I feel like theres a time in near future where LLMs will be too cautious to answer any questions they arent sure about, and most of the human effort will go into pleading the LLM to at least try to give an answer, which will almost always be correct anyways.
It's not going to happen as the user would just leave the platform.
It would be better for most API usage though, as for business doing just a fraction of the job with 100% accuracy is often much preferable than claiming to do 100% but 20% is garbage.
There is nothing useful you can do with this information. You might as well memorize the phone book.
The model has a certain capacity -- quite limited in this case -- so there is an opportunity cost in learning one thing over another. That's why it is important to train on quality data; things you can build on top of.
Just because it's in the training data doesn't mean the model can remember it. The parameters total 60 gigabytes, there's only so much trivia that can fit in there so it has to do lossy compression.
Fair indicators, although I'd say the fresh user account and the lack of any concrete meanings in the statements are even more telling.
Saying this partly in self-defense – as a human who likes using en dashes and ellipses in my manual writing…
Besides, I wonder why people do this. What's the motivation? Is there some way of financially profiting from writing comments on HN that I'm unaware of?
If I may recommend, replace output LM386 stage with any dual opamp (e.g. another NE5532 or TL072, slightly different schematic of course), they can drive 32 ohm headphone speakers without issue and have significantly (~100x) lower white noise.
You can drive even 8 ohm headphones to unpleasantly loud levels with any opamp and a pair of transistors to beef up the output, along with a resistor to sort out the biasing. I did something like this as a headphone driver amp for "desktop mobile" radios used as part of a communications centre for a large festival. Motorola had a device that would do it, for about 500 quid each. I built the thing in the PDF at the bottom (I must have rerendered this at some point, it was definitely not done in 2022, more like 2012).
Using cheap bag-of-1000-for-a-fiver Chinese transistors off eBay I was able to get incredibly quiet output, to the point that I needed to add a muting gate because the radio was objectionably noisy. I notice that the exact transistors are not mentioned but any small-signal NPN and PNP ones will do - I used BC548 and BC558s, like I use in everything.
It will be way quieter and way more stable than an LM386.
Edit: I'm a lot better at drawing things in Kicad these days, and would have left the capacitors at the input a lot tidier.
Only in the first two stages. Output stage is LM386 which will be the source of the most of the noise. Replace the LM386 with another NE5532 (but modify the schematic of course, LM386 is single audio amp and has different pinout)
It works, I got 16mya in 5 minutes, one swipe is like 10000 years. My first thought was it should be faster but no, it should be slow to really show the scale. While I was scrolling I was hearing "age of war" soundtrack in my mind.
Thank you. Yes the huge amount of time it takes to scroll is kind of a feature, not a bug (in my mind)! The huge numbers of ancestors needed to see a step change is the point I was hoping to make.
Gasoline cars will be banned in 2035 and there ought to be some kind of on ramp so these giant American trucks probably won't meet emission limits anyway, right?
I couldn't see any citations or references in that video or its description. It presents it as him solving the problem himself, but I'm sure other people have written about solving the Game of Life in reverse with SAT solvers prior to this...
Garden of Eden[0] is a related concept where time-reversal comes up. Mentions NFAs, DFAs, backtracking and brute-force analysis. I see no mention of it but the latter was probably doing custom SAT implicitly, even if there was no access to our nice fast modern solvers.
I'm not sure how I am supposed to interpret your comment, what did you mean by that?
My aim was to start a positive dialogue with zppln about why the book didn't land with him or why it was off-putting. I'm happy to discuss it with you as well, but if your comment was just meant to be negative then I'll pass.
Former game dev here, drenched in cold sweat! Drop anything you are doing and immediately erase any mention of the 4-letter genre (words 3,4,5,6 in the title of your post). It is trademarked term and the holder goes VERY aggressively after it, including it's 4 letter acronym. I strongly suggest deleting the GitHub repo if it contains that phrase and create new GitHub repo without it. Use something like "interactive adventure" or something like that.
Even Netflix got sued over a Black Mirror spinoff that used the phrase. They countersued to invalidate the trademark; the suit was settled under undisclosed terms. I would love to know how that went down and why there wasn't an obvious path to success for Netflix...
