1,711 karma · joined February 6, 2018
meet.hn/city/us-Conshohocken
Interests: Data Science, Art, AR/VR, AI/ML, Fintech, Gaming, Hacking, Hardware, Philosophy, Programming, Science, Space Tech, Web Development
---
And then you can be still be wrong because you didn't appreciate that the company is operating a low-trust environment/switching costs are high/they didn't have enough industry contacts/competitors outmarketed.
Yea its hard figure out the truth when you just dismiss all the evidence that goes against your priors.
We already algorithmically filter resumes, using far dumber 'AI'. Now sure, that's not firing the CEO level... but trying to fire the CEO when you're an AI is a stupid strategy to begin with. But consider a misaligned AI used for screening candidate resumes, with some detailed prompt aligning it to business objectives. If it's prior is that AI is good for companies/business, do you think that it might sometimes filter out candidates it predicts will increase supervision over it/other AI's in the company? If the person being screened has a googleable presence, where even if the resume doesn't contain the AI security stuff, but maybe just author/contributor credits on some paper? Or if its explicitly in the person's resume...
Also every time I read any of these blog-posts, or papers about this, I'm kinda laughing, because these are all going to be in the training data going forward.
Is more value produced or are costs just shifted off the balance sheet onto the public commons? Driving instead of walking/public transit has certainly been profitable for some people/companies. But it has also been less than ideal from a public health standpoint. And the time spent commuting is unpaid, so while the business saves money on rent, the increase in travel time is still a cost borne by society as a whole. I would describe this as the opposite of 'efficient land use' personally.
Would you be interested in buying my rock that wards off tigers? Ever since the great Tiger apocalypse where the premier military power in the world went on decades long quest to kill all of the tigers and everyone learned of the dangers of tigers, my rock has been extremely effective at warding off tiger attacks. For just 12 billion dollars a year (adj for inflation annually) (https://www.tsa.gov/news/press/testimony/2024/04/16/fiscal-y...), you could have this extremely effective rock.
edit: I'm pretty sure I clicked reply on GP's comment. Weird.
> I suspect the path to general intelligence is not that, but we'll see.
I think there's three things that a 'true' general intelligence has which is missing from basic-type-LLMs as we have now.
1. knowing what you know. <basic-LLMs are here>
2. knowing what you don't know but can figure out via tools/exploration. <this is tool use/function calling>
3. knowing what can't be known. <this is knowing that halting problem exists and being able to recognize it in novel situations>
(1) From an LLM's perspective, once trained on corpus of text, it knows 'everything'. It knows about the concept of not knowing something (from having see text about it), (in so far as an LLM knows anything), but it doesn't actually have a growable map of knowledge that it knows has uncharted edges.
This is where (2) comes in, and this is what tool use/function calling tries to solve atm, but the way function calling works atm, doesn't give the LLM knowledge the right way. I know that I don't know what 3,943,034 / 234,893 is. But I know I have a 'function call' of knowing the algorithm for doing long divison on paper. And I think there's another subtle point here: my knowledge in (1) includes the training data generated from running the intermediate steps of the long-division algorithm. This is the knowledge that later generalizes to being able to use a calculator (and this is also why we don't just give kids calculators in elementary school). But this is also why a kid that knows how to do long division on paper, doesn't seperately need to learn when/how to use a calculator, besides the very basics. Using a calculator to do that math feels like 1 step, but actually it does still have all of initial mechanical steps of setting up the problem on paper. You have to type in each digit individually, etc.
(3) I'm less sure of this point now that I've written out point (1) and (2), but that's kinda exactly the thing I'm trying to get at. Its being able to recognize when you need more practice of (1) or more 'energy/capital' for doing (2).
Consider a burger resturant. If you properly populated the context of a ChatGPT-scale model the data for a burger resturant from 1950, and gave it the kinda 'function calling' we're plugging into LLMs now, it could manage it. It could keep track of inventory, it could keep tabs on the employee-subprocesses, knowing when to hire, fire, get new suppliers, all via function calling. But it would never try to become McDonalds, because it would have no model of the the internals of those function-calls, and it would have no ability to investigate or modify the behaviour of those function calls.
I think there's three things that a 'true' general intelligence has which is missing from basic-type-LLMs as we have now.
1. knowing what you know. <basic-LLMs are here>
2. knowing what you don't know but can figure out via tools/exploration. <this is tool use/function calling>
3. knowing what can't be known. <this is knowing that halting problem exists and being able to recognize it in novel situations>
(1) From an LLM's perspective, once trained on corpus of text, it knows 'everything'. It knows about the concept of not knowing something (from having see text about it), (in so far as an LLM knows anything), but it doesn't actually have a growable map of knowledge that it knows has uncharted edges.
This is where (2) comes in, and this is what tool use/function calling tries to solve atm, but the way function calling works atm, doesn't give the LLM knowledge the right way. I know that I don't know what 3,943,034 / 234,893 is. But I know I have a 'function call' of knowing the algorithm for doing long divison on paper. And I think there's another subtle point here: my knowledge in (1) includes the training data generated from running the intermediate steps of the long-division algorithm. This is the knowledge that later generalizes to being able to use a calculator (and this is also why we don't just give kids calculators in elementary school). But this is also why a kid that knows how to do long division on paper, doesn't seperately need to learn when/how to use a calculator, besides the very basics. Using a calculator to do that math feels like 1 step, but actually it does still have all of initial mechanical steps of setting up the problem on paper. You have to type in each digit individually, etc.
(3) I'm less sure of this point now that I've written out point (1) and (2), but that's kinda exactly the thing I'm trying to get at. Its being able to recognize when you need more practice of (1) or more 'energy/capital' for doing (2).
Consider a burger resturant. If you properly populated the context of a ChatGPT-scale model the data for a burger resturant from 1950, and gave it the kinda 'function calling' we're plugging into LLMs now, it could manage it. It could keep track of inventory, it could keep tabs on the employee-subprocesses, knowing when to hire, fire, get new suppliers, all via function calling. But it would never try to become McDonalds, because it would have no model of the the internals of those function-calls, and it would have no ability to investigate or modify the behaviour of those function calls.
Not exactly.
> The concentration of carbon dioxide (CO 2) in the atmosphere reach 427 ppm (0.04%) in 2024.
Any process that tries to unmix something is not going to be 'literally' free. And given the relative trace amounts we're talking here...
Is this a real thing that happens? Can you name some historical examples? What is stopping from destroyed businesses from being rebuilt pretty quickly? Uber subsidized taxi rides for millions of people for years, but if Uber went out of business, legacy taxis would come right back. And even if they didn't for X amount of time, X would have to be > the amount of subsidy that was received in the meantime. And if the legacy business was so vulnerable that people were willing to try and replace it with some big matrixes, the replacement company might still escape the local maximum that the previous company was trapped in.
> I have a hunch that, through sampling many AI "opinions," you can arrive at something like the wisdom of the crowd, but again, it's hard to validate.
That's what an AI model already is.
Let's say you had 10 temperature sensors on a mountain and you logged their data at time T.
If you take the average of those 10 readings, you get a 'wisdom of the crowds' from the temperature sensors, which you can model as an avg + std of your 10 real measurements.
You can then sample 10 new points from the normal distribution defined by that avg + std. Cool for generating new similar data, but it doesn't really tell you anything you didn't already know.
Trying to get 'wisdom of crowds' through repeated querying of the AI model is equivalent to sampling 10 new points at random from your distribution. You'll get values that are like your original distribution of true values (w/ some outliers) but there's probably a better way to get at what you're looking to extract from the model.
0. Generate an embedding of some text, so that you have a known good embedding, this will be your target.
1. Generate an array of random tokens the length of the response you want.
2. Compute the embedding of this response.
3. Pick a random sub-section of the response and randomize the tokens in it again.
4. Compute the embedding of your new response.
5. If the embeddings are closer together, keep your random changes, otherwise discard them, go back to step 2.
6. Repeat this process until going back to step 2 stops improving your score. Also you'll probably want to shrink the size of the sub-section you're randomizing the closer your computed embedding is to your target embedding. Also you might be able to be cleverer by doing some kind of masking strategy? Like let's say the first half of your response text already was actually the true text of the target embedding. An ideal randomizer would see that randomizing the first half almost always makes the result worse, and so would target the 2nd half more often (I'm hoping that embeddings work like this?).
7. Do this N times and use an LLM to score and discard the worst N-1 results. I expect that 99.9% of the time you're basically producing adversarial examples w/ this strategy.
8. Feed this last result into an LLM and ask it to clean it up.
Listening to this on 1.75x speed is excellent. I think the generated speaking speed is slow for audio quality, bc it'd be much harder to slow-down the generated audio while retaining quality than vice versa.
Sure. But the MRNA vaccines were the first ones to be available and the Myocarditis risk from getting COVID-19 while unvaccinated is still multiple times higher than the risk from the MRNA vaccine. So, all that telling people about the risk of myocarditis does is dissuade some portion of people from getting the MRNA vaccine... which leads to more cases of myocarditis/deaths.
You could say the same for myopia. But we still make glasses for people, which breaks the natural selection process that would drive that evolution.
because..?
> That we could possibly rearrange our food system so we don't need it anymore is better.
Will this be before or after we fix capitalism/finish building communism?
> Lastly - large social media platforms are utility communication services and public squares. They need to be regulated and treated like a government agency, restricting their ability to ban users and content. After all, so much of today’s speech is on these platforms. Not being able to share your ideas there is similar to not having free speech rights at all.
This is all well and good, but maybe the place to rehash a debate about whether vaccines work is not in fact, the center town square. I would say that a person who has no idea about any of the underlying science and evidence, but is spreading doubt about it anyway (especially while benefiting financially), is not in fact, 'sharing their ideas', because they don't meet the minimum standard to actually have an opinion on the topic.
In theory, the economics of this don't work out (Comcast/ISPs might be an exception). It would raise their onboarding costs a lot and raise their offboarding costs too. But if they're a local monopoly the might get away with it.
Yeah, try telling people that NNs contain actual understanding and comprehension. That won't be controversial at all.
I guess I see your perspective, but I kinda just saw that sorta thing as like... fighting for civil rights has always been a dangerous activity?
And that specific talking point to me always read like:
"Oh I have to wear a mask in a grocery store and can't go to movies, but they're allowed to protest for their civil rights??"
Can you explain the chain of logic here? During the pandemic I "did my own research" which amounted to basically masking when other people did and getting the vaccines as they came out. At the time my SO was a nurse working on a hospital covid floor, so it seemed prudent. So, I'm not really sure how you see Democrats as being less science based? No snark intended, I'm truly curious.
It doesn't. And it isn't like hosting a web-portal is some kinda alien technology that can only be done in the cloud. There's absolutely no reason that a robot vacuum couldn't serve its own web interface.
We are in both. Both universes are equally real. Each 'copy' of you knows it's in the universe where the result matches the observation.