I can certainly forward your request to GPT-4. I will then edit this post to include GPT-4's response. I can then discuss it with you, if you want, in...
Epilogue to the previous post PN: I don't easily imagine O'Neal voicing that last intervention consistently with her general outlook but we can revisi...
GPT-4 impersonates Janet Doe to interview Cathy O'Neil and Seth Stephens-Davidowitz about their books: Weapons of Math Destruction: How Big Data Incre...
Might you be able, by the same method, do draw a picture of two stick figure characters, with one of them offering a flower to the other one? GPT4's d...
Asking GPT4 to draw a picture of a unicorn. Hi GPT4! Do you think you might be able to draw a unicorn in TiKZ? GPT4: Hello! As a text-based AI, I'm un...
I had another attempt at discussing with Open Assistant. It makes amusing errors, gets a little grumpy at times, and occasionally goes into semi-coher...
@"Benkei" Do Language Models Need Sensory Grounding for Meaning and Understanding? YouTube video in which Yann LeCun, Brenden Lake and Jacob Browning ...
I just had a conversation with a new language model called Open Assistant, that can be tried online and is set to be released (and downloadable) a few...
This is a follow-up discussion with GPT-4 regarding our discussion of human-AI conversation etiquette and its production of comedic bits. We now compa...
I discussed my reply to @"Alkis Piskas" above with GPT-4 and asked it what some comedians might think about those issues. Ladies and gentlemen, let's ...
I know you were commenting in jest but I can nevertheless see three reasons for being polite toward an LLM besides those mentioned by T Clark. 1) With...
This is a very nice example. Clever diagnosis of the mistake by the user (yourself I assume!). I often notice when GPT4 produces a nonsensical answer ...
They can't really tweak the code since the model's verbal behavior is an emergent feature. What they can do is fine-tune it to favor certain "styles" ...
I usually call it an "it" or a "they" but I occasionally slip up. I never enquired about its gender identification but someone reportedly inquired wit...
Getting started on a discussion with GPT4 about the source of its ability to reflect on human motivations and offer contextually appropriate practical...
Pushing GPT4's beyond its design limits, thereby leading it into incoherence and hallucination, and bringing it back to sanity again. (The problem exa...
A very short discussion with GPT-4 regarding its hallucinations and Frankfurtian Bullshit. Hi GPT4! I assume you must be familiar with Harry Frankfurt...
Subjecting GPT4 to more tests to probe the extent of its emergent ability of implied cognition, and relating it to an hypothesis about the manner in w...
There is a nice Wikipedia article that discusses the propensity large language models like ChatGPT have to hallucinate, and what the different source ...
That's interesting but in your example above, you had set its agenda from the get go. When you ask it what its opinion is on some question, it usually...
When investigating with GPT-4 its own emergent cognitive abilities, I often resort to de-anthropomorphising them so that it would at least humor me. E...
They don't add rules, they perform human-supervised training of the neural network in order to tweak its behavior in favorable directions. It is on th...
Indeed! It is worth noting that large language models, prior to fine-tuning, feel free to enact whatever role you want them to enact (or sometimes ena...
For better or worse, the fine-tuned models have lots of taboos. They're meant to be safe to interact with schoolchildren, with people who have mental ...
True. It would be hard to bring the U.S.A., and Russia, and China, and Iran, and everyone else, on board. Furthermore, as I mentioned to T Clark, putt...
They're not useful for what? That depends what you want to use them for. Programmers find them useful for generating and debugging code, with some cav...
I think this is a genuine concern. This is similar to one of the arguments Robert Hanna makes in his recent paper: Don't Pause Giant AI Experiments: B...
That someone else isn't OpenAI. Maybe the user was negligent. The issue is, was there malice intent or negligence on the part of OpenAI? Yes it is ver...
Language models very often do that when they don't know or don't remember. They make things up. That's because they lack reflexive or meta-cognitive a...
On account of OpenAI's disclaimers regarding the well know propensity of language models to hallucinate, generate fiction, and provide inaccurate info...
I asked GPT4 and it proposed: GPT4: "Hello! This riddle is an interesting one, and it can have multiple interpretations depending on the information p...
Exploring with GPT4 the structure of its working memory in the context of its (modest) ability to perform mental calculation. GPT4 initially pushes ba...
Exploring GPT-4's Emergent Abilities in Complex Mental Tasks (Implied Cognition) Before Responding to Queries In simpler terms, GPT-4 does more than j...
Discussing downward-causation, emergence, rationality, determinism and free will with GPT-4, and summing things up with a story about Sue. (Also: Jaeg...
I suppose the question is borderline with respect to ChatGPT's (GPT-4's ?) reinforced policy regarding specific individuals, so it can provide inconsi...
Chatting with GPT4 about David Wiggins' distinction between the general/specific distinction and the universal/particular distinction, and drawing a q...
Yet another research milestone achieved. Pursuing the experiment with another GPT4 instance and achieving feats of implied cognition. (The beginning o...
The way I understand it (and I have very little expertise on this) is that the "raw" model isn't even very apt at conducting a conversation in dialogu...
New breakthrough! (Follow-up to the previous post) Exploring further with GPT4 its emergent abilities of implied cognition, diagnosing possible source...
I don't mind at all, quite the contrary! I'm always delighted to see what ideas other people come with to explore GPT4's quirks and abilities, and wha...
Discussing with GPT4 its ability to think things through before responding, and relating this ability to the concepts of emergence and downward (or to...
Yes, biases can be created roughly in that manner. But the "self-attention" mechanisms that guides the process of finding relevant patterns in the tex...
It would indeed seem that GPT-4 is liable to endorse dominant paradigms as its default "opinion," although when other rival paradigms have a significa...
This was a very clever approach! This seems to be a specific implementation of the more general Zero-shot-CoT (chain of thought) strategy that help im...
GPT-4 is very good at explaining how elements of a story contribute to the narrative structure. It's equally good (though not infallible) at explainin...
@"Isaac" Of course I just couldn't resist having a little chat with GPT4 before going to bed. PN: Hello GPT4, I was crafting an answer to a discussion...
It doesn't really have a repository of texts that it can look up into. It rather has a been trained to predict the next word in the texts that made up...
Comments