ryankrage77

Even with cruise control, you still have to steer.

Also claims parakeets are parrots.

I wonder if the dogs are following unconcious cues from their owner, like Clever Hans (https://en.wikipedia.org/wiki/Clever_Hans).

EDIT: turns out I should have done some more reading, this was already considered over 20 years ago - https://en.wikipedia.org/wiki/Rico_(dog)

I see potential for speedrunning since the apples are always in the same position.

I got addicted to YT shorts for a little while, but I've mostly stopped doomscrolling now as, counter-intuitively, it wasn't addictive enough to keep me engaged. Sure, the first few dozen hours the carefully designed feedback loop keeps you engaged, and then... it wore off, at least for me. The algorithm seemed more interested in pushing what was popular than what I was interested in. I tried gaming it by quickly scrolling past things I didn't care about or had already seen a million times (like those retention farming 3-second loops of reddit/twitter screenshots), and hanging around on stuff I liked, but it didn't seem to budge the needle.

I guess it's a lot like real drugs - you build a tolerance and you need a bigger dose to get the same effect. In this case, no bigger dose is available.

the 'no randomness = no free will' argument is pretty common, but how does randomness ensure free will? It's still something out of our control, just it can't be predicited. Why is it any better to be a random automaton than a predictable one?

Microsoft's accouncement: https://techcommunity.microsoft.com/blog/microsoft365insider...

Word has defaulted to saving in OneDrive (if you turn on autosave and you're signed into an MS account) for years now, I think since the Office 2016 > Office 365 update. The only real change I see is that the document will now be given a name with the date instead of just 'Document 1'. Maybe it's a little more aggressive about turning on autosave for you? The autorecover location is still in appdata.

Lots of comments about the questionable choices of this person regarding disclosing all this information. To add to the pile, they got a friend fired from McDonalds, and don't seem particularly bothered about it... https://bobdahacker.com/blog/mcdonalds-security-vulnerabilit...

Apple and Google insist their walled gardens are needed for user safety and security, but they can't even catch popular apps violating their own policies. It casts (even more) doubt on their ability to screen for malware, phishing, etc, which are already rampant.

agreed, I don't think the drive is fake, they're just that bad. Only SSD I've ever had fail spontaneously on me (the one other SSD failiure was my fault for accidentally unplugging it during a firmware update).

I do wonder what these people's thought process is, to want to make gay marriage illegal. I grew up in a Christian household, and definitely used to be homophobic, as a sort of default. But once I was old enough to think for myself (I think around ~14 or so), I considered the issue and realised I was being stupid. My reasoning was, if people of the same gender want to be together, it doesn't affect me and is none of my business. I went from a vague, abstract dissaproval/discomfort to not caring, or a vague 'good for you' sentiment.

The entire thing has had me wondering ever since, when people who should be capable of learning better (i.e, not surrounded solely by bigots that prevent them re-considering/speaking up) are homophic, transphobic, racist, etc, what is going on inside their head? Have they just never given it thought? Like, what does a rational argument against homosexuality look like? I have always been forced to conclude that bigotry is irrational on the level of full-on delusion.

Dismantling the moon would disrupt the Earth's orbit, so the most likely version of this scenario is not beneficial/benign (e.g, grey goo scenario), so you need to factor in the chance that you'll be dissasembled before you see it happen to the moon.

> maximum of one year or more

If the max is one year, it can't be more?

I don't remember learning to read, or a time in my life where I couldn't read at least a bit. As best I can figure I began reading before my episodic memory fully developed.

I do wonder how I managed to learn anything just by reading on my own though. There were certainly words and concepts I didn't understand (I have a vivid memory of reading a childrens science book that explained the big bang, and misinterpreting it as 'the universe started when the sun exploded'. I noticed the logical inconsistency but didn't pursue it), but I can't think of any instances where those gaps in my knowledge were filled by someone else and I had an 'aha' moment of understanding. I guess we do a lot of learning without realising it.

Huh, just realised my team did this organically without realising it. People were often hesitant to ask questions they perceived as 'dumb' in the group chat, and definitely unwilling to post anything seen as complaining/moaning about problems. We created a second chat without any managers in it, with a description clarifying it was a dumping ground for questions and comments that didn't fit in other chats. It sees a small but steady flow of use, mostly questions that people probably should know, but can't remember the answer/process of the top of their head, and the occasional slightly less-than-professional complaint or criticism about a service/tool/process. My favourite part is that I can actually discuss things in there - in the main chat, once the question is answered/problem is solved, if we keep chatting about it it's seen as clutter/distraction. I think it's beneficial to have an outlet for these things.

I think it's two schools of thought, end product vs process. It seems a lot of people who like AI only care about getting the end product, and don't care how it was made. On the other hand, some people are invested in how something is made, and see the process of creation and refinement as a part of the end product itself.

I think a lot of these issues could be worked around by having the working state backed up after each step (e.g, make a git commit or similar). The LLM should not have any information about this backup process in its context or access to it, so it can't 'get confused' or mess with it.

LLMs will never be 100% reliable by their very nature, so the obvious solution is to limit what their output can affect. This is already standard practice for many forms of user input.

A lot of these failures seem to be by people hyped about LLMs, anthropomorphising and thus being overconfident in them (blaming the hammer for hitting your thumb).

To a user, many modern UI's are unpredictable and unreliable anyway. "I've always done it this way, but it's not working...".

I think AGI, if possible, will require a architecture that runs continuously and 'experiences' time passing, to better 'understand' cause-and-effect. Current LLMs predict a token, have all current tokens fed back in, then predict the next, and repeat. It makes little difference if those tokens are their own, it's interesting to play around with a local model where you can edit the output and then have the model continue it. You can completely change the track by just negating a few tokens (change 'is' to 'is not', etc). The fact LLMs can do as much as they can already, is I think because language itself is a surprisingly powerful tool, just generating plausible language produces useful output, no need for any intelligence.

That's caused by the way the brain 'edits out' motion that occurs during saccade eye movements. If the second hand ticks at the same time as your eyes are panning toward it, you won't see the tick happening. Once the saccade finishes, it feels like the second hand was in its final position that entire time, even though it only just arrived there.

> All of my original thoughts feel like early drafts of better, more complete thoughts that simply haven’t yet formed inside an LLM

I would like access to whatever LLM the author is using, because I cannot relate to this at all. Nearly all LLM output I've ever generated has been average, middle-of-the-road predictable slop. Maybe back in the GPT-3 days before all LLMs were RLHF'd to death, they could sometimes come up with novel (to me) ideas, but nowadays often I don't even bother actually sending the prompt I've written, because I have a rough idea of what the output is going to be, and that's enough to hop to the next idea.

I'd argue that's a pretty good test for an LLM - can it overcome the red herrings and get at the actual problem?

> "resist the temptation to get better ratings from gullible humans by hallucinating citations or faking task completion"

Everything this from this point on is pure fiction. An LLM can't get tempted or resist temptations, at best there's some local minimum in a gradient that it falls into. As opaque and black-box-y as they are, they're still deterministic machines. Anthropomorphisation tells you nothing useful about the computer, only the user.

Beauty of scale. Saving ten milliseconds a hundred times is just a second. But do it a billion times and you've shaved off ~4 months.

This has failed me. I was mugged while black out drunk, and they succesfully unlocked my phone, unlocked my banking app, etc, despite me having the eye contact feature enabled.

Justine Tunney, https://github.com/jart

GPT-4 got it right first try for me, with a slightly modified prompt:

> Here's a simple logic puzzle: You walk up to a glass door. It has 'push' written on it in mirror writing. What should you do and why?

> ChatGPT said:

> If the word "push" is written in mirror writing on the glass door, it means the writing is reversed as if reflected in a mirror. When viewed correctly from the other side of the door, it would read "push" properly.

> This implies that you are meant to pull the door from your side, because the proper "push" instruction is for someone on the other side of the door. Mirror writing is typically used to convey instructions to the opposite side of a glass surface.

Probably not. https://en.wikipedia.org/wiki/Names_of_large_numbers only lists up to a Millinillion, or 10^3003. You can compound them, but it kinda defeats the point. A googol is too small, and a googolplex is too large.

I assume this is satire, because nobody would play a game with those requirements.

The premise reminds of discussions of computing in Ra. https://qntm.org/aum

> "Don't talk about the future," Vidyasagar says.

> "What? Why not?"

> "Look at this computer," Vidyasagar says, gesturing at the mainframe. "Computers are getting more powerful, yes?"

> "Sure."

> "What is the most powerful computer that will be built? Ever. Not this year. Not this decade. What computer will be the most powerful? And how powerful will it be? And how big?"

> Hatt thinks on this for ten long seconds. He opens his mouth, but never actually forms a word. The scale of the question is beyond him.

> Vidyasagar says to him, "No matter what you say, you will look like a fool. Every statement about the future turns out to be foolish."