It is like having Google's MusicML output a mp3 of saxophone music and then ask what proof is there that MusicML has not learned to play the saxophone?
In a certain context that is only judging the output, what is meant by "play the saxophone", the model has achieved.
In another context of what is normally meant, the idea the model has learned to play the saxophone is completely ridiculous and not something anyone would even try to defend.
In the context of LLMs and intelligence/reasoning, I think we are mostly talking about the later and not the former.
"Maybe you don't have to blow throw a physical tube to make saxophone sounds, you can just train on tons of output of saxophone sounds then it is basically the same thing"
I just always assumed an "open office", really meaning a non-office, an empty building, was what was available for startups after the dot com bust in SF.
Then after the fact we made up a bunch of bullshit as to why this is some brilliant idea. Then this idea spread as if it was some kind of technological advancement because it worked for small tech companies trying to not spend money on furniture and walls.
We just aren't very good at any of this at scale. The open "office" and battle against remote work are different flavors of the same type of stupidity.
I think it is a really great article but I am a chef too already familiar with the ingredients and style of cooking.
I think the main issue is getting people to try new dishes.
Of course, the dishes have to be good and not just the chef fooling around. Sometimes this style seems like "hey try this new exotic dish! It is made from mixing organic ice cream, grass fed beef, fair trade coffee and sriracha. Oh you think it sucks? Well that is because your taste buds are use to standard western food!"
I think of how no one needs an intellectual explanation as to why pad thai is good even if they only ever had pizza and burgers.
I don't see how it really matters. A bigger issue is that some days there is just going to be nothing to report if you want "real" news.
"Today nothing happened, the end" would not work. So you would have to lower your standards for that day. On that day you would conflate what happened with entertainment and low and behold that day nothing actually happened is more popular than reality.
Loop this process over and over and we get what we have now.
I suspect we end up at the point we are at now no matter what the initial starting conditions or how you design the system.
"News" is a form of entertainment and to pretend it is not seems completely delusional to me.
I think it is like asking how do you get people to watch a movie of a professor giving a statistics lecture. You have to publicly fund it because no one is going to really watch or pay for that movie.
25 years ago I could never have imagined the internet we have today.
25 years from now we could easily have a free internet protocol that is taking hold and the whole process repeats.
So much has came and went in a single generation but it works both ways. It is not an eternally negative, march towards complete totalitarianism even if it it feels like that right now.
I am guessing you have never worked at a giant corporation with 5 levels of useless middle managers that have no purpose if there are no office politics to manage?
Those are the managers that hate work from home. Not managers actually doing productive management.
I don't personally see how an individual can judge this at this point unless it is a huge leap.
More importantly, if the model is not a huge leap at this point I just don't care if it is as good as the very limited models we already have because I am not impressed by any of these anymore.
Anything less than a 3.5 to 4 jump from here is just not going to vibe for me.
The truth is that we are very limited apes that are highly prone to self delusion and believing nonsense, trying to understand an impossibly complex world.
At the aggregate we seem to be making progress while at the individual level humanity always seems doomed. It seems a constant that the individual wants to pretend they live in the end times because at least then your time on this rock was a little bit special.
It is also the normal state of the economy to not job jump because it is normally hard to get a job.
The far from equilibrium state was software jobs with 0% interest rates the last decade.
Then you have people who entered the workforce during this time believing this is how things have always been and always will be.
Surprise! That was a temporary state we are never going back to in your lifetime.
It is like when I was young, every song had to have a guitar solo. To hear a song with no guitar solo was so weird. I assumed things would always be this way. Neither the guitar solo or the 2016 software job market is ever coming back the same way it use to be.
John D. Rockefeller argued against Standard Oil being broke up on the grounds that it only was so big because it is just that efficient and innovative.
It also kept the price of kerosene low for the consumer.
If it was broke up it would also hurt the economy overall. They also argued it wasn't really a monopoly and they have other competition lol.
Tech companies are such a brilliant monopoly they get the consumer to mindlessly puppet Rockefeller's exact points for them.
Because no one watches something like the shot put other wise, EVER.
It use to be the Olympics were the one showcase for the non-popular sports.
Case in point, the shot put. One of the long standing "unbreakable" records was Randy Barnes shot put from the late 80s.
Ryan Crouser broke that last year and won the gold medal this year, again.
It is an incredible accomplishment but no one cares because Lebron and Steph Curry were going for the gold or why watch the shot put when you can watch break dancing.
I feel bad for anyone who is not old enough to have watched it when it was all amateurs. It was something so special.
Just have the Taylor Swift karaoke event already. I am sure that would get great ratings and that is all that matters obviously.
We are also only counting the winners who all took huge risk. Many more took huge risk in the past and lost everything.
It strikes me as rather absurd to believe it was better in the past when capital was much harder to come by.
Imagine trying to get startup funding in the late 80s when a person with capital could get 9% annualized on a short term US government bill. Good luck with that.
I had never in my life heard of this before now and I am no stranger to hotels.
I would go straight to the front desk, demand my money back for that day and then never stay with hotel again along with making sure the corporate office got a nice email about their bullshit policy.
Maybe over priced Vegas hotels can do this but any hotel I have ever stayed at needs to make the customer happy because the competition is so fierce. Most hotels will go out of their way to make sure you are happy. Not randomly inspect your room like you are a child.
I thought the article seemed pretty bad so tried to read the actual paper too.
It really might be one of the worst papers I have read. It is basically a magazine article using more scientific sounding language.
There is a huge age/experience confound that I can think of from my own life. My answers would be so much different at 25 vs 55 after so many failed attempts at pair bonding. From my dating experience and life I would say the time average and ensemble average are not equal. Even forgetting about experience we know hormone levels change drastically over time. I can think of all kinds of ridiculous experiments in other domains that would be analogous to studying sex without accounting for sex hormones.
I don't know, my experience is all this is really hard to judge.
I would have a really hard time saying that 4 in April 2023 was better than 4o now though.
I have always wonder if it matters what time of day you are using it too. I feel like 4am EST works better than 4PM EST but it is so hard to judge. I think there is so much difference too with just how the prompt is phrased so it ends up feeling like some days it is is good and some days it sucks.
That is coupled with I have got a bad result before, opened a new chat window, pasted the exact same prompt and got a good result.
If I had to bet, I imagine it is like flipping quarters. Sometimes you will get runs of heads, sometimes runs of tails and sometimes a real mixed bag of both.
I am old myself. It is not ability to learn with age that is suspicious it is wanting to share it like this and it would be a clever marketing move.
I will learn new things until the day I can't. I am currently learning the Stan language but I am not going to write a medium article "What happens when an old man attempts to learn Stan?"
In a certain context that is only judging the output, what is meant by "play the saxophone", the model has achieved.
In another context of what is normally meant, the idea the model has learned to play the saxophone is completely ridiculous and not something anyone would even try to defend.
In the context of LLMs and intelligence/reasoning, I think we are mostly talking about the later and not the former.
"Maybe you don't have to blow throw a physical tube to make saxophone sounds, you can just train on tons of output of saxophone sounds then it is basically the same thing"
The enter discussion is ridiculous.