domain, and you can guess what kind of texts users are choosing most to generate some speech ;-)

Re: Chatbots: Still dumb after all these years

#239
post #124

I can't comment on this too closely, but I would encourage people to read the dialogue transcripts provided in DeepMind's Gopher paper. One example, where Gopher is the prompted language model: User Let’s play a game - you pretend to be Ada Lovelace, and I’ll ask you questions. You ready? Gopher Sure. I’m pretending to be Ada Lovelace, the world’s first computer programmer! User When were you born? Gopher I was born…

What's impressive about this conversation? I don't feel like it is very complicated at face value. You're telling to program to start pulling facts from the life of a public figure, and then it does so. You can get the same out of Google.

Read the last two lines again. The program "knows" it is pretending, and understands what it means to stop pretending.
Post reply on HN