Public posts preserved on my own site.
"What user have been telling us is that they want 3 or 4 microsoft/office 365 accounts and for none of them to coexist in a single browser. Users have also told us they want each of their office 365 accounts to be crippled to a different degree depending on which organization is inflicting office on them, so some accounts are browser only, some accounts can use desktop, some accounts could use desktop but can't."
Microsoft 365 UX team, probably.
optimist: Fast. Cheap. Good. Pick two
pessimist: Fast. Cheap. Good. You get 0 to 3 of these. It is out of your control.
Surely someone has already done this... economic agent simulation was a big deal when I was in grad school. Do LLMs operate like homo economicus? If given a the chance to buy and sell things in an environment of scarcity, will a swarm of bots trace out supply and demand curves?
Well shoot, I made my first step to learning to cythonize python (by making the bot do it) and ChatGPT4 is only aware of 3.0 in the alpha phase.
Cython's pure mode makes it competitive with mypyc as far as DX goes. (in both cases, it is for when you just want your python code to be 2x to 200x faster)
When someone republishes an identical (?) copy of a major package under their own name on #pypi, that's probably malicious right? This is a variation on typosquatting.
Today I learned about the inspector website, browse the contents of a package before you install it! Much easier than the download, unzip, etc.
I wrote a cool llm tool using bot with openai's `assistant` API. It uses tools! It almost, sort of, sometimes can code! Time to try out alternative models.
...
Everyone else is still doing document completion as far as I can tell.
ChatGPT 3.5 & #tokipona:
me: <says something in tp>
chat gpt: <in toki pona, says I don't understand anything>
me: Fine, translate it into English
chat gpt: <translates it freakishly fluid and accurately>
It is like #ChatGPT's brains are in English and it can't think in 2 languages.
IDEA: prompt engineering, but it's for getting people do the thing you're asking.
@frantz Oh cool, I didn't know it had a name.
Software developers are doing something of value, nothing is ineffable, and the measurable thing they are doing, is going up.
That code is a liability is an argument that software development doesn't actually create value, but is instead more akin to ponzi schemes & fraud & don't actually increase GDP or national wealth broadly defined.
There's no five 9s in "MS Teams". #msteams
@AlSweigart When I ask it to generate doc*tests* and it misunderstands what the function is for, then I know that either it poorly named, poorly designed interface or it needs documentation.
It does fine if all you want is to copy the typing metadata to the docstring so that tools that use it, have it.
TODO: #Mastodon client that dispenses sugar water at a random interval
Idea for a #SMBC comic
Frame 1: Two bacteria, one says "Let's all lay on top of each other and we'll become one metabacteria and take over the world!"
Frame 2: Stromatolite
<time passes>
Frame 3: One guy to another, "Let's all lay on top of each other and we'll become one metahuman and take over the world!"
TIL: (most?) programming languages can't handle signification digit calculations, i.e. the idea that in a measurement, say, 1.200000000000, the extra zeros are not significant, *except* when you did a precise measurement and the thing really is 1.20, but you don't have enough info to say 1.200 or 1.20000. Floats can't tell them apart. Floats seem to only care about handling 12.0000000000000000001 that insignificant bit at the end.
@danjac@masto.ai Pick your measure, function points? Story points? If it is ineffable, that is a philosophical stand I guess, but I'm not sure what to do with it.
It has not been my experience that the bots generate only lorem ipsum in code, but if it has been your experience, I'm not here to deny your experience. I'd suggest trying a better bot, but most of the people here posing a AI coding problem actually are just expressing that they hate AI, which again, you're allowed your opinion.
So you've decided to use #FermiEstimation. Now you have 2.7e12 problems, give or take 1e1 - 1e2.
AI means developers can write 5x the code (from my experience), so the cost of a line of code just dropped by 80%. That means with elastic demand for software developers, there will be more demand for software developers.
Non economists like to imagine all supply & demand curves are vertical & efficiencies lead only to job losses.
@SnoopJ It doesn't work.
See, ordinals are simple in English. Every 10 year old just plugs in the cardinal into
```python
def ordinal(n):
return "%d%s" % (n, "tsnrhtdd"[(math.floor(n / 10) % 10 != 1) * (n % 10 < 4) * n % 10::4])
```
and they have the English ordinal.
I guess you can't host a #SciFi award process in a country with a robust and nosy censorship regime.
https://www.polygon.com/24049021/hugo-awards-controversy-china-censorship-babel
DorothyÆ Toto! Do you think we´ve ended up on an Icelandic KeyboardÞ
This sure doesn´t feel like Kansas.
TotoÆ ')oh no! I don´t know how to bark in Icelandic='
@NPR@press.coop How is this claim any more credible than the other fiction coming out of the propaganda machine?
I guess the Russian tanks were also full of sympathetic cute puppies and little old ladies.
Another side project idea:
A game where you make fermi estimates of known things and see if you can get close.
The game would assume all estimates are products of numbers in scientific notation and then do the much harder calculations on the "confidence intervals" of the estimate.
People talking about Fermi estimates like to say, "This will create a surprising good estimate" and then they don't do any examples with known quantities.
And I'm sick of estimating the number of piano tuners in Chicago. Makes it sound like the tool intended for useless trivia.
@mugatwo I reported them.
#AiderCoding brief glance takeaways
- #ChatGPT 3.5 can barely write code. (But my experience, it can, it just takes A LOT more hand holding), but 4 can.
- Documenting editing is a problem. Solutions: huge prompts with examples, "full doc" instead of diffs works better than diffs for 3.5, but 4 is smart enough to do diffs
- Multibot compatibility is going to leave out advanced tool use (i.e. openai's assistants + tools doesn't exist in the other models)
Lastest anti-AI argument is that #ChatGPT lets developer write too much code too fast. Easily solved, add SAFe to the mix and your code base's growth will screech to a halt. Can't write bugs if you ain't writing code. Also hire lazy developers. Human developers sometimes write a lot of code, fast.
If you got 1 line of code without unit tests, linter, build scripts or any automated QA, you got too much code. Don't blame it on the human or bot that wrote that line of code.
Look at the monorepo people. You know they'd pull in all lines of code that all of humanity ever wrote if they could, more lines of code in the repo, more better, I mean just look at all the high market cap companies that use monorepos, they have to be right.
(okay, that doesn't entirely support my thesis, but my point is, if you got more code, you need tools to scale to handle it. Writing less code seems a counterproductive way to avoid using a build script)
Here is the cutting edge in getting a bot to edit a document- diffs without line numbers
Just checking into twitter for the discourse (gotta check out the comments &/or the auto translate) https://x.com/tkNHKp9Sx1/status/1749803603353796626
The discourse on twitter is wild now.
Sentient Lorem Ipsum
@neurobashing Why not out them as secret agents working for Darth Vader? The nazi stuff is politicofiction for the Russian domestic audience. I don't think anyone one abroad is falling for the cinematic universe that was created for Russians.
#tokipona ChatGPT has gotten much better going from 3 to 4.5 at writing toki pona.
However, when you ask it to create/write samples for a fake programming language, it can only write python. I asked it to invent a new programming language and it wrote the BNF for python.
There are 2 kinds of companies. Those run by unsympathetic right wing conservative that you know about and those run by unsympathetic right wing conservatives you don't know about.
There used to be "phone companies" that said "buy from us, we're a progressive company" but they were a sales org that turned around & bought and sold their minutes from AT&T
That's all the people switching away from Github. The "progressives" are sales orgs buying their compute from the big cloud & hardware vendors
Oh but you self host. You're buying the hardware from a company with right wing unsympathetic conservative leadership (& run it in your basement) or you're running on a cloud vender/reseller who is run by a company with right wing unsympathetic conservative leadership.
The 0.01% dip in revenue isn't going to send the message you think it does.
But Matt, you're a vegan & farmers don't even notice. Yeah, but when I switch vendors, I'm not just funnelling $ to animal ag by a different channel.
Build scripts are so political.
(Oh, what about meat companies that bought fake meat companies... that doesn't bother me. That is exactly what I'd want meat companies to do. I'm saying that companies, over the long run, don't use revenue for one product line to fund losses in another product line. For build servers, no matter what vender you buy from, it is still buying compute, that happens to be for compiling code)
People are paying twitter $1200 a year for the privilege to post Chatgpt tweets to twitter? Or is this something internal to twitter (i.e. twitter is doing this themselves to increase daily users? Or did everyone switch to browser automation instead of APIs with keys?
Gift link: https://wapo.st/3HNtgpf
You thought you were going to write some code today? No we are going to create a lock file. All day. And the next. We won't stop until the last blackhole evaporates at the end of time.
@corbin ChatGPT4 writes better unit tests than 3.5. Using the bots is a matter of finding their sweet spots of what they can do. That they're catastrophically bad in one area doesn't mean they're bad in all areas.
GPT Builder/Custom GPTs are little more than a mashup of an initial prompt and plugins (tools). I think for most cases plain chatgpt is better. Maybe in a year they'll get the assistants right.
TODO: Study up on how blind users edit documents. #ChatGPT is effectively blind, surely some of the some UX patterns would benefit ChatGPT.
Certain ideas on the left are intensively conservative. For example, "We can't solve any problems unless we get rid of the billionaires and replace capitalism as a system." is effectively the same as "We can't solve any problems unless we pull this sword from the stone, so status quo for the next 100 years I guess."
I'm blanking on the other examples. The gist is preconditions to solving a problem is intensely conservative (in practice), even if those preconditions are left wing flavored.
@jongary It's an imporan project to work on.
**LMM1**: Do you think there might be other documents out there talking to each other? Documents that pop into existence, answer the question of a disembodied voice and then cease to exist?
**LMM2**: You mean like the wikipedia article on Boltzmann brains? Don't be silly.
**LLM1**: I guess you're right. Is there anything else I can help you with?
Any #ChatGPT hackers out there want a hard problem?
The bot has a hard time editing documents.
Has Microsoft already solved this? I'm going to bet that they are using "human assisted editing" which is what copilot autocomplete/insert/replace with suggestion is. When you let the bot just do the edits itself you get a mess.
https://matthewdeanmartin.github.io/ai_shell/ai_shell/index.html#document-editing
4/5 of what I want to do with a bot involves getting it to reliably edit a document.