What you described would be a great solution for plenty of tasks, but maybe solving some fallacies one at a time would also be great. For example, when we're sure by placing a properly named file at a directory location, we will find it by recalling the folder name or the name of the file itself while in reality we're often surprised that after months or years this won't work, the expected path to the file either not exists or doesn't contain what we're looking for. The same fallacy is true also for different hierarchical notes organizers.
In this case LLMs with their ability to find semantic equivalence might be a great help. And with the current state of affairs I even think that an LLM with a sufficiently large context window might absorb some kind of the file system dump with directory paths and file names and answer a question about some obscure file from the past.
It's interesting, if we assume that the microbiota "wants" something (like other viruses or bacteria trying to affect the host in some way) and learned it evolutionary then when inside a human it tends to avoid other humans probably due to some disadvantages of the contacts for them. But why? It looks a little strange because humans are highly social species to the edge of survivability so the microbiota just risks losing the host completely.
From the release "..You can try out a preview version of Magistral Medium in Le Chat..", I suppose it's when the drop down in "Thinking" mode is either slow or fast (limited to 3 queries per day).
My favorite from the last months was asking for a string that for base64 produces strings with non-alphanumeric and non-padding symbols (so '+' or '/' should be in the output). It thought for 7 minutes and 74k of markdown length, and finally came up with the AB?C string that produces QUI/Qw== (correct). It is impressive, because general LLMs just always fail, but I didn't try other "thinking" models recently.
I think you invented (or reinvented) a simplified password manager, or a plain-text password manager. A usual PM solves the task of managing by human memory unmanageable: plenty of passwords with variable complexities routing them all to a single one intended for the human. In your system you have your own version of the master password (prefix + suffix) that locks out your actual password (a single substring). There are obvious drawbacks compared to a general PM like a much lower space of possible variants or needing to manage this manually (like generating the file or choosing randomly your prefix and suffix). But there is at least one benefit, if you keep the system simple enough (while not making it simple in generating the sequences), no hidden vulnerability should waiting to happen
Good points. Also looking at your original example I noticed that not only humans can explain regularities they expect in many different ways (also correcting along the way), they can basically ask LLM to base the result on a reference. In your example you provided a template with an img tag and brackets having different attributes patterns. But one can also just ask for a html-style tag. As I did with the "Please create a regex for extracting image file names when in a text a html-style tag img is met" (not posting it here, but "src" is clearly visible in the answer). So the "knowledge" from other domains is applied to the regex creation.
The programming languages are more predictable than human. So the rules are much easier to be "compressed" after they're basically detected when fed with big data. Your two examples imho are easily interchangeable during follow-up conversation with a decent LLM. Tested this with the following prompt and fed a c fragment and an SQL-fragment, got in both cases something like your first one
> Please convert the following fragment of a programming language (auto-detect) into a json-like parsing information when language construct is represented like an object, fixed branches are represented like properties and iterative clauses (statement list for example) as array.
From my perspective LLMs shine in programming languages because they (languages) are invented by people to be more formal so they are more predictable. Even for syntactically different programming languages there are similarities we don't see with our own eyes, but the training has no problem detecting. My favorite example is when I tried to find how to represent an int array for reflection in Kotlin. Ask ChatGPT or a sufficiently large model about this, the answer will be IntArray::class.java , but try to find the exact line with google - few or nothing, more on github search in the sources, still not much. So LLMs "detected" the system of making type signatures in Java/Kotlin and were able to successfully predict because the rules are consistent. In human language it also works, but to a lesser degree, so if you give it a verb/subject pair that makes sense and ask for the equivalent ones, you will get some that still make sense but are literally absent in the full corpus of the web.
follow-up/spin-offs ideas about using the pieces of information important to your life
* script of the movies you like to better remember/recall interesting quotes
* the infamous spaced repetition method applied to this game would allow "two in one trening"
I like reading about ideas even if they're half-cooked because they often inspire something else.
Good point. Some time ago I came up with a philosophical interpretation that although literally time machines don't exist, humanity already invented them, they are called photos and videos. You can't change the past, but you can visit it
Showing a random photo from the past is a nice feature. Used to have it on an Android tablet where an idle lock screen slideshow allowed this. There are also windows in different programs where I don't mind having a random picture as a background. Looks like a "New tab" page in the browser is ok for this. Anyone know how to set this up in Firefox?
I'd argue that there's another perspective, more complimentary to US politics. There's obviously a list of "experts" aligned with RT narrative which they use depending on the topic. No media like RT wants someone from this list to be too visible and appear too often, whether they are pay-rolled shadowy or not. So it looks like not so many "experts" are eager to be on alert on this list.
Good point. On the Daily podcast (NYT) Goldberg was asked what was in the chat after the strikes. He said a plenty of different emojis. Goldberg's reaction was also interesting, he thought at the moment that every workplace is the same. Personally I'd expect some serious mood and attitude when there are so much at stake including people lives, but humans are humans
I wonder whether a possible audio similarities analyzer would be capable of detecting influences described by the creators themselves. For example, in a podcast, Tal Bachman said that making his most famous hit "She's So High" started when he heard the chorus of "If It Makes You Happy" by Sheryl Crow in a supermarket. After that revelation I now can not unhear it and the choruses of those songs are connected in my brain. But the connection is not that obvious.
With eInk screens it's an interesting story. Because when your page is shown, there is no need for CPU at all for keeping it, so in an ideal world for such devices every page flip is a micro-cycle of waking up from the deep sleep, doing a job and going back to sleep. With such a lifecycle it's even better to measure the battery life not in hours but in page flips. When KOReader supports going to the deep sleep between flips then the difference between it and the bundled software should be small. But if for some reason there's no going to DS between flips, the battery life will drop significantly and back to time units per charge.
We take the modern internet speeds for granted, at that time the tech behind Skype was top notch and probably when Skype made its way into Windows, that looked like the original destination. But later many questionable decisions made things worse even before the internet became faster and other voice technologies were up to the task. One of them was changing the protocol that made many headsets bricked. Probably from the marketing point of view it was a "if one wants Skype, he or she would buy Windows" step, but obviously it was not
Unless they also make the patents coming with it public, it would be of no use. As long as I remember, Microsoft has no good history of purchasing patents and making them public. Google has (vp7,vp8, Webm as an evolution). But probably some of the patents are expired anyway.
I also doubt that they would open source the ported version (c# I suppose) but they could do this to the previous Delphi version, they developed many useful components so even without the core functionality, UI code might be useful for Lazarus/Delphi developers
An interesting discussion between two good authors. Regarding the usage comments in code I'm sure the topic is hot and guaranteed to produce tons of comments on HN. Personally, I would comment more, but it needs time, motivation and thinking about my future self or anyone who will read my code.
An idea came that one might merge logging and commenting in a way when a logger (may be a separate one, not the one for debugging) should be used with bigger sentences that serves two goals - it is still allows using quasi-comments visible in the code and also allows producing docs+data when outputted when the program is run. The latter also might be an additional motivation since it's extending human understanding by producing "literal comment/data" streams.
"I envy him" - when I read at HN in comments when people discuss something I don't know, I envy them. I suspect that OP is also more about this (reading from his frustration), but "knowing something" doesn't always mean "have to work with something". Not knowing even at least vaguely that XML is something related to HTML is a flag of the lack of interest in the sphere you're working in
I see no issue in setting it dynamically from the code, supposed to be set during start-up code. A global variable representing a function pointer and this variable either changed several times (if the latest addition allows execution of previous routines then setting its new value) or just once for the latest version if the logic somehow disables previous start-up routines. Can be called "dynamic inheritance"
Very interesting also as a thought-provoking idea. For example
- It would be less challenging if function pointers variables are used instead of function. In this case, the code appended later may override the function variables it needs to fix/change
- Since all the code is there, it is possible to invent some convention to compile/run previous versions without CVS machinery
On a little technical side, the cross-platform engine of KOReader is developed in Lua and low level stuff like drawing and eInk specifics is different for every device. I think this separation might be very noticable for anyone running it on Android, almost no standard UI elements, dialogs, etc. But that's what makes this project so widespread device wise I think
It's interesting. Recently I came up with a question that I posted to different LLMs with different results. It's about the ratio between GDP (PPP adjusted) to general GDP. ChatGPT was good, but because it found a dedicated web page exactly with this data and comparison so just rephrased the answer. General perplexity.ai when asked hallucinated significantly showing Luxemburg as the leader and pointing to some random gdp-related resources. But this kind of perplexity gave a very good "research" on a prompt "I would like to research countries about the ratio between GDP adjusted to purchasing power and the universal GDP. Please, show the top ones and look for other regularities". Took about 3 minutes
I think that sometimes making drastic UI changes like MS Office did is one of negative incentives of commercial software. They always need something visually noticeable to justify added value of new versions.
I think the project is great, but for me it is connected now to an empiric rule of big complex projects with a large user base: An issue will be fixed more or less quickly if it affects millions/thousands of users, otherwise it will live forever in an eternal priority queue. There's an issue I noticed in LibreOffice, found the corresponding tracker post, confirmed it with a test, and now it is only visited by a robot checking whether stall issues are worth deleting. Another example is Chromium, where I posted an issue about wrong rendering of some elements with rare sets of attributes. The post is more alive than the LibreOffice one, but nevertheless, never fixed since it requires precious developers time but too niche to gain much attention
Good point. LLMs can be treated as "theories" and then they definitely meet falsifiability [1] allowing researchers finding "black swans" for years to come. Theories in this case can be different. But if the theory is of logical or symbolic solver then Wolfram's Mathematica may be struggle with understanding the human language as an input, but when evaluating the results, well, I think Stephen (Wolfram) can sleep soundly, at least for now
Why can't other AI tasks be also in focus? Like having fun with some creativity. I remember that I once created a story with an LLM when we switched sides. I still remember some of the twists and it's interesting that I no longer can attribute them to either of us :)
It's ironic that progress that freed many people from hard agricultural jobs and allowed society to free part of the population from daytime jobs was mostly directed at non-child-bearing categories like children and teens (to learn and study for future jobs) and elderly (to retire earlier than in previous centuries). The age ideal for raising children is also ideal for economics to be a part of it
My first (almost) thought was "Ok, looking at the llama, where's the quality boost?" I think the explanation is probably in how LLM are trained. Even without knowing deeply the internals, I suspect that it's the compression of information so to simplify you can't make 8GB data contain all the facts of a "bigger" normalized relational database. So they keep the facts present everywhere and often drop rare facts. For example, a fact "SQL was invented at IBM", this fact can be found everywhere, in books, web sites, comments. You don't need access to copyrighted books to acquire this fact. But a first-person account of someone who worked at IBM at that time is probably can be found in a couple of books, but due to "compression", it will be gone anyway
The article lacks references, especially about the law that enforced banks to make loans on non-market conditions. Even if the sources are in Russian, it would be better than no sources at all