"AI will most likely lead to the end of the world, but in the meantime there will be great companies.”
Is there any serious journalistic source suggesting that this was anything other than an offhand joke? This article links to a youtube clip of the comment with context removed, but hair raising comments.
Taking the most uncharitable view of any person, you could imagine someone who was evil enough to cause the end of the world after their own lifespan where they faced no inconvenience, but not the circumstances from the quote
The quote as it stands is preposterous enough that I don't think a human capable of functioning in society would seriously say such a thing.
Are people wilfully misinterpreting the comment, or do they truly believe this an actually held opinion? If so, can they explain how they think someone could hold an opinion like that?
>What if you mark the untrusted user input explicitly in the prompt,
I think the more robust approach would be to have whatever embedding vector the model attributes to untrusted input and to directly attach that vector after every layer of transformation. Set a mask of where to apply that vector programmatically for every external input.
That way it gets forced back into line if some sort of internal rationalisation tries to semanticly drift away .
I wouldn't say so about Occam's Razor which is a heuristic.
The relationship between compression and intelligence, while not equal is definitely there. It looks like 3Blue1Brown is going to be doing some videos on this aspect.
>Truth is, it’s not that 90s anymore. Hardware has advanced to the point that you can have general purpose game engines that can be molded to any type of game. You do not need purpose built engines anymore.
I disagree, I think there's an over-emphasis on generating high quality individual frames and a expectation of what it is you should be able to do in games.
You can have a game that is photorealistic but you turn around and have your gun barrel poke into the wall and disappear. How many games can you throw enough junk into a river and make it change course eroding a new path for itself as it goes?
Some games rely on clear specific rules of an engine for the player to know because the rules are an integral part of the game, and any inconsistency in implementation creates a feeling of being cheated. Often you can implement such things in standard engines, but you are working against them the entire way.
You could have a game where a player sees a pylon and knows that because it is made of metal you could melt one of the legs and make it fall over. but to do that the entire construction of the game rules are integrated into the world. Most games teach the player that things like pylons are static objects unless they need to be destroyed for a plot point in which case just this one is different. Perhaps the player just has to learn that pylons are one of the class of destroyable things. Making emergent properties goes engine deep.
>And someday, if you can imagine, we’ll just have AI churning out visual representation of game state, turning game development purely into a declarative data driven exercise.
And therein lies the problem. A game engine is game state. You can make it pretty any number of ways, The engine will still be the thing deciding what you can do, and it is the things you can do that makes it play.
This kind of argument has been made since the days of renderware.
I have seen a number of projects go from
'We're building our own engine'
To
'we should have just gone with $engine_of_the_day'
To
'We were so lucky we chose to make our own engine'
If you want to make a game like fortnight, the Unreal is your pick. If you want to try something that hasn't been done before you could do worse than rolling your own engine.
Especially if you are looking for where the fun is, the idiosyncrasies of your own engine gives you a world with it's own flavour if you incorporate that flavour into your design process you could create a feedback loop that turns into something special.
Without even considering the veracity of the reports, cannabis is not the same thing as THC.
In illegal markets product safety is not assured, there are plenty of examples of cannabis being treated with other drugs, not to mention simple unsafe horticultural practices.
I had a cold a year or two ago, I had used the last of a bottle of actually working cough medicine, attempting to buy a replacement failing to do so, I found it had been banned. Turns out the active ingredient was morphine, which is apparently quite an effective decongestant in very low doses(and variants have been used as such for thousands of years)
That started me on a quest for a good cough medicine. My research turned up some interesting results.
Different countries have their own go-to medications for decongestants. Actual different drugs, not just the same thing under a different name.
None of them work.
Anything that does work is banned unless the people making it have been doing so for a couple of hundred years, then it seems that they just pretend it's not there.
When challenged the industry defended their sale of ineffective products by saying if they didn't work, people wouldn't buy them.
Anecdotal evidence is taken as gospel in many public health circles. When an attempt to have a substance banned was rejected on the basis of no evidence of abuse, it appears that means 'try again later', not gather evidence. You only have to wait for someone who shares your ideology once and it gets done and getting it undone is too much work for anyone to consider.
It's almost enough to send you off to a music festival to gather your own supplies for home remedies.
Cost per tokens is as valid as price per unit volume of fuel.
Changing the fuel type, efficiency of your vehicle, driving distance, or driving conditions will all change how much it will cost you.
Fuel cost per unit volume does not become meaningless just because you are neglecting all of the other factors involved. That would be throwing away the only data point you have been using.
This is just asking for someone to amalgamate all of the factors involved into one simple, easy to game, index.
If the models are quantifiably different on cleaner codebases, surely this could be leveraged into a measure of code cleanliness.
I'm not sure if you would call it an objective or subjective measure, a fixed model would be consistent which would provide an objective base for comparison, but other models would be different, so it would be subjective in relation to the model itself.
I was a little disappointed that the upper range of gravity on earth only goes to 9.8337. Just a little more and there would have been somewhere on earth that was an exact match.
It would have been the ideal (if chilly) place to start a cult.
I'm not sure it should, or even would, work like that.
The future I see is more specialised software. Perhaps sewage management workers need something done, which currently is put up for tender and sales people make claims of what it is you really need, promise the world and a pony, then management looks at the list of promises from everyone. They don't much notion of the problem at hand and the promises are deliberately vague on tricky details, so they go with the one that was cheapest. Then the workers get something that doesn't work and they spend a year doing service requests to get what they want while going massively over budget.
Instead of all of that, if a few of the front line workers had enough knowledge to wrangle a LLM, they can solve the problem they actually have. In that instance you are not paying for the software but paying someone for their time to make it.
This used to be how things worked in general. It wasn't until things could be duplicated for free that there was incentive to jealousy hold onto your ideas and use the monopoly of intellectual property to serve copies to people.
This makes it a pure numbers game and gave vendors an incentive to lock customers in. Enshitification is the blunt end of intellectual property hitting you in the face.
I don't plan on charging for the things I generate with AI.
I could conceive of being paid to make a piece of software using AI, but barring any privacy concerns I'd want the end product to be released as public domain.
Are models any good at descerning motion from multiple frames?
For instance if I gave models multiple animations of a bouncing ball as individual frames. Would they be able to tell which bounce was the more realistic motion.
(Is this a potential new benchmark? maybe also variations of stair dismount)
I think if an AI solves a problem that has been known and unsolved for a substantial period of time dispute attempts then the solution could only be considered non obvious.
If we make AI that can do 6 of these things before breakfast then we should think of them as easy to obtain.
The distinction is that non-obious to a human was a property that denoted a degree of specialness. If AI could do those things with ease then they cease to be special.
It was that factor that led people to be awarded some form of monopoly over the creation. But if it is no longer particularly special, then it should be public domain.
I'm not suggesting you tell no-one about your ideas, but if you can't convince people who know the field, turning to laypeople instead is the hallmark of a crank.
Extraornary claims should require extraornary proof, not a credulous audience.
"AI will most likely lead to the end of the world, but in the meantime there will be great companies.”
Is there any serious journalistic source suggesting that this was anything other than an offhand joke? This article links to a youtube clip of the comment with context removed, but hair raising comments.
Taking the most uncharitable view of any person, you could imagine someone who was evil enough to cause the end of the world after their own lifespan where they faced no inconvenience, but not the circumstances from the quote
The quote as it stands is preposterous enough that I don't think a human capable of functioning in society would seriously say such a thing.
Are people wilfully misinterpreting the comment, or do they truly believe this an actually held opinion? If so, can they explain how they think someone could hold an opinion like that?