prometheus1992

This only shows the lack of understanding the majority of readers have about post training LLMs. This capability did not emerge out of nowhere. There are problems in the post training datasets which are incredibly similar to the ones in TFA, where you essentially abstract out the 'jailbreak' or 'cheat' - in simple language - they are training the models to find short cuts or cheat. So it shouldn't be surprising that the model generalized that ability. But, both the writers and readers of this type of content act out surprised and mesmerized and it makes me think that we need to improve our understanding of how LLMs are trained and built.

I don't like the direction this is going in - so a third party is going to be able to control a robot in my house?

we shouldn't aim for economies of scale on this problem as there are massive negative externalities attached to it. Hopefully, a chinese lab will release an open source version soon.

100M free tokens? wow.

This is a showpiece at this point. but I can see these on the roads in ~10 years; but it felt more like a frog car than a flying car? in my mind flying quadcopters felt much safer in videos than this frog. It feels overly complicated based on what i have read; rocket boosters?? why? quadcopters and other flying taxis seem much more economical and using tech that's been reliable for at least last 20 years.

not trying to be controversial here; but isn't it weird that you can buy semi automatic guns and there's restrictions on radio equipments bc these equipments can mess with the bandwidth used by public services - but similar arguments hold for guns as well.

you are absolutely right

because iPad is my primary reader at this moment.

>>The only thing that sucks both about the iPad and most ereaders is the lack of physical buttons to flip pages, which makes 1 handed operation clumsy at best

I was just thinking about this yesterday; to have either a voice controlled flipper or an eye tracker where i lock and blink at the bottom right corner to flip or something like it; but couldn't figure out the battery drain that these systems will cause to have the models always on.

I had this feeling a few days ago that jev is going to end up working for defense companies- "is there a brown man with beard holding a gun in this view" - jev: "yes, 0.9", "is there a child standing in the same view" - jev: "no, 0.65" - decision - blow it up. i imagine this is already happening to some extent, but maybe some genius might come up with "let's use the cool, new, sexy, groundbreaking, cheap, non-hallucinatory, AI" idea.

for anyone who already has a colored reader - is it better than ipad? i enjoy reading on my ipad and i can read for 2-3 hours on it daily without feeling any strain on my eyes.

>>The biggest issue with sharing a border with the state is the fence line

Is it the physical blockade that the fence provides or just "where is the boundary?" - my guess is that it's both. While physical fence is hard to sustain due to trees falling but could there be an electronic solution for keeping the boundary line intact by putting some sort of gadgets underground which you can always ping to see where is the boundary?

if anyone is well versed with this tech; can this be used to target regular people? how can an average joe protect themselves from this?

Thanks for sharing. i hope you get through this.

pangram is supposed to detect AI writing, gemini is AI!!

so what was karpathy expecting?? everyone will give up on their life and start feeding the monster in the screen? that's literally what the tiny sliver is doing; they are food for the AI model.

>You could look around in 2016 and 2026 and honestly not much has changed

I find 2026 much more repulsive compared to 2016.

easy to trick pangram.

what i understood from the above comment is that the government should first do the basic policy making (set some non-negotiables for the ai companies etc) before jumping into solution building. let the market figure out the tech solutions.

I think this is a reasonable take and I agree with you but its a bit unsettling that the event of fund raising has become a more important variable in the equation than the fundamental innovation itself. And I don't think this is a good long term strategy- it almost gives me "let's con our way out of this" vibe.

I can totally sympathize with the author and admittedly I sometimes put myself in the same camp out of frustration, given the current events and news. I will however say that I have also come to realize and accept that human greed can turn any good idea into an awful mess. We are not programmed to understand when something is good enough. We don't know what is enough money, enough resources, enough power, enough speed, enough attention etc. No matter how much technology we develop we won't be able to solve for human greed. Look at OpenAI - they can allegedly solve navier stokes but can't report honest revenue numbers?

>>I was training custom ML models back in 2017

but you truly do sound like an angry 19 year old from your arguments.

- accurate - on what? on trust me bro benchmarks?

- zero-shot model are fundamentally general purpose.

- fast and cheap ; models on hf are FREE and fast enough.

maybe you entered the AI space during the vibecoding era but there had been ton of useful models before that. especially zero shot models- both for text and images.

why wouldn't they benchmark the accuracy against jev too?

I really don't understand how this can be. I have sat in fund raising meetings with VCs in toronto and my experience is that there is shit ton of due diligence at the tech level. a product which has no moat, was already available, was duplicated within a couple of days is valued at 7B - i thought we were past the peak of the hype cycle.

>>Incredible lack of curiosity

Maybe on your end because that's the exact point the author explained - these companies need to adhere to the scientific community rather than expecting the opposite.

If the write up - the natural language part - sucks - why are you expecting humans to waste their time on understanding the proof???

This is super creative!

is there no israeli lobby in washington though? have you heard of AIPAC?

you seem pissed that they wasted a noble on someone that you don't approve of. do you not think that it might be YOU who might not have the ability to understand every field with as much clarity as you understand software after reading a few lines.

so these are just indian companies and microsoft which is lead by an indian guy? is the government signaling something here? i know amazon, meta, google are all doing the same thing- but amazon and meta are licking the ass.

giving the same excuse of "oh these companies post jobs where they are not seen" is the lamest excuse you can give. are you saying this government is so poor and incompetent that it cannot verify where the job was posted? how many people saw the jobs, how many clicked and applied?? are you really saying that the most power country, with SI is not able to do this basic check??