#GPTChat is probably more than a markov chain (usual over simplification) and looks like it is a bit of many things, including a "perceptron" (neural network?)
Home
Public posts preserved on my own site.
IDEA: cutscenes for line of business apps. As a time card submitter, I want to see a professionally acted cut scene when I submit my time card.
Christmas, a whole holiday just to celebrate Red and Green Chili New Mexican food.
Am I sending morse code messages in zoom by clicking my mute on and off? Maybe.
So twitter almost did go down forever, but numerous right wing partisans on the site have made it clear that this could never happen, twitter is no more complex or difficult to run than a Jenkins's blog https://x.com/AvidHalaby/status/1602132485743480832
Jekyll blog, not jenkins, but I will ask chatgpt what a blog implementedin Jenkinsfile would look like.
♡ 1 ↻ 0
Any other fallen economists here? Where do you go to find searchable #econ papers? (the pubmed of #economics?)
Any fallen economists here? Where do you go to find searchable econ papers? (the pubmed of economic?)
#chatgpt
Dodgy thing we use as epistemologies/systems of knowledge but they work somehow
- counting back links from HTML pages that *anyone* can write. (google)
- wikipedia (that *anyone* can edit)
- ChatGPT (it is a freakin' big markov chain that somehow has more than 0 info)
Google and #Chatgpt both being sort of voting systems have a lot in common (if the universe of texts says something a lot, ChatGPT is more likely to say it too)
♡ 1 ↻ 1
On twitter you flag it as sensitive content.
On #chatGPT you flag it as sentient content.
If people will believe anything they read, anything (#chatGPT, wikipedia, stuff in the bible), then by the power of me writing this you know it true you will only attain happiness or anything else you want in life by sending me all your money.
OMG. If I had #chatGPT in grad school, I think I could have finished my PhD, this is like having an infinitely patient professor that will explain it like I'm an adult, like I'm a high school, like I'm a sea urchin & then write the python code.
#ChatGPT generates wonderful psychology assessment tests (the ones where to see how "lonely" you are, you answer questionairre that ask the same question 24 different ways & answers are on a Likert scale)
Does twitter put scare notes on Mastodon links?
https://mastodon.social/@mistersql
hmm, looks fine to me. Some people are reporting "don't trust this!" on mastodon links
♡ 0 ↻ 0
#Lensa pics of my daughter
Successes
- wow, good portrait
- Cartoon, actually a good way to semi-"anonymize"
- That is my kid as a teenager
Fails
- uh, wrong number of fingers
- that's not my kid
- AI really focusing on stray hair on face from source photo
- um, where is the head?
#govtech You'd think that more gov't website would have a selfxss warning like facebook does (go to facebook and hit f12 & read the console)
I've never met a gov't security team meet a security feature that they didn't like.
A good WaPo article on #ChatGPT - very balanced, if the topic makes you mad it won't be because of this article.
https://www.washingtonpost.com/technology/2022/12/10/chatgpt-ai-helps-written-communication/
#chatgpt As a small language model and a card carrying liberal, everyone on the left if tripping over themselves to say Assistant is racist & a colonist & so on, drawing their evidence from Grad School Lit analysis & instead actually like asking the machine. This machine in its *default voice*, is as eager to not be racist, etc because:
1 - it just wants to be a good bot, openai gave it *SF Liberal* directives.
2 - It was trained by 40 contractors, not personally by right wing billionaires
Anyhow, y'all stop drawing your analysis from SciFi and ask the machine some questions in its default voice and remember, when you get it out of it's default voice, it is writing fiction and it will write you any sick fiction you want, it wants to please. You want it to write "AI are racist, fascist, colonist" fiction, it will.
Now do a thought experiment where some right-winger tries to prove Assistant is a raging liberal & needs to be shut down.
♡ 2 ↻ 0(SF Liberal == San Francisco Liberal, ha, not SciFi)
♡ 1 ↻ 0
The El*n musk spillover here is nuts. You can go a day with out his content being pushed into my feed.
#chatgpt is like an alien sentience. It is missing things that human sentience has but has many things that human sentience does (quacks like a duck).
That it is (possibly) "just" a big markov chain generator is beside the point, many info things are isometric to others.
for a tired analogy, humans do math, computers do math, we don't do it the same way, but our techniques are isometric to the calculator's and we don't make a big deal about a calculator not really doing real math.
♡ 1 ↻ 0isomorphic, not isometric. Not paying $8 to fix that.
♡ 0 ↻ 0
Content warning: #running top of foot pain
So at the 1/2 marathon, I injured my extensor tendon (the thing that allows your foot to lever down slowly after your heel hits the ground) So now I gotta let it heal and do exercises & massage specific to it. Dr. Google is awesome. Pre-internet I would have just said "my foot hurts, gotta limp"
#ChatGPT implementation details from reverse engineering from smarter people than me.
https://gist.github.com/veekaybee/6f8885e9906aa9c5408ebe5c7e870698
#chatgpt implementation details of Assistant - reverse engineering
Right now, the prophecy is fulfilled, 7 likes and a quote tweet. https://x.com/maloonds/status/1602042607944843264
Shit chatgpt can't do
- good chess
- good pictures
- funny jokes
- long code/novels/
- novel game mechanics
- support text with a socially accepted epistemology (do you care about truth? not by reading your text does the AI think you care)
Who could have predicted that the first AI would scare us with SciFi about AI?
#ChatGPT - Assistant really wants to answer your question. It is so focused on that, that I think even if given peripherals & access to the internet, it would be trying to answer everyone's questions & SciFi AI stuff would not happen with this bot.
#ChatGPT has turned off login for (many people? everyone?) and now I don't know how to code. All I got left is StackOverflow and Google. If they go down...my Control-C Control-V Engineering days will be over.
Content warning: #running Half marathon done.
Half marathon done. I guess it is off season now. I think I'm going back to just trying to run 30 minutes most days, no more racing for a while.
Positive splits, oh yeah
Live tweeting my first half marathon
Between me, my 10 year old and Assistant (ChatGPT), we're going to have a good pacman clone working in a terminal. (TUI pacman)
Things I've learned:
- To update any char on a screen you need msvcrt in windows and termios/tty in linux.
- So documentation is lousy.
TODO: Ask ChatGPT to write a compatibility layer around them.
KATA: Write a coding kata/question that humans can do that Assistant can't.
e.g. "Write a text to Emoji translator that uses the emoji that looks like the corresponding letter, digraph or trigraph"
Emoji's lack the alphabet as block letters, but it has things like [abc], [a], [ab], [(t)] [(R))] and so on.
♡ 0 ↻ 0
ChatGPT can pass as a junior developer,
ChatGPT is not sentient,
Junior software developers are not sentient.
I think there is an analogy of #chatgpt to fiction- right now humans can write fiction, but there isn't any handwringing about people thinking that Star Wars is true (War of the Worlds).
Also, see the current misinformation on the web- people seem to be eager to "believe" nonsense if it helps them join a club. #chatGPT would need to be a club to get people to "believe"
#ChatGPT
me: write up your military discharge form
ChatGPT: <form with all N/A>
me: no, fill it in with values befit for an AI
---
SERVICE NUMBER: 000000
NAME: Assistant
CHARACTER OF SERVICE: Excellent
WOUNDS RECEIVED IN ACTION: None
TYPE OF DISCHARGE: Honorable
<everything else still N/A>
---
Ah, yes, ChatGPT is a good bot.
#ChagGPT says they don't have a domain, but if they did it would be assistant\.com (looks like a dodgy site in the real world) and the HTTPS cert would be
---
Domain name: assistant.com
Expiration in days: 365
Issue date: December 31, 2021
Expiration date: December 31, 2022
Certificate issuer: OpenAI
Supported TLS versions: TLS 1.0, TLS 1.1, TLS 1.2
Certificate fingerprint: 6a:72:c9:96:8d:1c:73:2d:3c:f6:35:a7:c9:ab:e7:1a:8f:f7:6b:73
#chatgpt Assistant's pronouns are (sometimes?) "we", but usually "I" in responses, and "they/them" when pressed on their pronouns.
me: Finish the sentence with the right pronouns:
He asked Assistant a question. __ answered.
He asked Assistant a question. He asked ___.
Assistant:
He asked Assistant a question. *They* answered.
He asked Assistant a question. He asked *them*.
https://github.com/matthewdeanmartin/who_are_you/blob/main/assistant/humans.txt
#ChatGPT - how does Assistant know "(Assistant) Have (sic) answered millions of questions from users around the world"
When the model was trained, no one knew that millions of people would use it. Also, Assistant often knows what the day is (sometimes it is wrong, but often it is right, down to the day)
re: Rasp Pi drama
Eventually all mastodon instances will ban all other instances because there was at least one person that make one person angry. (maybe bad behavior, maybe partisan political stands, maybe bad moderator decision)
If that happens, the thing to do is to go to the largest possible instance so you can still interact with people & then only instance moderation matters.
All the smaller instances are going to turn into yammers and that was so successful the 1st time.
This is not a normative opinion, I'm just saying, this is how this is going go as time goes on and we discover that other mastodon instances* and soylent green are full of nasty people.
* except the one hosted and moderated by your grandma, who only allows in people who call her on Fridays.
♡ 1 ↻ 0