It's amazing to me how nobody seems to know about the short story "The great automatic grammatizator" by Roald Dahl. Nobody got closer than him. I feel like I should be reading about it all the time and no one seems to have ever heard of it.
“There are many other little refinements too, Mr Bohlen. You’ll see them all when you study the plans carefully. For example, there’s a trick that nearly every writer uses, of inserting at least one long, obscure word into each story. This makes the reader think that the man is very wise and clever. So I have the machine do the same thing. There’ll be a whole stack of long words stored away just for this purpose.”
“Where?”
“In the ‘word-memory’ section,” he said, epexegetically.
Roald Dahl also wrote a story about two dudes who wanted to try each other's wives without getting consent from said wives so they swapped places in the middle of darkness and then the next morning, one of the dude's wives said to her husband, "Holy shit, whatever you did last night was amazing. I never liked doing the hot dog dance before but if you can keep doing what you did last night, I'll always be down!"
Because the horror of dahl’s adult stories are as pervasive even if knowing the ending. I reread many times and still get the same sense of impending doom barbarically twisting fates in the mind - what if it was true?
Interesting, this one did pop up on my radar a while back, i always keep my pulse on the playtester space :) Although cardtavern one strays away more from the free form play that simulates a kitchen table game of magic, piles and structured zones ect, much like untap.in has (as i mentioned, users want fast structured play and QOL mechanics built in). The ops vision seems to be its a table with some hide/reveal and shuffle mechanics, do what you want after that. Which i appreciate, allows a bit of fun.
I don't quite understand how to use this, but if people are looking for a way to play magic digitally I might recommend the site I built at cardtavern.com.
Not sure he's "on the outs", he on Shopify's board.
Sidekiq's solo dev (Mike Perham) has for many years made a generous donation to Ruby Central. He informed them that he didn't want his money to be spent platforming dhh at their conference, they ignored his request, he stopped his annual donations.
Me too, and because of that I feel it's even more important to use language like racist, white nationalist, and fascist when describing him and his ilk, because that's what they are. Softening the language only leads to those beliefs becoming more normalized than they already are.
If you’d like to read, in his own words, his “coming out” as an ultra right wing racist piece of shit, feel free to look on his blog for the post titled “As I Remember London.”
I built a daily puzzles site at https://dailybaffle.com, and I'm working on promoting it and releasing the mobile app for it this month. Turns out it's a lot of work to promote things!
I've been able to get something like 25 interviews in 2 months despite having long gaps on my resume and nothing especially impressive to my name. So I suspect you might be going about this wrong. I haven't gotten an offer yet, that's another story, but getting the interviews hasn't been hard. Applying in NYC/SF, senior-only.
I honestly have no idea. The last place I worked is pretty well-known. Not big tech, but a recognizable name to most people. I send out a lot of applications: those 25 interviews are the result of 150 applications in the last two months or so. And then I have my linkedin set to be discoverable and looking for a job. Basically just fiddle with the options under Visibility and Data Privacy in the linkedin settings and a bunch of people start reaching out to you immediately. I also think I have a nicely formatted resume, really readable.
So are the majority of these applications the result of recruiters finding you via LinkedIn, or have you been applying direct as well? What application path have most of the interviews come from?
Location has always been a huge factor in these discussions. There are usually significantly less opportunities outside of hubs. It’s a cart/horse problem- because companies go to those hubs to hire due the talent pool.
The part that eludes me is how you get from this to the capability to debug arbitrary coding problems. How does statistical inference become reasoning?
For a long time, it seemed the answer was it doesn't. But now, using Claude code daily, it seems it does.
IMO your question is the largest unknown in the ML research field (neural net interpretability is a related area), but the most basic explanation is
"if we can always accurately guess the next 'correct' word, then we will always answer questions correctly".
An enormous amount of research+eng work (most of the work of frontier labs) is being poured into making that 'correct' modifier happen, rather than just predicting the next token from 'the internet' (naive original training corpus). This work takes the form of improved training data (e.g. expert annotations), human-feedback finetuning (e.g. RLHF), and most recently reinforcement learning (e.g. RLVR, meaning RL with verifiable rewards), where the model is trained to find the correct answer to a problem without 'token-level guidance'. RL for LLMs is a very hot research area and very tricky to solve correctly.
Because it's not statistical inference on words or characters but rather stacked layers of statistical inference on ~arbitrarily complex semantic concepts which is then performed recursively.
This answer makes sense if you know that LLMs have layers, if you don't this answer is not super informative.
If I were to describe this to a nontechnical person, I would say:
LLMs are big stacks of layers of "understanders" that each teach the next guy something.
Imagine you are making a large language model that has 4 layers. Each layer will talk to it's immediate neighbor.
The first layer will get the bare minimum, in the LLM's of today, that's groups of letters that are common to come up together, called "tokens". This layer will try to derive a bit of meaning to tell the next layer, such as grouping of letters into words.
The next layer may be a little bit more semantic, for example interpreting that the word "hot" immediately followed by the word "dog" maps to a phrase "hot dog".
The layer after that, becoming a bit more intelligent given it's predecessors have already had some chances at smaller interpretations may now try to group words into bigger blobs, such as "i want a hot dog" as one combined phrase rather than a set of separated concepts.
The final layer may do something even more intelligent afterward, like realize that this is a quote in a book.
The point is that each layer tries to add a little meaning for the next layer.
I want to stress this: the layers do not actually correspond to specific concepts the way I just expressed, the point is that each layer adds a bit more "semantic meaning" for the next layer.
DNNs aren't really "statistical" inference in the way most people would understand the term statistics. The underlying maths owes much more to calculus than statistics. The model isn't just encoding statistics about the text it was trained on, it's attempting to optimize a solution to the problem of picking the next token with all the complexity that goes into that.
One problem is that "statistical inference" is overly reductive. Sure, there's a statistical aspect to the computations in a neural network, but there's more to it than that. As there is in the human brain.
This literally says nothing - are we supposed to infer that they are putting the product into maintenance mode and will no longer be developing new features for it? This is a masterpiece of corporate nullspeech.
I love the idea of this site but have always been disappointed by the fact that it's more of a slideshow than actual animations. You have to do a fair bit of interpolation if you aren't experienced.
reply