harperlee

Have you pointed another agent at it, so see what does it think about the environment? It could be used as a honeypot of sorts, seeing that agents are increasingly being trained to be relentless, even when faced with impossible tasks…

As an english-as-a-second-language speaker, I learnt quite late that the original meaning of drone is a male bee, and that the modern drone is called like that because it is a small flying thing that buzzes with their propellers. Perhaps that’s useful for others.

Awesome way to show you’re not an LLM, by the way.

Arguably studying derivative works, such as a volume on economics, would be sufficiently acceptable as to void your calculation…

“Quantity has a quality in itself”. How many books can you read in your life, and how faithfully can you transcribe them, vs. just retaining a fraction of them? What about LLMs?

FWIW, I'm on the email list of Hillel, and when discussing that the book is done, he said the following, which makes me thing otherwise:

    This marks the completion of a project that took five years of work, six bookwriting professionals, fourteen domain experts, and fifteen public alphas.

    This has been, without a doubt, the biggest and most exhausting project I've ever done. The examples in the discarded drafts alone could make a second book. The cursed knowledge I've gained on LaTeX and typography could fill a third (or at least a couple of entertaining blog posts). Self-publishing was simultaneously the worst and best decision I made.

    Now excuse me I am going to sleep for a month.

There is some research around minikanren that touches exactly that (minikanren is akin to prolog written in scheme, where order is just an implementation detail, and cuts and other non-relational things are looked with suspicion).

Would be interesting to see a movement akin to Asahi / OpenWRT that enables me to run my side project on my fridge or tv!

I'm sure the general population being asked whether they read books have a very clear idea of what a book is.

I don't know a lot about formal verification, but:

> So you can formally prove that your e-commerce refund management logic is correct, except for proving that you /processed the refund/. You can't even prove anything about recording the refund in your database, say nothing of proving anything about your interactions with your payment processor.

You could say the same thing about the viability of functional programming on a CRUD webapp, but languages like clojure have been used to great effect here. The fact that thera are important, even fundamental, bits that you cannot verify, doesn't take out value from the fact that you can eliminate whole dimensions of issues.

I won't hesitate to give this to my children, in a year or two when they are reading more fluently and don't need lots of pictures.

Price is also a factor here!

It was posted at 5am in New York... not sure that that was a US view, so the fact that the platform is US-owned doesn't seem so relevant, if there's a global audience.

That being said, I do agree it is a legit thought (and moreso, completely on point in the subthread discussing downsides), and that it shouldn't be downvoted.

"Marjane Satrapi died of sadness a little over a year after the death of Mattias Ripa, her husband and the love of her life"

The simple humanity in this candid description brought a small tear to my eyes. I'd say that the classical approach to this is a dry, clinical description of a depression stage, or a description of a how and not a why. Very welcomed in the age of AI slop!

More strongly, if the pattern matching of a phenomenon totally / perfectly models the phenomenon, and you end up with a perfect model of the phenomenon, that enables you do do causal prediction, how can you NOT call it understanding? What more is there?

Don't know about your landlord specifically, and I'm no landlord, but there's also a bunch of people (including homeowners) that will wait until summer starts to test and then complain about HVAC... just on peak season for HVAC maintenance, where the waiting times are long (and the price will probably be higher).

Then use them! Then those projects will cease having a single contributor or being unmaintained. This doesn't need to be a binary decision (either there is no risk or I won't use it), just choose the project / scope with knowledge that there is some risk in a small community; so internal use rather than client facing, specialized uses, etc.

APL was designed to be written on a chalkboard (if I remember the story right). It is quite dense, and programs are quite small. Reading is slow and requires you to ponder about what was written. You can hold a lot of content in a small amount of 'ink'.

Now, an idea: HN is always complaining that an ipad (or any other tablet) is a consumption device, as it is not designed to be used with keyboard/mouse. Do any of you know if there is an app where you can write APL with a stylus, and has the ability to evaluate expression on the fly, similar to a repl? That would be an awesome thing to do.

Depends on the test design; is an agent competing against other agent in a given match, or against a test? Plus! Does the test's ELO fluctuate?

Hopefully this becomes a reversal in the trend of giving less and less context to the user.

I'm not against the considerations of the article regarding the user and its state of mind, but please do add as much technical detail as possible!

Even if an error message is a cryptic error code, that's better than a "Something went wrong" message. This is not better, or even friendlier, UX. An error code can be referenced, can be searched on the internet, can be passed around on a ticket or on a call... add parameters to your error template, reference the name of the file, the item name that does not respond, the HTTP error code... just give the user some transparency, some agency. Help the client build up a mental model of the error: when / how / why might it be happening.

I'd say Java, because it has a massive footprint amenable for training, and a strong type system (does not have sum types though and those are trendy).

You'd have to steer the LLM to use the style you want, and not massively overarchitect things though, but that's going to be an issue nonetheless.

Would be cool to have a plugin that shows % of bot per user, based on their history of comments.

> It's already demeaning to expect them to "learn an accent"

The concept of an accent is broad, but at least part of it you need to learn together with the language, as speaking a non-native language with a thick accent is partly based on the fact that you have yet to learn.

Without being exhaustive, things that might fall into the "speaks with an accent" concept in this thread:

   - Prosody. Prosody can vary per region but a distinctly alien prosody to a language is a barrier for the receptor of the message, that expects a given language and a range of prosodies. E.g. as I know french quite well, hearing english with a heavy french accent makes my brain try to understand what's being said as said in french, and interferes a lot.

   - Sound shifts for particular phonemes. While some of it might be local to the language in certain registers (idea --> /ide"er"/, three --> /free/), others are clearly issues in the target language pronunciation (eg. japanese people having trouble with the l phoneme, spanish people adding an /e/ sound prior to an s-mobile, or v versus b for spanish people also).

   - Connected speech. Where do you end words, how do you omit sounds, etc. Also massive hindrance to understanding.

   - Grammar. Alien grammar is a hindrance to communication. You need to learn that.

Interesting! Could you expand?

There is a lot of hate in the comments but there is some merit to the post existing:

  1. Even if the task is unreasonable, it is good to showcase that the LLM will perform poorly - warning not to be used for diabetes.

  2. As it is a probabilistic model, the approach was to execute it multiple times and look at the distribution. They also tried to minimize variance: "All at the lowest randomness setting these models offer.", the post mentions. Yet the variance of the responses is surprising.

  3. A multimodal LLM should be in general able to discriminate between crema catalana and a cheese sandwich, and provide a textual, uncalculated range of how much calories the item has (internet is full with tables for calorie counting and things such as this https://fitia.app/calories-nutritional-information/cheese-sandwich-1205647).

  4. It is not clear that the "expose" surprised / outraged style is just a communication vehicle or if the author really thought that e.g. LLMs could be hypothetically able to provide confidence estimates.