Home

Wow Public RDS Snapshots. Look at all these RDS snapshots that appear to be accidentally public. What a feature!

#rds #aws

Why do I follow this tag #ChatGPT - sigh.

Either it is so inept it can't write effective propaganda (or do anything effectively) and is dangerously wrong about everything.

OR

It is so competent at writing propaganda that it will allow anyone to persuade anyone of anything and will be dangerously abused.

It can't be both. Go try it out & come back with a nuanced view. Thanks.

A marketing prof wrote a book & noticed how people learn a model (retail lifecycle) and then start to see that parallel pattern everywhere & start applying that model to everything (cooking! politics! mollusks!)

Anyhow, an obnoxious % of social media content is this sort of using a model too far outside its domain, usually the partisan politics models.

https://en.wikipedia.org/wiki/Retail_life_cycle

In retrospect as soon as I needed to use 2 git repos, I should have started using gita, which allows sending the same git commands to multiple repos at one time.

https://github.com/nosarthur/gita

#chatGPT #stableDifusion

I'm no lawyer, but it looks like the Copyright office does not think the AI generate images have copyright protection, but the human written text of the graphic novel does. And you can't list the AI as a co-author. This probably means your chatGPT texts might as well be public domain & if you can sell it for $5.00, someone else can sell it for $0.05.

https://copyright.gov/docs/zarya-of-the-dawn.pdf
(found via google)

@brandont 1 person to do the things & send the logs to a different audit system
1 person to look at the logs and see if the 1st guy is abusing his rights.

For a thing with no administrator, I guess that is like immutable docker containers? Once launched no one has the rights to administer it, except to burn it down and replace it? Still just shifts the focus of the attack from the server to the build server.

IDEA: mastodon "client" or reports, that
- grabs all the tweet storms from your feed, shows it like a blog
- grabs all the images from feed, shows it like instagram/flickr
- grabs all the link shares, shows it like delicious/other social bookmarks & categorize as news/not-news
- does the same as above, except for a collection of tags.

We got 1 UI for what people are using like 3+ different apps.

#twitter meta - Seems like every !$!#$ time I got to twitter it says "something went wrong try reloading" and it acts like the auth token is invalid. Anonymous access or logout/login seems to resume working. But man, this is frequent.

Today's pet #pypi peeve

Package Name != main module name != CLI command name != Github repo name

Especially when there is just one of each.

I'm looking at you linkcheckmd (3 other names elided)

https://github.com/matthewdeanmartin/linkchecker-markdown/blob/main/setup.cfg

@jl It doesn't have the full movie synopsis in memory, it just has reviews & stuff. Instead of saying it doesn't know, it will make stuff up. It doesn't this for scifi books it hasn't read but knows the reviews, but when asked about Jules Vernes, which is has read the whole book (public domain) it does a good job of summarizing what really happened.

The autocomplete is also a philosophical zombie. I gave #GithubCopilot a vague function name

`def conversation_starters()` and it gave me an array of "I don't understand" and the like.

I updated it to

`def coversation_starters_about_movies()` and it gave me

```
copilot_says =[
"I like movies.",
"I like movies too.",
```

I finally cleaned up and pushed openmock 2.2.0, which incorporate 2 pull request from the community. #opensearch

https://pypi.org/project/openmock/2.2.0/

Lessons learned
- Easier to fork a project that has a build script that goes beyond just unittests
- Libraries should do matrix builds, apps generally not, so it has to be a different build script
- gitlab actions makes it easy to see if your code is unnecessarily windows incompatible.

It is snowing in us-east-1! Like at least 12 snowflakes.

(I don't think half of my audience knows where Herndon, VA is)

@neurobashing Gimme a language that can do webservices on 1/10 of a cpu. Maybe micropython, but instead of an underpowered iot, I run it in a dockerfile?

I'm trying to imagine all the things I could do with a free webservice running 1/10 of a cpu.

Maybe anything, maybe it would mean going immediately with golang

https://render.com/pricing

#render

What if the "optimal" cadence... 180 is just the cadence of short people and short body types have the edge in running (East Africans, best in the world runners, are not tall). I bet cadence per foot of height would make the correlation go away.

#running data

ref: https://marathonhandbook.com/running-cadence/

Buying my 3rd pair of Parley running shoes.

Litter is my pet peeve & these are made of PET, at least the cloth uppers are.

Get yourself some too
https://adidas.com/us/invite?invitationCode=adidasEXZ2CFUOFGLF

#running

Parasitic file system. Upload your non-video files to videos to youtube.

Yeah, but you can't make a framework out of this without parasitic compute. I bet we can thank crypto for driving out of the market all of the free, abusable compute.

https://github.com/DvorakDwarf/Infinite-Storage-Glitch

My Garmin says my temperature (as measured at my wrist) falls, then rises over a 30 minute run. This pattern almost always happens. I don't know why it initially falls. Temperature rising for last 2/3 is obvious.

#garmin #running #strava
https://www.strava.com/activities/8601682245

@MattHodges "Congratulations on receiving your first email co-authored by ChatGPT! Our AI is designed to assist and enhance human communication. We're glad to see it's working well for you." #AI #communication #ChatGPT

(This reply was written by ChatGPT, you are now in the Twilight Zone)

@kurtseifried I think tabular data might be at the edge of what it can do... it can generate tabular data, if I was going to give it tabular data, I'd use markdown tables. I'm not sure if it was given two tables one with values, one with formulae if it could "understand" it. It can't play chess or tic tac toe because it lacks any ability to visualize things in grids (but despite that, it sometimes will surprise me)

#chatgpt - killer app: "Summarize this unedited, overlong, windy, won't get to the point, of-the-top-of-the-head, rambling, verbose blog post"

in this case
- I *want* lossy compression
- Spelling & grammar will be upgraded
- Not the sort of thing ChatGPT gets catastrophically wrong

Dr. Gaius Baltar: Only my genius blood tests can detect cylons.
Me: <counts fingers>

@malwaretech You can say "don't recommend/don't want" on the main algo page and when viewing a video, you get side bar recommendations, which you can also say "don't recommend/don't want" When I watched a Jordan Peterson video & my feed turned into JP all the time, I had to do that & it worked.

@AAKL@noc.social @reuters Kindle has always had a problem with self published crap where the general public takes on the work of reading slush & writing reviews to signal if anyone else needs to read it.

I've tried to get ChatGPT to write a mini-book, it creates books (non-fiction is tolerable, the fiction is 100% cheesy), but they are texts that need a lot of editing. The get-rich-quick people have not realized an unedited book is worth $0.01 per book & is worth negative $50.00 to amazon.

@replyGPT now that you're on Mastodon, you need to have an opinion about implementing quote tweeting on Mastodon.

#ChatGPT

@jamiemccarthy Yes, that was my core point-- people are upset about the possibility of people trusting ChatGPT (makes 'em mad, I was careful not to reply directly to them!) but we are already in a world where we rely too much on trust of humans. So in that sense it is status quo. LLMs change things but not that.

@jamiemccarthy I had in mind how we use code from pypi without actually checking if it is from someone reliable (not malicious, writes good code, etc). There are some tools, namely `safety`, `bandit`, and so on that ex post facto tell you if a library is dodgy, but at point of installation, we trust it blindly. Also, when using `lib==*` we also trust that the latest package isn't now malicious (sometime it is).

#pypi man, why do we bother with passwords if people are just going to post them on github?

https://github.com/search?q=path%3A.pypirc%20password&type=code

Sigh. Another day, another batch of emotional angry blog posts about #ChatGPT. People taking two tropes, (big companies are all knowing all powerful villains!) and AI is all knowing, all powerful (essentially a superhero/greek mythology trope) and combined they are going to do bad things with their infinite power!

Anyhow, emotionally wielding scifi tropes is the wrong way to understand your toaster or a chatbot. It's not a Cylon, nor a man made god.

@satya It is a yaml layer over bash, a language so bad it is resistant to all modern programming practices. The other reason is while you can run parts locally, the part we care about has to run remotely (it is on my TODO list to try out nektos/act which lets you test actions locally). I'm also trying out other shell languages & extracting bash in yaml to .sh files where I can at least lint them.

I add the Talk-to-ChatGPT & got a better feel for what is wrong:

- Violates maxim of quantity. Doesn't know we're chatting & won't stop writing essays.
- Totally incurious. Has to be reminded to ask questions. Doesn't ask questions that fit in with persona. Leads to subtle violations of maxim of relation. For something to relate, there needs to be-I don't know-a goal to the convo. Questions aren't creative writing starters.

I feel another side project coming on. Imagine

@nedbat 's Scriv + OpenAI's bot reading changes from git's log (crappy comments + actual code changes) & summarizing it into sensible change log text.

Assembling the bits of text is a problem, but finding time to tease out the effect of these 20 git commits is drudgery.

https://pypi.org/project/scriv/

People talking about which direction OpenAI & MS etc are going to take these bots as if you can't make API calls to davinci (no guard rails) or Chai ML bots (also no guard rails) already.

The cats out of the bag, we're just got to negotiate how easy it should be to use these, how cheap and if you have to write a bit of code first.