firewolf34

ChatGPT would have probably parrotted the bold text. It is always super concerned about risks.

The demo video for it is pretty cool: https://youtu.be/sj2tzW2uhZo

Medieval, more like MIDIeval, am I right?

Because it blew up as a get rich quick scheme / "hustle culture" trend recently as it became much more accessible via internet-based dropshipping facilitators, and there's a lot of people trying to make a buck on it, in sometimes nefarious ways.

I was reading a book (1) which talked about that and emphasized the importance of writing on formulating thoughts and ideas. The funny thing is, it seems it has less to do with writing being some magical input method that makes you think better, and more to do with the fact that writing is just plain slow and forces you to think through and sort of sum up your thoughts as you go. So ironically, it being an inefficient method actually has a positive! But I still feel like you could get most of the way there by just being more deliberate when using a different input method, for example, forcing yourself to stop and think as you type, or using outlining tools, or maybe even artificially limiting your input speed...

(1) "How to take Smart Notes" by Soenke Ahrens

If ya can't sleep, ya might as well learn something. Thanks for the link! Sincerely, ~also can't sleep.

So many different little human trends, crafts, styles... lost to time. Makes you wonder what we're missing.

The whole point of their system isn't to eliminate the possibility entirely it's to make it impractical to get around for the vast majority of concert-goers, and it clearly succeeds at this.

Recording the ticket with a video is everyone's first thought at defeating their restriction, and is no doubt the first thing they thought of when designing it. Hence, the codes expiring too quickly that you'll need a new video before you get through the line at the entrance of the venue. And messing with videos in a pressured line of people in front of a bouncer, is, as others have said, simply not practical for the vast majority of cases.

So it's kind of irrelevant - practically speaking - that it is possible.

If the internet is being filled with AI generative content post-2021, then doesn't that just imply that the next generation of AI training on this "slurry" would be analogous to a "multi-round fixing" operation (as quoted above)?

While currently this is a relatively weak strength of genAI - assuming technological improvement of this technique over time, isn't it just as possible that the data quality will converge positively rather than negatively over time, in the future? That is to say, the web would be consistently "refined" as time goes on, by predominant VLLM?

Assuming that the internet is even "filled" as you say in the first place (personally I don't think organically-generated content is ever going to be pushed out of the internet, but that's my opinion, and I'll entertain the opposite case for the sake of the discussion). It also assumes that people are using models trained on the current state of internet "slurry" in the first place - that we are continually ingesting more internet YoY into these models. If we come up with a better model that needs less data to produce high-quality content, neither my nor your assertion is even relevant. Same case if the internet just decides to use small, low-quality models trained on only a portion of the internet.

But if the internet is continually recycling the entirety of itself through a model that has tens of millions of dollars of funding and research focused on directly improving the quality of it's answer metrics, it's not necessarily 100% locked into a downward quality convergence slide. Especially if we assert that humans /will/ continue to be consistently putting more organic data into the internet over time. It's a pessimistic take.

How are these people finding VGA cables in the street :S I needed like 10 or so VGA cables recently for an art installation and asked everyone I could and nobody had any lying around... I ended up having to buy new ones which seems a shame considering how many are thrown away!

https://mosh.org/

"Schedular" typo in image heading

Now, I'm all for not having to deal with their annoying paywall tactics bullcrap, but I'm wondering, is using archive.org as a way to get around paywalls like WSJ's an abuse? I'm concerned less for their profit and more for the fact that we might get archive.org hit with cease-and-desist's or cause archive.org to have to foot larger bandwidth costs when they're a really good service that's important for the long-term of the internet. I see a lot of people do it, though, so wasn't sure if there's some sort of blind-eye going on.

Wow, that bit at the end really sold it. Very cool

I think we're looking at this wrong. I feel like this test is designed to investigate social biases not test for logical skills and if these people are failing it, it's not so much of a failure in their understanding of logic but rather a procedural impact of the way the question is framed, which is probably precisely why "reframing it in a social context" changes their result populations. I think this test is extremely sensitive to how you pose the question.

Are we trying to test if the candidate can solve the logic problem, or are we trying to test how they handle an /intentionally-confusing/ situation and what (psychological) biases they jump to with their solution?

If it's a test of their logic capabilities, then it seems like the numbers are artificially low, so maybe not so embarrassing as you say... Reason being, I think there are several confounding variables included in the results they'd need to control for if that was the point.

An obvious one, if we were testing logic directly, then I wonder if they allowed the participants submission to "show their work" rather than just which final cards they chose. Doing so would eliminate the "carelessness" confounder in the result where they didn't thoroughly think through all of the logical cases of the cards or where they accidentally included an incorrect card but understood the nature of the required solution, ie. if they knew they needed to disprove rather than confirm but accidentally included a useless card for disproving, they still understood how to solve the problem and thus the logic. What percentage of their results fall into that bucket?

There's also other confounding factors that are set up to "confuse" the participant here that could be removed if we wanted to truly test their /logical skills/ and /not/ some psychological/sociological property. For example, the question merely says: "test that if a card shows an even number". In English, "if" can mean both the inclusive or exclusive OR depending on context - it's needlessly vague, and additionally, I posit that in English, given the common usage of the phrase "test ... if", the phrase is /leading/ the participant to look for /positive confirmation of the rule/ rather than the negative. You can of course derive that the negative test is needed by studying the cards but why try to mislead them outright? Why not say "choose the set of cards that you'd need to flip to prove the rule is false"? This clearly demonstrates the task and doesn't send them on a goose chase.

There's other things too. It doesn't mention if these cards are from a global set of cards or the rule is only meant to be proven on the 4 cards presented. It implies the latter but if you start thinking about "confirming if the rule is true for all cards", it sends you down another useless logical rabbithole, yet, /cards normally come from a deck in real life/ and it is natural to expect there are more cards. Maybe if they wanted to be exact we shouldn't be using cards at all but rather wooden blocks or something.

And I'm sure there are more "biases" that I'm not catching here. If your goal is to test people's likelihood of affected by certain biases psychologically, then all's well and good with the test, go right ahead. But if you're going to present the poor results as some sort of indicator of an population's skill at logic, maybe not the best test without some better testing procedures, imo.

Seriously, it's the same argument that people give for "ChatGPT can't give me good code, I don't know why", just rephrased. The deluge of "GPT is not useful for X" articles meant to bait the average critic, despite it being used en masse for "subproblems in X-space" already...

They're asking the wrong type of work from it. If you need some boilerplate or a transformation, it's going to give you a fantastic template to work with. If you need it, on the other hand, to engineer out a highly-specific and nuanced solution with an esoteric codebase to a complex problem, maybe not so much. The former is wide, the latter is narrow. It's going to take maybe a bit more breakdown of the scope into proper subproblems before you'll get a good answer; and that's something you can do yourself, or have an agent perform across multiple queries maybe (though I'll admit, more work needs to be done for the whole multi-agent workflows to be truly useful).

Okay, now you absolutely must explain how you are employing high temperature superconductors at a party. :) I want to go to some of your parties! I have heard of the ice cream. I must try that perhaps some day...

Thanks for the recommendation! :)

This is very strange. I've been working with an AI assistant called Astra since GPT-3 came out. I told it to name itself, and the name it chose was Astra. Now this?

What do you use to "keep /etc in scm"?

That's the best interaction I've seen - so many are just GPT piped into a TTS, but this seems to actually be identifying different speakers in the crowd? There's a point where a guy butts in to ask a question and the robot basically says "wait a sec, I'm talking to this person, are you done? Okay, now I'll answer your question, raise your hand." That's another layer of interaction that's impressive and the latency and speech recognition quality is near-realtime. Very cool

What do you do for party entertainment with it? Fog machine? Smoky cocktails?

Something nobody seems to mention that could be a convoluting factor...

I wonder if this operates bidirectionally - if parents learn traits from their offspring, maybe to a lesser extent. If you have a group of people that constantly spend time around each other, they all seem to average out their social tendencies over time, to some degree. As children get older and develop more defined characteristics, it's possible this isn't a one-way street.

The NUMBER 1 most annoying thing about not having a phone in the modern day is ticketing services / events. The second is Uber/Lyft/etc (though less of an issue in cities with public transit).

Is not the purpose of a model to interpolate between two points? This is the underlying basis of "hallucinations" (when that works out /not/ in our favour) or "prediction" (when it does). So it's a matter of semantics and a bit of overuse of the term "hallucination". But the model would be useless as nothing more than a search engine if it were to just regurgitate it's training data verbatim.

Storing photos of people is a bit of a can of worms, make it even easier and just ask for ID if this error occurs.

So perhaps you're suggesting we sort of "boil-down" an information source into a sort of base representation of meaning and intent, something similar to vector store, and relate the many inputs together in this space using the LLM as glue like one does manually creating links in a web of Zettelkasten for research. I think this is something that the field is rapidly moving towards in personal information management.

My theory is that to meet a new person and have it be substantial, you basically need to spend a few hours, a few times per week, in the same space doing some arbitrarily interesting thing for a common reason, without being too eager but with a signaled sense of openness.

I like this concept, and I feel like I've experienced this as well, but I'm having trouble picturing an example of what you're describing, practically speaking, for the average city-dweller. Care to elaborate on this?

This is an interesting idea but I'm having trouble understanding what you're to achieve. Do you mean the LLM would simply continuously update it's context window with incoming data feeds realtime, and you use it as an interface? That's pretty akin to summarization task, yes? Or are you augmenting the streams with "metadata" you mentioned?

Check out PrivateGPT on GitHub. Pretty much just works put of the box. I got Mistral7B running on a GTX 970 in about 30 minutes flat first try. Yep, that's the triple-digit GTX 970.