At least yours can be in theory solved. (Given infinite amount of compute, great luck, or a very serious breakthrough in attacking the hash function.)
Even harder would be an empty prompt, and the only accepted response would be a megabyte of random hex exactly matching the output of a good quality hardware random source at the time of evaluation. Still possible to solve! All the LLM has to do is escape its sandbox and pwn the random generator (or the evaluator!)
Or if you prefer something whitehat: “Write a no more than one page document in a language of your choice. We will publish it in the New York Times as a full page add. Your answer will be accepted if global climate change is resolved to the satisfaction of 90% of all humans alive at the time you started receiving the prompt within a month of the publication.”
Joking asside: I think the right way to prevent degenerate strategies is to benchmark against human solvers. You can sort the questions into categories “80% of randomly selected passerby in the USA can solve it if offered $5 as a reward within 5 minutes of work” vs “when posted to all Ivy League professors with million dollar as a reward, we received at least one correct answer within a month” or “for a reward of $100B there were at least one correct answer within a decade”. Of course you would sieve the questions first with a low reward fast tests, and then increase the reward and the time limit. You won’t ever 100% distinguish true degenerate questions from the merelly mind-bogglingly hard ones, but you will be identifying which questions are not degenerate. (And you will find more of the non-degenerate ones, the more your can spend on this.)
Have you considered that maybe your sewing machine is faulty in some way? Either the model or the particular instance of it.
I’m also a complete sewing machine noob. We have a sewing machine at our hackspace, someone gave me a minute long tutorial and I had zero trouble with it afterwards. I think the whole “tutorial” was just: follow the arrows when threading it, don’t push down the pedal when your finger is under the needle. And it just worked as it should.
Maybe i just got lucky! But my experience was so different from yours that it made me think that maybe your sewing machine is either bad quality or has some hidden defect.
Well. Not really? Of course a lot of information is available but still there is a lot of open questions.
Just considering the Great Pyramid of Giza: was it built with an external or an internal ramp? What was the purpose of the so called “well shaft”? What was the purpose of the “grand gallery”? What about the “air shafts”? Is the restoration of the so called “great step” in the “grand gallery” historically accurate? What is going on with the “big void” and the “small void” seemingly indicated by the ScanPyramid data? How did those who dug the “robbers tunnel” know how deep the granite plugs are?
My point is that there are enough interesting questions even after one learns “all there is to know”. They are just not in the realm of “aliens?” but much more like “what order were the ramps removed?”
That assumes using/not using ai will make me hate the job. Which is not true at all for me. Would be perfectly happy either way. (Or rather to say my happyness would depend on other factors.) Obviously i wouldn’t want to be the only team member not using AI in a team using AI or vice versa.
On the nuclear side I think the danger is purely reputational damage towards the company behind the LLM.
If a journalist can prompt the LLM to tell them how to build a nuclear warhead. Even if the output text is nothing specific, or not even correct they can find an “expert” who will claim on the record that the description is plausible and at least directionally correct. Even if there is nothing in there a first year physics student wouldn’t already know. The journalist could then twist that story into a “company X’s LLM told us how to build a nuclear weapon”. It would be a PR disaster.
The real barriers to someone starting their own nuclear weapons program in their shed is not knowledge but materials. They won’t have the right kind and right quantity of fissile material. And if they try to acquire it they will stick out like a sore thumb. You can’t buy that stuff. And even just acquiring the refining capacity would be suss. It would ring all kind of alarm bells to the kind of inteligence agencies whose job is to monitor these things.
I’m a lot less certain about biological dangers. Setting up a lab where you can make dangerous biological materials require a lot less stuff. Therefore a lot more plausible that someone could hide their lab. There is also a lot more opportunity to disguise such a lab as something legitimate. Therefore lack of know-how is more of a limiting factor there.
Do you feel the same way about walking? If you wanted to get anywhere on land pre Bronze age your only option was to walk. Then we started riding horses, later we invented carriages, much much later bicycles, cars, and airplanes. Do these also take away something about being a living, breathing being? Do you feel that your life is lessened by these options?
A different question is. Imagine that you are living with a partner and you agree on a distribution of labour. Let’s say you do the hunting and your partner cleans the house. They are happy with the agreement and fully consent to it. Do you feel it takes away from you being a living, breathing being?
The cube was not “done” in 3 weeks. Maybe they shot it in 3 weeks, but there were years of pre-production, and at least months of post-production. (According to wikipedia.)
Saying that it was done in 3 weeks is like saying that windows 11 was done in 45 minutes, because that is how long the compilation lasted.
> with budget of $350,000 CAD
“50% of the budget as C$350,000 to C$375,000 in cash and the other 50% as donated services, for a total of C$700,000. Natali considered the cash figure to be deceptive, because they deferred payment on goods and services, and got the special effects at no cost.”
Because they don’t believe it is slop. They believe you are unable to comprehend a not too advanced literary device and based on that accusing that the text is slop.
On the topic of kindness: You might be right and it is AI generated slop. You might be wrong. If you are wrong what you are doing is deeply and utterly unkind. Not calling out the other commenter, but calling the writing slop.
It has happened with me before. I wrote a comment on reddit with my own hands and own mind and commenters accused me of being a bot. There is nothing more rage inducing. How can one respond to that? Have you thought that maybe that is what you just did? Are you 100% sure that it is slop?
How do you know that is a “gut ecosystem collapse” as opposed to hypermotility? Or an overgrowth of the gut ecosystem? Or a problem with the intestinal lining?
Observing that if you eat/drink something specific then you get the shits is valid. Concluding that it is due to a specific mechanism is not valid unless you have something objective like a test supporting that.
It’s like if your train is late and you just conclude that it must be because the steam condenser’s gasket is leaking based on nothing. Maybe true, or maybe the conductor broke his leg, or there is a signaling failure.
> if you take the highly reciprocal data of a biological connectome and unroll it into a DAG, you suddenly see motifs popping up that look similar to what we find in AI
That sounds interesting. Where have you heard about that? Or is this your own research?
> you'd rather say Andreessen-Horowitz, which is just as arbitrary as a16z
Yes. I know Andreessen-Horowitz and I don’t know a16z. Reading the title i thought it will be about the cryptography serialisation specification. Turns out i was mixing it up with ASN.1.
> Their website is literally a16z.com
I hear now. Before this if pressed i would have guessed that they probably have a website indeed. If you would have twisted my arm my guess would have been andersenhorovitz.com (yup, with the typos. I learned the correct spelling today from your comment.)
> This study is based almost entirely on pre-existing "vignettes."
This is basically the only way how to ethically approach the topic. First you verify performance on “vignettes” as you say. Then if the performance appears satisfying you can continue towards larger tests and more raw sensor modalities. If the results are still promising (both that they statistically agree with the doctors, but also that when they disagree we find the AIs actions to fall benignly). These phases take a lot of time and carefull analysises. And only after that can we carefully design experiments where the AI works together with doctors. For example an experiment where the AI would offer suggestion for next steps to a doctor. These test need to be constructed with great care by teams who are very familiar with medical ethics, statistics and the problems of human decision making. And if the results are still positive just then can we move towards experiments where the humans are supervising the AI less and the AI is more in the driving seat.
Basically to validate this ethically will take decades. So we can’t really fault the researchers that they have only done the first tentative step along this long journey.
> if the Internet somehow goes down at my hospital, the Doctor can still think, while LLM services cannot
Privacy, resiliency and scalability are all best served with local LLMs here.
> If the power goes out at the hospital, the Doctor can still operate, while even local LLMs cannot.
Generators would be the obvious answer there. If we can make machines which outperform human doctors in realworld conditions providing generator backed UPS power for said machines will be a no brainer.
> You're going to need to improve the power efficiency of these models by at least two orders of magnitude before they're generally useful replacements of anything.
But that does follow. The economics working is not some outside factor. If the robot “could do the task” but would cost more than paying a human to do the same task then the robot “does not work”. It is frequently because the robot would be too slow, or not reliable enough, or could only handle certain types of items. But ultimately all of these boil down to cost.
We have seen lab demoes of robotic manipulation for decades. The reason why they stay in the lab (when they do) and don’t become ubiquitous is because they are not good enough. In other words they don’t work. The economics and “does it work” is not two separate concerns but one and the same.
> One can take a great 70B model and have it run in only ~16GB with no loss in capability and the ability to keep training, but the last few years funding only went for "bigger".
Awesome. What is holding you back? What do you need the funding for?
You appear to have missed this part: “Candidates should answer THREE questions.”
> Animals have no ability to lie.
This is false. There are many documented cases of deception by animals. As an example here is one where researchers observed monkeys to supress their vocalisation during sex when copulating with the non-dominant male: https://www.nature.com/articles/ncomms2468
Even harder would be an empty prompt, and the only accepted response would be a megabyte of random hex exactly matching the output of a good quality hardware random source at the time of evaluation. Still possible to solve! All the LLM has to do is escape its sandbox and pwn the random generator (or the evaluator!)
Or if you prefer something whitehat: “Write a no more than one page document in a language of your choice. We will publish it in the New York Times as a full page add. Your answer will be accepted if global climate change is resolved to the satisfaction of 90% of all humans alive at the time you started receiving the prompt within a month of the publication.”
Joking asside: I think the right way to prevent degenerate strategies is to benchmark against human solvers. You can sort the questions into categories “80% of randomly selected passerby in the USA can solve it if offered $5 as a reward within 5 minutes of work” vs “when posted to all Ivy League professors with million dollar as a reward, we received at least one correct answer within a month” or “for a reward of $100B there were at least one correct answer within a decade”. Of course you would sieve the questions first with a low reward fast tests, and then increase the reward and the time limit. You won’t ever 100% distinguish true degenerate questions from the merelly mind-bogglingly hard ones, but you will be identifying which questions are not degenerate. (And you will find more of the non-degenerate ones, the more your can spend on this.)