News for hackers. Stuff that matters.
Re:HuggingGPT: Solving AI Tasks with ChatGPT and Its Friends... by UniverseHacker on Friday March 31, 2023 @02:11PM (#35390953)
I agree, I suspect AGI is possible right now with a similar system only slightly more sophisticated than this one. The right "glue" for existing models, and plugins to existing data sources all coordinated in a system. GTP-4 would do the managing, and handling, and some simple template API and handler script would allow it to call instances of itself or other models, track recursion depth, and automatically remind GTP-4 to stay on track and always use the correct 'templates' for requests. It could also remind it to create a condensed summary of it's goals and state that gets repeated back automatically to act as long term memory, and increase the effective context window size (edit: this is single variable external memory).

I am afraid to explain this, because I have tried a preliminary version of it that I supervised step by step, and it seems to work. I think it is obvious enough that I won't have been the only one to think of it, so it would be safer to put the information out there so people can prepare.

I see a big disconnect on here between people saying GPT-4 can't do things like this and is just a "stochastic parrot" or "glorified autocomplete," and people posting logs and summaries of it solving unexpectedly hard problems outside of any conceivable training set. My theory is that this disconnect is due to three major factors: * People confusing GPT-4 and GPT-3, as both are called "chatGTP" and most people haven't actually used GPT-4 because it requires a paid subscription, and don't realize how much better it is * Most popular conceptual explanations about how these models work imply that these actually observed capabilities should be fundamentally impossible * Expectations from movies, etc. about what AGI will be like, e.g. that it will never get confused or make mistakes, or that it won't have major shortcomings in specific areas. In practice this doesn't seem to limit it because it recognizes and fixes its mistakes automatically when it sees feedback (e.g. in programming)

Re:HuggingGPT: Solving AI Tasks with ChatGPT and Its Friends... by nomel on Friday March 31, 2023 @02:43PM (#35391442)
> I suspect AGI is possible right now with the right "glue" for existing models, and plugins to existing data sources all coordinated in a system.

I think the regulation of the feedback loops required for sustained, deliberate thought, and interaction with the world, will be the most difficult piece of the AGI puzzle, and might not exist today.

New ideas are "hallucinations" of our "existing data", that we eventually get around to proving. An AGI will require these "hallucinations", inhibition and excitation of them. I think it's going to be a tricky balance [1], for sane output.

1. "Creative minds 'mimic schizophrenia", https://www.bbc.com/news/10154775

Re:HuggingGPT: Solving AI Tasks with ChatGPT and Its Friends... by marcosdumay on Friday March 31, 2023 @02:31PM (#35391284)
I believe you will quickly run into a barrier due to the small size of the context window. And increasing the context window gets harder and harder the larger it is.

There's a sibling talking about internal memory. That's how our brains solve the issue. AFAIK, nobody knows how to train something like it.

Re:HuggingGPT: Solving AI Tasks with ChatGPT and Its Friends... by wyager on Friday March 31, 2023 @02:36PM (#35391361)
I wouldn't be surprised if AI research moves towards training models to use bolt-on peripherals like scratchpad memory. Transformers showed us an interface for addressing, so I wouldn't be surprised if someone figures out a way to use a similar addressing scheme and made it read/write.
Re:HuggingGPT: Solving AI Tasks with ChatGPT and Its Friends... by UniverseHacker on Friday March 31, 2023 @02:45PM (#35391477)
You are right, but it is unclear to me where this limitation will fall in terms of actual abilities.

GPT-4 can divide large tasks up into logical steps and instructions that can be worked on independently by other instances, and it can create "compressed" condensed explanations of it's own state that can be stored externally and repeated back to it, or passed back and forth between instances. With the unreleased 32,000 token context window, that is really a lot of context when you consider that it can heavily compress things by referencing what it was trained on.

Re:HuggingGPT: Solving AI Tasks with ChatGPT and Its Friends... by IanCal on Friday March 31, 2023 @03:18PM (#35392019)
> With the unreleased 32,000 token context window, that is really a lot of context when you consider that it can heavily compress things by referencing what it was trained on.

Also that means you can just iterate with different contexts until you can deal with the problem. How many problems need 50 pages of context on top of what gpt4 knows for solving the next step?

Re:HuggingGPT: Solving AI Tasks with ChatGPT and Its Friends... by Workaccount2 on Friday March 31, 2023 @05:29PM (#35393935)
>How many problems need 50 pages of context on top of what gpt4 knows for solving the next step?

Perhaps the unknown unkowns?

I envision a scenario where a super intelligent AI seemingly runs off the rails and becomes obsessed with overtly complex problems that are totally intractable to humans. Where we'd be like ants trying to make sense of cell phones.

Re:HuggingGPT: Solving AI Tasks with ChatGPT and Its Friends... by jules on Saturday April 01, 2023 @02:25AM (#35397703)
Maybe it can do this with sleep, where it gets further training on the new input and output it produced that day.
Re:HuggingGPT: Solving AI Tasks with ChatGPT and Its Friends... by Sinidir on Saturday April 01, 2023 @03:54AM (#35398204)
That just gave me the idea that there could be a main instance that keeps a summary of the problem (List of paragraphs for each subproblem) and other instances that hold the larger text that these summaries are based on. Then on demand the main instance could get needed extra information for a subpoint back. Kinda like having forgein keys in a databas with which you can join in the relevant row!
Re:HuggingGPT: Solving AI Tasks with ChatGPT and Its Friends... by nomel on Friday March 31, 2023 @03:05PM (#35391802)
I think an attention/focus system might be applicable. I think humans have the same limitations, with tricks to work around it. As Jim Keller once said on an Lex Fridman interview, a good engineer is one that can jump around between the different levels of abstraction. We think in a high level, then "focus" and jump down when needed.

I think something like this could be hacked together, with summarization and parallel "focus threads", containing prompts that focus on details. These could be pruned/merged back together, to add to the "summary" higher level abstractions.

I use this approach already, to some extent, when a conversation gets too long and I need specific details. I'll start a new prompt, include a high level summary, and then "focus" on specific ideas to get more specific answers about a particular detail.

Re:HuggingGPT: Solving AI Tasks with ChatGPT and Its Friends... by sdenton4 on Friday March 31, 2023 @04:06PM (#35392793)
Approaches like RETRO tack vector search over a fixed database onto an attention based model - that's how you get long term memory. People are already working on it for the current crop of LLMs.
Re:HuggingGPT: Solving AI Tasks with ChatGPT and Its Friends... by brotchie on Friday March 31, 2023 @02:20PM (#35391111)
Yep, the tipping point for me was seeing a demo of GPT-4 writing code, then being shown exceptions where the code failed, and successfully fixing the code given that feedback.

Seems that for any task where you can generate an error signal with some information ("ok, what you just tried to do didn't succeed, here's some information about why it didn't succeed"), GPT-4 can generally handle this information to fix the error, and or seek out more information via tools to move forward on a task.

Only thing that's really missing is somebody to crack the back of a memory module (perhaps some trick with embeddings and a vector db) is all it takes for this to be AGI.

Re:HuggingGPT: Solving AI Tasks with ChatGPT and Its Friends... by alchemist1e9 on Friday March 31, 2023 @03:02PM (#35391745)
> Only thing that's really missing is somebody to crack the back of a memory module (perhaps some trick with embeddings and a vector db) is all it takes for this to be AGI.

Here to say I agree with embeddings and vector db for memory systems. However also to disagree this can lead to AGI.

Re:HuggingGPT: Solving AI Tasks with ChatGPT and Its Friends... by basch on Friday March 31, 2023 @02:14PM (#35390992)
It also, in my pay opinion, needs a memory store.

I need to be able to save something to a variable and call back that exact variable. Right now, because it’s just pure text input of the whole conversation, it can forget or corrupt it’s “memory”.

Even something like writing a movie script doesn’t work if it constantly forgets characters names or settings or plot points.

Re:HuggingGPT: Solving AI Tasks with ChatGPT and Its Friends... by UniverseHacker on Friday March 31, 2023 @02:16PM (#35391042)
My description above includes a memory store. It is a single variable that it can save and recover, and will automatically get sent back to it at regular intervals if it forgets it saved anything. Only a single text string variable is needed, as GTP-4 is smart enough to realize it can compress, pickle, etc. as many things as it needs into that.
Re:HuggingGPT: Solving AI Tasks with ChatGPT and Its Friends... by alchemist1e9 on Friday March 31, 2023 @02:43PM (#35391460)
I think we might be working something a bit similar and I agree memory is an issue. However it’s not really AGI that will emerge from this. It’s looking very useful but it’s still observable but it’s not a “mind” like an AGI.

To add to your comments, I suggest that local vector store of embedding vectors of local content is how I’m going with memory issue. Langchain like. That way all previous progress is tracked and retrievable. That way the system can retrieve all it’s “memories” itself. The recursive multi agent pattern is big deal in my opinion.

Re:HuggingGPT: Solving AI Tasks with ChatGPT and Its Friends... by nomel on Friday March 31, 2023 @03:07PM (#35391847)
> but it’s not a “mind” like an AGI.

Please define this. ;)

Re:HuggingGPT: Solving AI Tasks with ChatGPT and Its Friends... by WA on Saturday April 01, 2023 @01:33AM (#35397441)
Please define AGI first ;)
Re:HuggingGPT: Solving AI Tasks with ChatGPT and Its Friends... by nomel on Saturday April 01, 2023 @02:53PM (#35402935)
I asked first?
Re:HuggingGPT: Solving AI Tasks with ChatGPT and Its Friends... by basch on Friday March 31, 2023 @06:30PM (#35394630)
lay*
Re:HuggingGPT: Solving AI Tasks with ChatGPT and Its Friends... by pmoriarty on Friday March 31, 2023 @11:01PM (#35396642)
Langchain[1] can be used to add memory to an LLM.

[1] - https://www.pinecone.io/learn/langchain-conversational-memor...

Re:HuggingGPT: Solving AI Tasks with ChatGPT and Its Friends... by knome on Friday March 31, 2023 @03:18PM (#35392025)
I'm prone to agree. Your memory would still require being parsed by the language model, and for it to have a sufficient window as to allow this. Alternately, the human mind keeps, what, 7 items 'present' at a time? When the 32k variant comes out, the ability to use a side-along memory module on a running self-summarizing conversation will be even easier than it is now. I expect a more human-like memory will emerge eventually, once we can figure out a good way to project and retrieve via arbitrary weight values into it.

I'd like the "stochastic parrot" crew to explain how interactions like this somehow don't involve the extraction of and manipulation of abstract meaning.

  H: translate the following sentence to instead use olympic curling related metaphors ( do not explicitly mention curling, and do not use similes in forming the translation ): man:"I've been spinning my tires all day without getting anywhere" woman:"running out of gas?" man:"nah, I'm gassed up, geared up, oiled and revving. I just can't seem to get in the right lane, you know?" woman:"anything knocking?" man:"nah, I'm tuned just fine. I can't seem to get my GPS to connect, you know?" woman:"take a few laps without a destination, see if you can figure out the lay of the land" man:"you know, that might help. thanks barbara" 
  R: man: "I've been sweeping the ice all day without hitting the target."
  .  woman: "Losing momentum?"
  .  man: "Nah, I've got the energy, the technique, and the focus. I just can't seem to find the right path, you know?"
  .  woman: "Anything coming in your way?"
  .  man: "Nah, I'm on point with my aim. I just can't seem to get my strategy to align, you know?"
  .  woman: "Maybe glide a few times without aiming specifically, see if you can figure out the ice conditions."
  .  man: "You know, that might help. Thanks, Barbara."
It seems obvious to me that it is understanding the metaphors into underlying meanings to be able to translate them like this. I think half the pushback is just ego driven nonsense, similar to the sorts of people that make similar claims on the capabilities of animals.

It's all just an argument that "boats can't swim".

Re:HuggingGPT: Solving AI Tasks with ChatGPT and Its Friends... by UniverseHacker on Friday March 31, 2023 @03:42PM (#35392396)
I think this is a good example of deep conceptual understanding, I suspect if you posed this same problem to adult humans (e.g. me) they wouldn't be able to do as well.
Re:HuggingGPT: Solving AI Tasks with ChatGPT and Its Friends... by michaelmrose on Friday March 31, 2023 @09:21PM (#35396021)
An actual intelligence would have a mental model of the world deeper than the connection between one token and the next and a model of the human beings he was interacting with and an expectation of effecting that model to achieve actual differences in the observed world. In other words Google translate doesn't actually speak Chinese.
Re:HuggingGPT: Solving AI Tasks with ChatGPT and Its Friends... by rlt on Saturday April 01, 2023 @02:17AM (#35397645)
> An actual intelligence would have a mental model of the world deeper than the connection between one token and the next

Why are you so confident in this? We don't really know how human "intelligence" works. Whose to say it's not doing something similar?

It's a bit like saying an audio file doesn't contain "music" because it's just a binary representation of the original work.

Re:HuggingGPT: Solving AI Tasks with ChatGPT and Its Friends... by TedDoesntTalk on Saturday April 01, 2023 @04:06AM (#35398275)
> We don't really know how human "intelligence" works

Or cepholapod intelligence. Or magpie/raven/crow intelligence.

Re:HuggingGPT: Solving AI Tasks with ChatGPT and Its Friends... by aezart on Saturday April 01, 2023 @04:44AM (#35398507)
If I ask a human what their favorite flavor of ice cream is, they will answer based on their actual preference for different kinds of ice cream.

If I ask a chat LLM what its favorite flavor of ice cream is, it will select a statistically likely answer to the question.

That's the difference.

For reference: my favorite ice cream flavor is cake batter, but only if it's the one at Cold Stone. If it's from somewhere else, I go with cookie dough.

Re:HuggingGPT: Solving AI Tasks with ChatGPT and Its Friends... by UniverseHacker on Saturday April 01, 2023 @08:39AM (#35399807)
It seems to me that most of these arguments boil down to “it didn’t think or respond like I would expect a human to, therefore it isn’t intelligent.” This isn’t open minded to the diverse possibilities of non human intelligence, nor is it demonstrating a lack of understanding or real limitation. It doesn’t seem to have any sense of self beyond playing characters it was asked to play, yet those characters are able to intelligently select actions towards goals they were assigned by a prompt using creativity, complex reasoning, and deep understanding. This is a difference that isn’t an inferiority or shortcoming.

I would even argue that humans are more similar to that than we think. The main difference is that one of the characters is locked to the surface most of the time (not even very perfectly).

Re:HuggingGPT: Solving AI Tasks with ChatGPT and Its Friends... by rlt on Sunday April 02, 2023 @01:15AM (#35407406)
GPT’s only sensor input is text, but if it were able to taste maybe it would learn a favorite ice cream flavor too.

Why do you prefer cake batter ice cream? Because of some sequence of events leading up to present point in time (not unlike training), and/or random genetic variations (not unlike setting “temperature” parameter in GPT to >0)

Re:HuggingGPT: Solving AI Tasks with ChatGPT and Its Friends... by simondw on Sunday April 02, 2023 @09:24AM (#35410257)
Someone's actual preference is stored in their memory. If you give the LLM access to a database that includes "preferences", I'm sure it will be able to retrieve that.
Re:HuggingGPT: Solving AI Tasks with ChatGPT and Its Friends... by michaelmrose on Monday April 03, 2023 @01:54AM (#35419042)
If it's preference is a variable that exists because a human thinks it would make sense to another human it has no connection with reality.
Re:HuggingGPT: Solving AI Tasks with ChatGPT and Its Friends... by WA on Saturday April 01, 2023 @01:29AM (#35397422)
I can’t understand how you come to this conclusion so easily. It makes no sense. Computers are better at some things. My calculator can do 12345 x 54321 in a heartbeat, I can’t. What does this prove at all?
Re:HuggingGPT: Solving AI Tasks with ChatGPT and Its Friends... by chillfox on Saturday April 01, 2023 @06:54AM (#35399170)
That calculators will take most of the math jobs and companies wont have math departments anymore...
Re:HuggingGPT: Solving AI Tasks with ChatGPT and Its Friends... by UniverseHacker on Saturday April 01, 2023 @09:39AM (#35400229)
That specific example proves conceptual understanding of the subject matter.
Re:HuggingGPT: Solving AI Tasks with ChatGPT and Its Friends... by espadrine on Saturday April 01, 2023 @05:13AM (#35398638)
> I'd like the "stochastic parrot" crew to explain how interactions like this somehow don't involve the extraction of and manipulation of abstract meaning.

They don’t claim that it doesn’t. What they point out is that humans have a tendency to ascribe intent and agency to the text it outputs. But the LLM is optimized for prediction, not survival, unlike humans[0]:

> Text generated by an LM is not grounded in communicative intent. […] Our perception of natural language [is mediated by] our predisposition to interpret communicative acts as conveying coherent meaning and intent

Some of the dangers they raise associated with this is that it will not realize that words it chooses are PII, dangerous to give to who they are talking to, or biased in a way that can cause societal harm:

> If the LM or word embeddings derived from it are used as components in a text classification system, these biases can lead to allocational and/or reputational harms. […] A Palestinian man [was] arrested by Israeli police, after MT translated his Facebook post which said “good morning” (in Arabic) to “hurt them” (in English).

What they encourage is to view it as a tool and assess how it may fail. For instance, you might object that the MT was simply incorrect; but realistically, sentences can be translated in many ways and with many intents (eg. Allah Akbar has a lot of contexts!) and the LLM may not be given the full picture of the situation.

The stochastic parrots paper is heavily misrepresented or used by people that don’t seem like they read it. For instance, it was cited in a recent petition asking to stop work on powerful AI, prompting a response from the authors[1] pointing out that it misunderstands the paper and emphasizing that they disagree with the petition.

[0]: https://dl.acm.org/doi/pdf/10.1145/3442188.3445922

[1]: https://www.dair-institute.org/blog/letter-statement-March20...

Re:HuggingGPT: Solving AI Tasks with ChatGPT and Its Friends... by knome on Saturday April 01, 2023 @10:18AM (#35400502)
>The stochastic parrots paper is heavily misrepresented

I listened to a lengthy video on the "stochastic parrot" from the Alan Turing Institute yesterday after having made that comment, and it was mostly on the topics you mentioned, subtle and overt bias, etc. I kept waiting on them to explain the "it's not real" bit that I've seen implied by the phrase where I've seen it used ( or perhaps, simply read into it myself ), but it never really came in. At the very end, one commentator made a comment implying it's not really all there and the next responding didn't disagree outright, but you could see a kind of "well yes, but actually no" kind of expression go across their face as they talked around it.

Re:HuggingGPT: Solving AI Tasks with ChatGPT and Its Friends... by UniverseHacker on Saturday April 01, 2023 @09:00AM (#35399990)
I like the boats can’t swim summary of this disagreement. I think these two groups have been talking past each other, and this is the crux of the disagreement. Some people seem to expect AI to converge to being more like a human as it gets better and are stuck on how inhuman it is. They are not considering the endless possible types of non human intelligence. These differences should not be mistaken as inferiority, in many cases they will offer huge advantages.
Re:HuggingGPT: Solving AI Tasks with ChatGPT and Its Friends... by lukeschlather on Saturday April 01, 2023 @03:00PM (#35403004)
I agree in general, though I think you're being too optimistic about GPT4 finding and fixing its mistakes. I haven't been using it much, but the only thing where it performed flawlessly was translating a simple song into Spanish. At the same time, GPT is a total sycophant and this makes it rather hard to work with. Whenever I give it feedback it agrees with whatever I have said and then proceeds to produce paragraphs of text which include several contradictions of the things we've "agreed" on.

I think the "stochastic parrot" idea is too harsh, but it does reflect that GPT is fundamentally incapable of looking at a small block of text, turning it into a set of logical facts, and recognizing if those facts are consistent or not. I think it is something that is possible to build with a refinement of GPT, but I'm not sure it can be done simply with glue code.

Re:HuggingGPT: Solving AI Tasks with ChatGPT and Its Friends... by sharemywin on Saturday April 01, 2023 @10:49PM (#35406662)
At least not the models we have access to
Re:HuggingGPT: Solving AI Tasks with ChatGPT and Its Friends... by ActorNightly on Sunday April 02, 2023 @03:48AM (#35408199)
Honestly the whole fear mongering around AGI is kind of annoying. We are very far away from AGI, and its driving focus away from the main issues that people should be paying attention to.

An AGI, by definition, can start with zero information, and learn on its own. Where are the models for that? The closest thing we have is Mu Zero, but we have to give it an objective function, which destroys the whole idea of AGI in the first place (as it should be able to generate its own objective functions) and it takes a shitload of resources to run for a very limited rule set. Until you have something like this in place, no version of Chat GPT will even come close to AGI. The problems that chat GPTs are solving are turning out to be not as complex as we thought, thats all that the "sparks of AGI" are.

What will happen assuming the current course of iteration on the models is that humanity will get is essentially better and better software that can solve a large range of complex problems. You will have essentially a large body of human knowledge that is compressed in some form, with the ability to interpolate between branches very efficiently. This will create issues because you will now be able to ask multi modal questions like "what do I need to do economically to ensure ultimate compliance of the population?", and the solution will involve things like which government offices to target with what information, generated videos of propaganda, special products that promote addition or some other health changes, e.t.c.

To prepare against this, the solution isn't to lock down the models and hide the key while telling everyone how AGI is going to kill us all, the solution is to allow everyone access to this and let the AIs fight against each other and reach an equilibrium.