Birds of feather and all that. I was lucky enough to end up in a class full of kids much smarter than me.
As for LLM replacement for talented teacher,even though I kinda worry that it would be subverted by organizations and various interests intent on stripping anything of value from LLM thus rendering LLMs role as a talented guide/teacher role largely useless, I personally found exploring new subjects even more engrossing than ( at one point in time, following down the rabbit hole of Wikipedia entries on some obscure subject ).
Part of the problem is that this thing would need to be marketed as safe, but safe is staying within rigid parameters that do not allow for a genius level individual to grow. Smart is probably a lot easier so safety features will likely not be triggered that often.
The other problem is that only some kids will take advantage of that mode. Not everyone is inclined to explore like that.
I have no real solution here. My kid is not at the age I need to worry about it yet, but I am slowly starting to plan my approach and I think tuned LLM with heavily restricted digital access will be the initial approach.
<< there should be an app like an 'auto-RAG' that scrapes RSS feeds and URLs,
I am not aware if that exists yet, but the challenge I see with it is rather simple: you get overwhelmed with information really quickly. In other words, you would still need human somewhere in that process to review those scrapes and the quality of that varies widely. For example, even on HN it is not a given a link will be pure gold ( you still want to check if it fits your use case ).
That said, as ideas goes, it sounds like a fun weekend project.
<< That's not the state of today at all, and probably doesn't represent the near or medium future.
Thank you for saying this. I briefly wondered if my particular company is just way behind or particularly dysfunctional and disorganized ( a possibility for sure ). I do agree with you observation on LLMs effectively lowering entry level skill ( yesterday I was able to dissect xml file despite it not being something I could normally do without any prep work and despite mildly unusual - I thought - formatting choices by vendor ). There was still a fair amount of back and forth for what some enthusiast would call 'perfect prompt' and interesting bugs that had to be addressed, but, having seen the daily mess at my company does not exactly make me a full blown evangelist. I see it more as a get to the wrong answer faster. That is the part that concerns me.
<< Yep. I often reflect on some of my career mistakes, especially when evaluating current decisions within the context of a job or interacting with other institutions, and what I wish I'd learnt earlier would be to "read a room".
While I agree and even accept this answer in theory, I have a hard time putting it in practice. Just today it seems I unnecessarily ruffled some executive feathers by pointing out some -- otherwise clear -- issues no one dares to mention and I am wondering now if that was even worthwhile. After all, I am not paid for extra for it. Regardless of the choice made by executives, the only thing that would change is the amount of support work I would do for it.
I know for a fact that 9 out of 10 it is better for me ( and my career ) to stay quiet, but sometimes I just can't stay out of my way.
I did toy with the idea of trying do analysis of HN aliases and keywords. It never went anywhere, because I forgot about it, but a longer weekend is coming:D But yeah, language betrays, who we are in references alone.
To be honest, I am genuinely surprised an attack never materialized. But then I also remember mentioning my thoughts on the matter to my wife, who was aghast that I would even consider such a scenario. Maybe, on average, people are actually decent and it is people like me, who come up with weird hypotheticals.
>To be honest, I am genuinely surprised an attack never materialized.
I'm not. The TSA is hated by the populace the way the population hates every wasteful boondoggle jobs program. Foreigners hate them for profiling but that's pretty far down the list of grievances. Foreigners looking to strike at America and Americans looking to get off the porch likely have dozens of more preferable targets.
I doubt the TSA makes any would be attackers list of top five most deserving agencies.
Edit: A fed sponsored false flag attack on the TSA could make a good comedy plot. It might need to be a TV series in order to have time to fit in all the jokes, references, tropes and wise cracks you'd be obligated to make when covering such subject matter.
You mean I’m not the only one who has wondered if you could etch a knife shape onto the back of an acrylic clipboard? Something that wouldn’t show up on X-ray, but could be punched out with little effort on the plane?
My wife says it’s a wonder I’m not on the no fly list.
Oh, I never considered that:D It is an interesting use of materials, but seems unnecessary. I was thinking about a scenario in which an assailant would attack an actual soft spot in an airport. That did not happen. That is the part that surprises me ( and I am glad I remain surprised ). Then again, you start reading about new and exciting ways of detecting guns and you start wondering if that is basically trying to fight an old war.
<< My wife says it’s a wonder I’m not on the no fly list.
For better or worse, my work almost guarantees that I will be looking for mildly interesting subjects so I am likely on some sort of grey list ( no idea what IC would call it internally ). No 'no fly' list yet, but these days it only takes being extra rude it seems[1] so these days that list is a major inconvenience, but the signal it generates is less useful:D
Reinforced doors and a belief among everyone from passengers to pilots that a successful hijacking is likely to lead to death rather than inconvenience reduced hijacking after 2001. I would be surprised if searching passengers more aggressively or requiring people to prove their address to get a driver's license had much to do with it.
Yeah, I think passenger game-theory is the number one reason, and that was already in place on September 12th, 2001.
Passengers are now aware that hijackers might actually just buying time until they can trigger a murder-suicide attempt, and many will believe it is likely enough that they need to fight for their own lives. The old assumptions that "almost everyone gets out alive by passively cooperating" no longer hold.
Who's we? Al Qaeda reduced it down to ~4 per year because now passengers no longer assume to survive and act accordingly. Locked doors and plain clothes security are just icing on the cake.
Look at all those other pre-2001 hijackings, and ask yourself what would have happened if most of the passengers were terrified that the hijackers were preparing to destroy the entire plane and everyone on it, regardless of any demands being made or met.
Would-be hostage-taking hijackers know it too: Their business plan, as it were, has been ruined for a generation by their suicidal colleagues.
Airline and security protocols are no longer "comply with everything so that nobody gets killed". Doors are locked. Passengers will likely bum rush you. There might be armed security on any given flight. The odds of success are unbelievably long compared to what they were in the 80s. Just not worth it vs a "boring" bombing or whatever.
It gets worse. In US most of the bigger corps institute some sort of means to authenticate you via cellphone, which means that if you want to be remote, phone is effectively a necessity ( which one usually does ). Only a year ago, it was still possible to avoid having to have a cell ( although that meant you had to be in person -- an interesting trade off in itself ).
At my last workplace, I somehow managed to get away with only Microsoft Authenticator on my phone, with no actual remote management capabilities enabled. That's pretty much exactly where I draw the line: if I have to have a device to perform work functions, the workplace needs to supply it. I'm not going to put work data on my personal machines, and I'm definitely not letting a third party root my phone for me "for sekhurity", and apply work policies on my personal device. I'm okay with work 2FA on my phone, but only without MDM, as an exception for where otherwise there's no reason for me to have a work phone.
These days a lot of folks can probably do more than just authenticator on their personal device. Teams and Outlook, for example, are both able to run with the MDM-level controls the company wants but without the device-level MDM. It's part of the app and has no control over anything else.
A company I previously worked for had a policy that if you had any company data on your phone, they had the right to force you to unlock it and look through it (not sure if they ever actually did but it was in the employee handbook). When IT tried introducing a system that required me to Auth with my phone I refused, citing the policy, and they helped me setup a workaround Yubikey.
Might not be possible everywhere but worth a shot. Also always helps to make friends in IT.
Most places can issue you with a physical token instead, like a Yubikey.
It's just unusual, so the first line in helpdesk don't always know about it. And people seldom want to start a battle with the bureaucracy on their first day on the job...
Manna was fairly eye-opening ( and you can see some parallels to today's LLMs to me. I will admit that I read it without knowing much about the author way back when and being fairly amazed at well he knew human nature and likely course that invention would take.
What a time to be alive, I read it the opening fully expecting to see an open source automated trashcan that takes itself to the curb each Monday. I was disappointed to find out it is about an actual garbage collection algorithm.
It is interesting and paints rather annoying future once those are cheaper. I am glad this research is conducted, but I think here the measure cannot be technical ( more silly guardrails in software.. or even blobs in hardware ).
What we need is a clear indication of who is to blame when a bad decision is made? I would argue, just like with a weapon, that the person giving/writing instructions is, but I am sure there will be interesting edge cases that do not yet account for dead man's switch and the like.
edit: On the other side of the coin, it is hard not to get excited ( 10k for a flamethrower robot seems like a steal even if I end up on a list somewhere ).
As for LLM replacement for talented teacher,even though I kinda worry that it would be subverted by organizations and various interests intent on stripping anything of value from LLM thus rendering LLMs role as a talented guide/teacher role largely useless, I personally found exploring new subjects even more engrossing than ( at one point in time, following down the rabbit hole of Wikipedia entries on some obscure subject ).
Part of the problem is that this thing would need to be marketed as safe, but safe is staying within rigid parameters that do not allow for a genius level individual to grow. Smart is probably a lot easier so safety features will likely not be triggered that often.
The other problem is that only some kids will take advantage of that mode. Not everyone is inclined to explore like that.
I have no real solution here. My kid is not at the age I need to worry about it yet, but I am slowly starting to plan my approach and I think tuned LLM with heavily restricted digital access will be the initial approach.