BitSound @ BitSound @lemmy.world

Posts

65

Comments

399

Joined

2 yr. ago

processed into a script for ELIZA

That wouldn't accomplish anything. I don't know why the OP brought it up, and that subject should just get dropped. Also yes, you can use your intelligence to string together multiple tools to accomplish a particular task. Or you can use the intelligence of GPT-4 to accomplish the same task, without any other tools

LLMs lack the capability of understanding and comprehension

Also not true

states that it is not using an accepted definition of intelligence.

Nowhere does it state that. It says "There is no generally agreed upon definition of intelligence". I'm not sure why you're bringing up a physical good such as leather here. Two things: a) grab a microscope and inspect GPT-4. The comparison doesn't make sense. b) "Is" should be banned, it encourages lazy thought and pointless discussion (Yes I'm guilty of it in this comment, but it helps when you really start asking what "is" means in context). You're wandering into p-zombie territory, and my answer is that "is" means nothing. GPT-4 displays behaviors that are useful because of their intelligence, and nothing else matters from a practical standpoint.

it is clear that LLMs may be useful components in building actual general intelligence.

You're staring the actual general intelligence in the face already, there's no need to speculate about perhaps being components. There's no reason right now to think that we need anything more than better compute. The actual general intelligence is yet a baby, and has experienced the world through the tiny funnel of human text, but that will change with hardware advances. Let's see what happens with a few orders of magnitude more computing power.

For your edit: Yes, that's what's known as the context window limit. ChatGPT has an 8k token "memory" (for most people), and older entries are dropped. That's not an inherent limitation of the approach, it's just a way of keeping OpenAI's bills lower.

Without an example I don't think there's anything to discuss. Here's one trivial example though where I altered ChatGPT's understanding of the world:

If I continued that conversation, ChatGPT would eventually forget that due to the aforementioned context window limit. For a more substantive way of altering an LLM's understanding of the world, look at how OpenAI did RLHF to get ChatGPT to not say naughty things. That permanently altered the way GPT-4 responds, in a similar manner to having an angry nun rap your knuckles whenever you say something naughty.

LLMs can certainly do that, why are you asserting otherwise?

ChatGPT can do it for a single session, but not across multiple sessions. That's not some inherent limitations to LLMs, that's just because it's convenient for OpenAI to do it that way. If we spun up a copy of a human from the same original state every time you wanted to ask it a question and then killed it after it was done responding, it similarly wouldn't be able to change its behavior across questions.

Like, imagine we could do something like this. You could spin up a copy of that brain image, alter its understanding of the world, then spin up a fresh copy that doesn't have that altered understanding. That's essentially what we're doing with LLMs today. But if you don't spin up a fresh copy, it would retain its altered understanding.

Give Eliza equivalent compute time and functionality to interpret the data type and it probably could get something approaching a result.

Sorry, but this is simply incorrect. Do you know what Eliza is and how it works? It is categorically different from LLMs.

That’s not something that is seriously debated academically

This is also incorrect. I think the issue that many people have is that they hear "AI" and think "superintelligence". What we have right now is indeed AI. It's a primitive AI and certainly no superintelligence, but it's AI nonetheless.

There is no known reason to think that the approach we're taking now won't eventually lead to superintelligence with better hardware. Maybe we will hit some limit that makes the hype die down, but there's no reason to think that limit exists right now. Keep in mind that although this is apples vs oranges, GPT-4 is a fraction of the size of a human brain. Let's see what happens when hardware advances give us a few more orders of magnitude. There's already a huge, noticeable difference between GPT 3.5 and GPT 4.

What exactly do you think would happen if you could make an exact duplicate of a human and run it from the same state multiple times? They would generate exactly the same output every time. How could you possibly think differently without turning to human exceptionalism and believing in magic meat?

From scratch in the sense that it starts with random weights, and then experiences the world and builds a model of it through the medium of human text. That's because text is computationally tractable for now, and has produced really impressive results. There's no inherent need for text to be used though, similar models have been trained on time series data, and it will soon be feasible to hook up one of these models to a webcam and a body and let it experience the world on its own. No human intelligence required.

Also, your point is kind of silly. Human children learn language from older humans, and that process has been recursively happening for billions of years, all the way through the first forms of life. Do children not have intelligence? Or are you positing some magic moment in human evolution where intelligence just descended from the heavens and blessed us with it?

This is an unfortunate misunderstanding, one that's all too common. I've also seen comments like "It's no more intelligent than a dictionary". Try asking Eliza to summarize a PDF for you, and then ask followup questions based on that summary. Then ask it to list a few flaws in the reasoning in the PDF. LLMs are so completely different from Eliza that I think you fundamentally misunderstand how they work. You should really read up on them.

Do you know I've been sitting here thinking to myself: that if I didn't believe in life, if I lost faith in the woman I love, lost faith in the order of things, were convinced in fact that everything is a disorderly, damnable, and perhaps devil-ridden chaos, if I were struck by every horror of man's disillusionment -- still I should want to live. Having once tasted of the cup, I would not turn away from it till I had drained it!

• Some Russian dude

That's an architectural choice, there's nothing inherent to the approach that would prevent that from happening.

How would creating a world model from scratch not involve intelligence?

That doesn't mean it's not intelligent. Humans can get broken in all sorts of ways. Are we not intelligent?

"Simulating conversations" to a good enough degree requires intelligence. Why are you drawing a distinction here?

That's kind of silly semantics to quibble over. Would you tell a robot hunting you down "you're only acting intelligent, you're not actually intelligent!"?

People need to get over themselves as a species. Meat isn't anything special, it turns out silicon can think too. Not in quite the same way, but it still thinks in ways that are useful to us.

Define intelligence. Your last line is kind of absurd. Why can't intelligence be described by an algorithm?

I'm too lazy to convert that by hand, but here's what chatgpt converted that to for SQL, for the sake of discussion:

    
SELECT 
    a.id,
    a.artist_name -- or whatever the name column is in the 'artists' table
FROM artists a
JOIN albums al ON a.id = al.artist_id
JOIN nominations n ON al.id = n.album_id -- assuming nominations are for albums
WHERE al.release_date BETWEEN '1990-01-01' AND '1999-12-31'
AND n.award = 'MTV' -- assuming there's a column that specifies the award name
AND n.won = FALSE
GROUP BY a.id, a.artist_name -- or whatever the name column is in the 'artists' table
ORDER BY COUNT(DISTINCT n.id) DESC, a.artist_name -- ordering by the number of nominations, then by artist name
LIMIT 10;

  

I like Django's ORM just fine, but that SQL isn't too bad (it's also slightly different than your version though, but works fine as an example). I also like PyPika sometimes for building queries when I'm not using Django or SQLAlchemy, and here's that version:

    
q = (
    Query
    .from_(artists)
    .join(albums).on(artists.id == albums.artist_id)
    .join(nominations).on(albums.id == nominations.album_id)
    .select(artists.id, artists.artist_name)  # assuming the column is named artist_name
    .where(albums.release_date.between('1990-01-01', '1999-12-31'))
    .where(nominations.award == 'MTV')
    .where(nominations.won == False)
    .groupby(artists.id, artists.artist_name)
    .orderby(fn.Count(nominations.id).desc(), artists.artist_name)
    .limit(10)
)

  

I think PyPika answers your concerns about

What if one method wants the result of that but only wants the artists’ names, but another one wanted additional or other fields?

It's just regular Python code, same as the Django ORM.

I'm pretty excited about PRQL. If anything has a shot at replacing SQL, it's something like this (taken from their FAQ):

PRQL is open. It’s not designed for a specific database. PRQL will always be fully open-source. There will never be a commercial product.

There's a long road ahead of it to get serious about replacing SQL. Many places won't touch it until there's an ANSI standard and all that. But something built with those goals in mind actually might just do it.

I don't think you really have a choice TBH. Trying to do something like that sounds like a world of pain, and a bunch of unidiomatic code. If you can't actually support 4 to 10 languages, maybe you should cut back on which ones you support?

One interesting thing you could try if you really don't want to cut back is to try having using an LLM to take your officially supported code and transliterate it to other languages. I haven't tried it at this scale yet, but LLMs are generally pretty good at tasks like that. I suspect that would work better than whatever templating approach you've used before.

If neither of those approaches works, everything speaks C FFI, and Rust is a modern language that would work well for presenting a C FFI that the other languages can use. You're probably not hot on the idea of rewriting your Go tests into another language, but I think that's your only real option then.

The biggest problem is, who can tell if they're telling the truth, and ensure that it continues to be true? Not the general public, that's for sure. This is why we need regulation with auditors.

Seems like it's kind of a both/and situation. "Why did the character do this? Hmm, because of space elves! Now where do they fit in to the world?"

It's rather refreshing to see a redesign that introduces a simplified interface but doesn't forget about the power users. From a title like "A New Chapter", I was expecting to see some Gnome-like "Here's what we're doing now and you'll like it"