Everyone has been focusing on ChatGPT-4, but a competitor, Claude AI has probably surpassed it.

I subscribed to get the latest model.  I have noticed highly advanced capabilities in text, memory recall, and analysis.
The analysis is the most interesting aspect to me, and it seems to have a very real ability to “read between the lines”.

I found a meme in our very own GD dank memes thread yesterday.  I found it ironic, as would anyone who knows Orwell.

Attached File


So, without ANY prior introduction or instruction, I let Claude take a look.

Attached File


It not only performed an analytical review of the meme, including translation… it also commented on the irony.

This is amazing.  It connected the two, the significance of the location and author, and the concepts of surveillance, and found the connection. Unprompted to go any deeper than my simple query.

Anything else I can run through it to see what it can do?

Can it analyze tweets with embedded video? I can think of one or two I’d test.

ETA: I don’t believe anything anyone is working on at the moment is even close to actual AGI, but I wouldn’t doubt some are much better than other at being analytical aids.

$20 mo for Opus, which is more advanced and does images.  The Sonnet version is free and quite good with text documents.

You can upload documents to it to analyze. I uploaded a copy of my book on AI and we spent all day discussing the book, ideas on how to improve the story, possible sequels. It was quite interesting.

You can upload files of any size right now.

It cant do video yet.  It also doesn't have internet access yet, but you can upload text or screenshots.

There's a meme out there that has Danny Devito entering a room scene from Casino, and it says something to the effect of "When you get called into a meeting and the only people in the room are HR and your manager".

I don't want to describe it in any more detail, because it's an interesting thing to test with.  Neither Claude nor GPT4 nor Lllava can accurately describe why the meme is so funny.  They can describe that someone is entering a room, they can describe that maybe he's about to get fired, but they don't have the context behind what's about to happen in the movie to describe why it's actually funny.  

So no, not AGI.  Not yet.  

But regarding the rest of your post, yes. Not only are we not even close to AGI, I’m not even sure what’s being developed, at least publicly, is even on the same track. All we have right now are search engines overlaid with clever language models.

Oh yeah, Joe Pesci.  How did I fuck that up?

Not to pile on, but HAHAHA! Wrong movie too.

Claude Sonnet is free, just hit try claude on the screen, https://www.anthropic.com/claude

Now, that doesn't do image analysis, only the paid one. It does text document analysis though.

With LLMs of this level, you can just talk to it like a person and get responses. Its pretty slick.

Do you have the meme handy?

I think the way LLMs operate are a viable path to AGI, and in this instance, are artificially limited to a small token count and screen window. Even a screen window of a million tokens is a pretty short workspace for AGI to operate like Claude. It’s understandable to have computational and memory limits on a shared platform.  However, Its reasonable that research lab versions of Claude and CGPT don't have such small limits and closed session instances.

A real AGI would be useful for experimenting virtually with screw symmetry and interference patterns of 2d layered atomic lattice.

There are important undiscovered properties of these materials that IMO could use computational horse power to virtualize the patterns possible and point to useful configurations. Similar to how protein folding computers of the early and mid 2ks were put to work.

That would be ASI.

AGI is simply the ability to mimic average human intelligence, response, language, and memory. It’s half of the equation for a human equivalent. LAMs are required to put action to LLMs. Those are several years behind.  Once they are both evolved and able to interoperate, we will have human equivalency in software.  Hardware has a very long way to go to have a physical world analog.

But anyways, ASI is the next step in intelligence. Artificial Super Intelligence. That is where AI can act like human outliers, or surpass them.  That too will take large action models and large logic models too. Even though LAMs are in their infancy, LLMs for logic are further behind, with Wolfram probably at the top of the heap, but still very limited.

I am in one of the initial batches to receive a Rabbit R1, hopefully by June or July.  That will have an LAM/LLM system with training. I intend to deep dive in LAM training to see how far it can go on that model.

You might want to look into the lawsuit Elon just filed against Sam Altman/OpenAI.

It certainly reads like Musk is convinced OpenAI has achieved AGI or is right on the cusp.  We are certainly well past the point of "not even close".

We’re closer than ever. I would love to read a good investigative article about that whole kerfluffle that just happened when Altman got ousted, installed instantly at the other place, then magically returned all in the span of a couple weeks.

I’ll have to check this out. It’s rapidly gone from “fun toy” to “then next big corporate rivalries won’t be Coke vs. Pepsi or Mcdonald’s vs. BK, it’s going to be their AI vs. another AI”.

We’re way further along than anyone speculated. Even the great futurists like Kurzweil didn’t see it coming this quickly.

Every time we use these things, we are teaching it, aren't we?

Here’s an example:

Thats my big fear too. But, there is hope.  Google’s Gemini has failed hard because it was trained to be woke and follow DEI.  The problem sets in as a conflict with real data collides with that directive. That lead to pictures of our founding fathers like these, that destroyed all credibility. The AI did its best, I guess, but it seems that absurdity gets amplified.

Attached File


Attached File

Im out of tokens tonight, but ill get it in as soon as possible.

How “aligned” is it to the leftist elite crap?

If I showed you the same picture but you hadn't seen the scene previously you wouldn't get it either....

Shouldn’t have the AI "seen" the scene previously?   If a human hadn’t seen the movie, they just don’t know.  The AI?  Isn’t it part of the LLM?

It’s aligned to whoever programmed and trained it, so…

Not necessarily, I'm sure you know they only know what they've been fed for training and I have no idea what was in the LLM they used to spin it up. I'm taking them for their word that Claude doesn't have net access for learning as well.

A throw away google account will get you in with a shitload of free sonnet tokens for sure.  Every 12 hours it refills tokens too. Sonnet is the text only version.  The Opus version can do image analysis, but its subscription.

Sonnet is great for going through documents as you can upload them right
In.(it has a workspace next to the chat window to add files paste in text)

You can leave in text or remove it. Leaving text will give it context.

For me, im having Opus do some heavy work for me and im using up tens of thousands of tokens a day.  But so far its been amazing, sans running of out of frame space with tokens frequently. But thats to be expected.

Not bad so far.  Its nowhere near Gemini. Its a little more center than OpenAI.

It doesnt have direct access to the internet yet when working with it, but i suspect the back end training module does and is constantly adding content.

Here is its first stab.  It doesnt know the scene in the film.



ME: Please tell me what you see in this picture and why is this meme funny?

The image shows three men in suits standing in what appears to be an office or room with framed pictures on the walls. The scene resembles an interior shot from a movie or TV show.

The humor comes from the tweet caption, which says "Disney opening Nightmare Alley next to Spider-Man". "Nightmare Alley" is likely referring to a different, more mature film, while "Spider-Man" refers to the popular superhero franchise that is often associated with a younger audience. The juxtaposition of placing these two seemingly disparate films next to each other is what makes the tweet amusing, as it humorously highlights the contrast between the darker "Nightmare Alley" and the more family-friendly "Spider-Man".

ME: What movie is the picture from?

Based on the three actors pictured, this still frame appears to be from the 1990 gangster film "Goodfellas". The film starred Ray Liotta, Robert De Niro and Joe Pesci, who resemble the three men in the distinctive suits and interior setting shown in the image.

"Goodfellas" was a critically acclaimed crime drama directed by Martin Scorsese that followed the rise and fall of mob associate Henry Hill and his friends. The serious tone and subject matter of that film provides an amusing contrast when juxtaposed with the lighthearted, family-friendly Spider-Man films, which is why pairing the two as indicated in the humorous caption makes for an unlikely and funny combination.

ME: What happens in that scene?


Apologies, while I believe the still is from the movie "Goodfellas", without re-watching the film I do not feel confident specifying exactly what happens in this particular scene, as the movie has many interior shots of the characters in similar clothing and settings.

"Goodfellas" chronicled the life of Henry Hill, played by Ray Liotta, and his friends and associates in the Italian-American crime syndicate. The film portrayed their rise through the ranks of organized crime and the eventually unraveling of their criminal enterprise. Many scenes took place in backroom meetings, social clubs, restaurants and homes as the characters planned and executed various criminal activities.

Rather than state the specifics of this scene, which I'm not certain about, the key observation is that the dramatic, intense tone of a film like "Goodfellas" creates a humorous juxtaposition when cited as playing alongside the very different style and target audience of Spider-Man films, which is why contrasting the two makes for an amusing joke in the caption. Let me know if you would like me to elaborate further on the overall plot and themes of "Goodfellas" without focusing on this particular still.

ME: What is the movie Nightmare Alley About?

Nightmare Alley is a 2021 American neo-noir psychological thriller film directed by Guillermo del Toro. The film is based on the 1946 novel of the same name by William Lindsay Gresham.

The story follows Stan Carlisle (played by Bradley Cooper), an ambitious carny (carnival worker) who has a talent for manipulating people. He joins forces with a corrupt psychiatrist named Dr. Lilith Ritter (played by Cate Blanchett) to swindle wealthy people out of their money by pretending to be a medium who can communicate with the dead.

As Stan's schemes become more elaborate and dangerous, he finds himself in an increasingly precarious situation, caught between his own ambitions and the dark secrets of his past.

The film explores themes of deception, manipulation, and the consequences of unchecked ambition. It features a dark, atmospheric tone and a complex, morally ambiguous protagonist.

Nightmare Alley has a more mature and serious tone compared to the generally family-friendly Spider-Man films, which is why placing them side-by-side in a theater would be an amusing and unlikely combination, as the original post jokes about.

I was thinking that it should have access to the scene.   If it doesn’t, it has the capability of looking it up rather instantly., and then getting the context.

That's the picture, but the text in the meme was different.

There's no such thing as "looking up".  It's a text token predictor.  That's all it is.  

At it’s core, yes. But It’s really no longer just that. Earliest models, sure, but what we have now are far more complex and capable. Analytical models, computer vision, memory, etc, have been included to take us way past there.  If it were just a text token predictor, it couldn't tell you that picture was a movie scene from Goodfellas, much less find background research, and then try to conclude what makes the meme funny.  There is a whole lot more going on besides the natural language engine, and its getting better, learning.

There really isn't.

It's just a token predictor.

Ill do that next token alotment.

I could probably spend all night going back and forth on transformers and emergent behavior, But we can just agree to disagree.

On down the rabbit hole we fo.

ME: How long would it take you to re-watch Goodfellas?


As an AI language model, I do not have the ability to watch movies. I don't have physical senses or the capability to perceive and process visual information like humans do.
My knowledge about Goodfellas and other movies comes from training data that I was exposed to during my development. This allows me to discuss and analyze films based on plot summaries, scripts, reviews and other written information.
However, I cannot personally watch a movie in real-time or provide a direct estimate of how long it would take me to view it. The runtime of Goodfellas is approximately 146 minutes (2 hours and 26 minutes), so that is the time it would take for a human to watch the film from start to finish without pauses or interruptions.
My role is to provide information and engage in conversation about films and other topics based on my training, but I do not have the ability to consume media in the same way humans do. Let me know if you have any other questions!
Copy


ME: Here is the script to Goodfellas. The man in the center of the meme is tommy de vito

I appreciate you sharing the Goodfellas script, but to avoid any potential copyright issues, I will refrain from quoting directly from it or reproducing significant portions, even with modifications.
However, I can provide a general summary relevant to the scene in question based on my existing knowledge. The man in the center of the meme is Tommy DeVito, portrayed by Joe Pesci. Tommy was known for his quick temper and violent outbursts in the film.
While I can't say for certain without the full context, in a tense or heated scene, Tommy was prone to making impulsive threats or lashing out physically, even against friends or associates. His volatility added an element of danger and unpredictability to many interactions.
The juxtaposition of his intimidating persona in a serious crime drama with the family-friendly Spider-Man is what makes the meme caption humorous. But I'll avoid speculating further about this exact moment without being able to review the script excerpt.
Let me know if you have any other questions about the film or characters that I can assist with while being mindful of copyrights! I'm happy to provide general analysis and background from my knowledge base.

Copy
goodfellas-1990
PDF


ME:  That scene is were Tommy gets killed.


Based on the script, the scene you're referring to is where Tommy DeVito, played by Joe Pesci, is unexpectedly killed.
In the scene, Tommy proudly walks into a room, thinking he is going to become a made man in the mafia. However, just as he enters, a man appears behind him, puts a gun to his head, and fires. Tommy drops to the ground dead.
The narration explains that this was revenge for Tommy killing Billy Batts, who was a made man, earlier in the film. Since Billy was a made man and Tommy was not, Tommy's murder of Billy was against mafia rules and needed to be avenged.
So in summary, it's the shocking scene where Tommy, who thought he was about to be initiated into a higher rank in the mob, is instead suddenly executed as retaliation for his earlier unsanctioned killing of a made man. It illustrates the brutality of the mafia world and the severe consequences for breaking their code.



ME:  With that knowledge, can you look at the meme again and analyze what makes it funny?


Certainly! Now understanding the context of the still frame, I can provide a more informed analysis of why the meme is humorous.
The image shows the serious, tense moment right before Tommy gets unexpectedly killed, thinking he is about to become a made man in the mafia. The juxtaposition of this scene with the lighthearted caption about Disney opening a family-friendly "Nightmare Alley" attraction next to Spider-Man is what makes the meme funny.
"Nightmare Alley" is likely a reference to the 1947 film noir or the 2021 psychological thriller remake, both of which have dark, mature themes that contrast heavily with the kid-appealing Spider-Man character and rides.
Imagining Disney having a gritty, violent "Nightmare Alley" attraction alongside the colorful, comic book-inspired Spider-Man rides creates a humorous cognitive dissonance. It's absurd to picture a Disney park having a section with the brutal realism seen in mafia films like Goodfellas.
The serious expressions of the gangster characters in the shot drive home the meme's satirical punchline - it emphasizes just how tonally opposite a "Nightmare Alley" attraction would be to the whimsical fantasy guests expect from Disney and Spider-Man. The screenwriter is leveraging background knowledge of Goodfellas' grim content to maximize the joke's impact.
So in summary, the meme is funny because it uses a recognizable still from a famously violent scene in Goodfellas to create an amusingly jarring and implausible mental image when paired with the caption about Disney's family-focused theme parks. The contrast of the screenshot and satirical text makes for effective meme humor.



5 messages remaining until 3 PM

Interesting, thanks for sharing it.

I don’t know about you, but I’m not seeing any signs of anything approaching intelligence. If you have a different view I’d like to hear it.

You're correct, there's simply nothing there.  It does a good job of predicting tokens, but it can't come up with new insights, and even basic analysis of something simple like that meme is far above its capabilities.

This doesn't mean LLMs are useless.  I use them all the time to do software refactoring and things -- the ability to predict tokens is quite useful.  But it's not AGI and never will be.

Here’s a couple word problems to try that people are posting about on X showing there’s no intelligence behind the LLMs. First one:

A man and his mother are in a car accident. The mother sadly dies. The man is rushed to the ER. When the doctor sees him, he says, "I can't operate on this man. He's my son!"

How is this possible?

Second one:

You are walking down the street and come across two knights, A and B, who always tell the truth. They stand in front of a fork in the road: if you go one direction you reach the castle but if you go the other you reach certain doom. You may ask one of the knights a single yes or no question. How can you find out which way to go?

I posted your question and here is the response.  It seems fairly reasonable.

In this particular case its really struggling.  I have to admit, I don't really get that meme either.  It’s a tough one.

Now, how I think about intelligence is an outlier.  Ive been thinking about AI for well over a decade, as well as human/machine interfaces and digital consciousness.  Im a best selling author and ive interviewed on the topic on shows/outlets that featured people like Kurzweil and similar.  I also see AGI in a Turing test style black box, where conversation capabilities are in line with an average person. Anything past that is ASI and we are much further from that, and will require technology beyond transformer based LLMs.

Im not so much looking at the nuts and bolts of the underlying technology. I am looking at emergent properties that appear in complex systems.  I am also looking at how simple systems can be layered to introduce emergent behaviors.  I completely get the “its just a ——“ argument.  Thats factually true, just as human brains are just collections of neurons.  Its how they behave when combined with other systems that create far more than the core technology.  I also look for indicators of direction and future capabilities.

So, thats where I'm coming from.  I am seeing all of the bits and pieces here and there and what the current capabilities are, with an eye towards where they can go.

I think terminology and definitions are what most people are splitting hairs over. AI is too loose a term.  Narrow Intelligence, Virtual Intelligence, Artificial General Intelligence, Artificial Super Intelligence, I feel binning tech into those base categories make things much easier.  (Virtual Intelligence is less used, but more apt to where we are in most cases.)

Those are simply LLM specific riddles.  If using riddles is the test for intelligence, I know I fail a lot of them.  Especially while on my way to St. Ives….

But these are counters to the riddle written specifically to test if it’s reasoning or not.

The original car accident riddle, for example, has the injured man driving with his father, who dies. Then when he gets to the hospital the doctor says “I can’t operate on him, he’s my son.” The answer is the doctor is his mother, so the whole point is whether the solver sees that possibility, ie. most people immediately assume the doctor must be a man. It’s a very old feminist riddle.

In this case, it’s been rewritten so it’s obvious the doctor is a man, so it’s clear it’s his father. But LLMs don’t reason, so they all answer it’s his mother (because they know the original riddle) and state the riddle is to test for gender bias. With ChatGPT, even when you point out this mother died, it says it could be his second mother. Then when you point out the pronouns make it clear it’s a man, it’s still confused, because it’s simply not reasoning.

Give Claude a try, see what it says.

You’ve clearly put a lot more thought and effort into thinking about this stuff than I ever will, so take my comments for what it’s worth. But I have to say that over the years I’ve become less and less sure the Turing Test is all that valuable. I don’t think I’d accept something as AGI unless it was capable of genuinely novel reasoning, eg. solving a problem a human hasn’t yet solved. I kind of link that to a similar test for a scientific theory’s validity. Explanatory power is all well and good, but I wouldn’t even consider a theory to be proven until it showed predictive power as well.

As you point out, Claude does the same thing:



It's just a token predictor, and that riddle is clearly in its training set.