This page is not created by, affiliated with, or supported by Slack Technologies, Inc.
2023-10-05
Channels
- # announcements (19)
- # babashka (28)
- # beginners (62)
- # biff (3)
- # calva (19)
- # cider (24)
- # clj-kondo (8)
- # cljdoc (15)
- # clojure (32)
- # clojure-europe (16)
- # clojure-nl (1)
- # clojure-norway (17)
- # clojure-uk (8)
- # clojuredesign-podcast (26)
- # cursive (64)
- # datomic (43)
- # deps-new (1)
- # fulcro (4)
- # honeysql (1)
- # hyperfiddle (46)
- # kaocha (16)
- # lsp (15)
- # missionary (51)
- # music (1)
- # nbb (4)
- # off-topic (55)
- # pedestal (11)
- # podcasts-discuss (1)
- # polylith (7)
- # practicalli (1)
- # releases (4)
- # shadow-cljs (120)
- # tools-build (34)
- # vscode (1)
- # xtdb (2)
BTW, I am quite amazed on how we usually discuss AI things, the idea of copyright, legality of the stuff, etc... and here I am unable to use a music that was recorded 90 years ago because it's still copyrighted 😢
Yesterday we had a discussion with a couple of friends and I remembered one important thing. If you're losing then you want to steal. If you're winning then you want everybody to respect each other and behave nicely.
Basically when you have amassed something then you want to prevent other's productive behavior so that your resource would be safe. And then you get this 90y copyright that screws behavior from people like you 🙂
If you're happy with status quo then you create additional special cases for yourself.
Or to put more bluntly:
The more if
statements you have, the more bugs can be found (you keep whole source code and only add stuff).
Or even more bluntly:
The more code you have the more bugs you may have.
I don't have knowledge about llm, so this question may sound stupid. I do my day job with handling some runbook scenarios. However, I don't find which one and the internal llm search engine also get polluted by noise created by our own, like designed docs or product docs. I am thinking if keep a new mindset like how to write my doc into a more searchable way like doing seo can help better search the knowledge.
I’m a bit uncertain how to parse your query, but finding a document sounds like a job for a search engine, not an LLM spontaneously. When an LLM is involved, it’s most likely being fed the results of a search query anyway, since they all have severely limited context windows. It might then select among those search results for you, but it might be the underlying search engine that’s disappointing you, not the LLM itself. How to optimize your document for findability would depend on the search engine.
Likely your internal LLM search uses Vector Search Database - which are pretty bad for many usescases. Unlike SEO it is very hard to optimise your documetns to rank well in a vector search result. I would just switch back to a normal search engine for documents or git repo wiki or folder structure etc. LLM themselves are incredibly good at reading documents without tables/diagrams/images - so if a person can read it then then the LLM should be able to too.
We use glean and I got the feedback that the indexing works like Google seo, I need to do proper links and clean our docs, didn't explore further details.
In case anyone hasn’t come across this old comment on code that reads photoshop PSD file format [https://github.com/gco/xee/blob/4fa3a6d609dd72b8493e52a68f316f7a02903276/XeePhotoshopLoader.m#L108-L136]
This is marvelous, I've never come across this one before 😆
I am not attempting to excuse any such things, but having worked in a large company on projects where in order to make any kind of progress, people who have no idea that the other one exists make independent decisions in order to move forward, it is extremely easy to understand why such things happen.
The author of those comments is eloquent, I must admit. 🙂
File format backwards compatibility for decades isn’t an easy task
Having worked at Adobe, I sympathize with the author's sentiment 🙂
It is said that software companies bake their org chart into their products. I suppose the longer a product exists, it also has all the re-org charts get baked into it.
Someone even gave it a name: Conway's Law. It is a real thing for some software, for sure.
And not just software, I am certain.
Certainly, although I feel software companies are more prone to it because they make thought-stuff, which is literally people's thoughts and ideas on replay.
I can't think of any reason why other large design/development projects like civil engineering, mechanical engineering, etc. would not be prone to the same results.
Hillel Wayne wrote an interesting series of articles about the claimed huge differences between software development and other kinds of engineering, and found evidence that there might be fewer differences than those who are only experts in software development, and not any engineering profession, might think: https://www.hillelwayne.com/post/are-we-really-engineers/
I believe he was motivated specifically by statements that were popularized by folks like Paul Graham, claiming that software isn't like other kinds of engineering.
There are some constraints to software development that take the place of physics and gravity, I believe. Not the same constraints certainly, but people tend not to create their own new programming languages or CPU instruction set architectures very often.
I like that post! I grew up in a mechanical engineering / systems thinking household and find endless amusement in the special-snowflak-y / NIH belief system that is rife in the software industry.
(or new database systems, or new replacements for React/etc. (although that last one seems to be created anew more often than my other examples))
Also I understand large software projects are at least as capital intensive as building an engine and all the mass-production stuff around it.
Yeah, 300-500 full time people for a handful of N years software projects do exist, and a few of those don't even collapse under their own weight 🙂
I was thinking of projects developing proprietary software at companies where I have worked.
> There are some constraints to software development that take the place of physics and gravity, I believe. Yes, I think these are, generally speaking, economic and behavioural systems. Feedback loops, buffers, supply/demand push/pull etc. The free-floating currency system would be a close analogue. It has no direct material or physical basis, yet it has self-regulating (and disregulating) properties as a system.
> people tend not to create their own new programming languages or CPU instruction set architectures very often This is a good point. And I think it stands in contrast with other engineering fields, where it is par for course to invent new tools to make the thing one wants to make. Even little machine shops will produce their own jigs and fixtures and shop floor design etc.
And vertically integrated manufacturing companies like Apple are known to redesign their entire production line and toolchain from scratch for new major versions of their hardware.
As well as to make all new versions backwards-incompatible with all past versions. Just to tie this side-track back to the original post 😀
Today I talked with a soon-to-graduate developer (third year of studies) and I told him that "I'm unemployed for more than 6 months because I chose a programming language that I like but one which doesn't have jobs". I also told him that it's a great programming language but he should just stick to what he learns in uni and so on. And to get a programming job he has to do a project that he could show in his job. He said he'll do a face recognition thing for his lecturer. He was interested to hear about the syntax but I no longer think that Clojure is a good way to get money. 6 months of unemployment means that the next salary has to be way higher than even Stackoverflow survey tells. So it doesn't work.
What other programming languages are you fluent in?
TS, JS, Golang, then maybe even some Java 8 and older as I used to work with these. Basic ones. I was doing some Python recently as well. I was writing a small thing in it. I do have a pretty good CV. But that doesn't do much. There are just too little job interviews.
OK, and you can't find work in any of those languages either? Is that a regional thing (you're in... Eastern Europe?).
Happy to review your CV and provide feedback -- I've been a hiring manager for over 25 years so I see a lot of resumes/CVs...
I was writing to everybody at some point. They just decline without any interview.
"Unfortunately". I received this quite a bit.
I would take @U04V70XH6 offer to review your CV. your issue doesn't seem like a language problem. where you're looking as well matter
(he has -- and it's an offer I'm happy to extend to other folks who want feedback on their resumes/CVs, esp. if they're struggling to get companies interested... as I said to Martynas via DM, I've probably reviewed ~200 resumes/CVs this year, most of which came from a JS req we had open back in January!)

Let's see how the story ends - it's not written yet! At times it might seem like everyone else is happy having a TS/Python/Golang party - in reality those have 100x more competion per job post - quite possibly many of them AI-exaggerated, which makes things worse for everyone involved. You can see how the job count has evolved in e.g. https://hnhired.fly.dev/ . I found it quite surprising that Sep had fewer posts than Aug.
Whoa. Mac Sonoma Personal Voice, recorded as QuickTime audio with condenser mic two inches above Mac Airbook speakers. And a fun correction: test playbacks played in QT it sounded choppy and warbly, so that is what I recorded for fun. Exported it does not sound bad at all. :thinking_face:
"Ledgers Maximo"? I had "Mac Sonoma"....
FYI, "Personal Voice" analyzes 150 samples to fake my voice. I guess this is Vision Pro stuff arriving early? 🤷
Scary. The real deal, for comparison:
Pretty good! The one thing that's still off-putting for me, in general, with text to speech is it doesn't seem to understand pauses for effect or mood properly. It's a bit hard for me to follow when it just bulldozes through text.
I noticed that some of the training phrases called for emotion, such as surprise or doubt. But yeah, that "Hello Clojurians" came out pretty flat. Was I too inhibited in the training though? I did notice that I was enunciating artificially well. When I have the energy I will do another trainimg.