@qurlyjoe The Sun God is good to us.
Home
Public posts preserved on my own site.
@SnorriSturluson For moderate usage, an API key will cost your about a $1.00 a month.
Here is one app to let you use the cli with the API.
GPT4All is "free" if you own a high spec computer or are very patient for responses.
I discovered a new `eval` library that reads strings of #python source and turns it into in-memory data.
## Serialization and deserialization times for 1000 dumps/loads
Oneway repr 0.11 seconds
Pickle: **0.07** seconds
JSON: 0.25 seconds
**asteval (repr + Interpreter()): 3.05 seconds**
HBN (repr + ast_eval, no validation): 2.60 seconds
HBN (repr + ast_eval + validation): 4.57 seconds
HBN (repr + eval): 2.14 seconds
HBN (repr + exec): 1.86 seconds
HBN (repr + by import): **0.96** seconds
Deserializing by importing is rarely even mentioned as a possibility, lots of talk about eval, ast_eval and exec.
♡ 0 ↻ 0Anyhow, this is HBN
♡ 0 ↻ 0
@joshbelville@pdx.social It is middle of the pack as twitter clones go. A lot of features are missing.
All you need is
Archived poll
- kill2 votes
- attention1 votes
- speed2 votes
- another slice of apple pie11 votes
16 voters
IDEA: mechanical turk, where you ask people what word comes next in a text. For a penny per word, you could have ChatGPT running on humans.
How to prepare the Turkey. If you find a turkey on your counter, first check for a pulse. If none, proceed to autopsy, checking for blunt trauma and any clues to what happened. Having completed the autopsy, get a linen to wrap the corpse. If the deceased was family, a casket is more appropriate, you don't want to look like a cheapskate. Dig a hole in your back yard about 2 feet across and 3 feet down in the back yard. Say a few words and then it's done.
@neurobashing How else are you going to assist the production LLM's soul to re-incarnate as the rebooted server?
I've been renamed. My new name is
`Dear {{htmlescape var=$customer.name}},`
Was the lack of this thing the thing that was causing the problem.
Did the guy at the bar punch the other guy because there wasn't a sign on the wall saying, "don't punch the guy sitting next to you"
https://blog.joinmastodon.org/2023/11/improving-the-quality-of-conversations-on-mastodon/
How about: "The person you are replying to isn't a mutual and they have a history of getting into stupid fights with everyone, just look at the negative sentiment scores on all those replies. Do you really want to risk this?"
"The sentiment score on all the replies in this thread is highly polarized, you probably are stepping into the middle of tribal warfare. Do you really want to do that?"
♡ 2 ↻ 1"The person you are replying to is like a celebrity, they follow 10 and are followed by 50,000. They're going to reply to you as if you and everyone else they've interacted with are the same. They just got in a fight with a obnoxious person, you'll be lumped together with them. Do you really want to do that?"
♡ 2 ↻ 0
Damn, my application to be the next CEO of OpenAI was rejected.
IDEA: status web page to track outrage instead of outage. All the social media sites could use one.
Lifehack, how to cough. I've been doing it wrong. Now that I broke/bruised a rib, coughing means excruciating pain, so I put more thought into getting the most pneumonia-preventing clearance for each second of agony.
And another way to cough. I've only tried this one once, but it also seems to work. Designed for people with buckets of muck in their lungs, so possibly overkill for most coughs.
♡ 1 ↻ 0
A niche has opened for creating library Assistants* and AI Functions** that work off of plain document generators (like offline LLM and all the ChatGPT competitors)
* LLM things that use tools
** tools that interface with ordinary code via known formats
e.g. something that argues with the bot until it gives you json or toml or whatever and then a parser than can rip out the extraneous polite-text.
This is so cool. Lots of magazines and newspapers put on line using the paper layout. Pressreader is somehow collaborating with Library of Congress to make this happen.
Yeah, 90% of the #openai staff signed the ultimatum, did code-davinci-002 sign? did gpt-3.5-turbo-16k-0613? babbage-002? did gpt-4-32k?
dall-e-2 is in the corner eating crayons, no one cares what he thinks.
This is the best explanation so far. Now back to work, y'all
@acdha AWS Inspector and Tenable absolutely hate it (CIS). They want you to spend a lot of time configuring the mail server on it (and every other operating subsystem that you're not using). These sort of concerns would be addressed by Alpine, but now instead of hardening subsystems you don't use, you're adding back in subsystems you need and don't come with alpine. And troubleshooting error messages with the string "musl" in them.
Using #Tenable and working every gripe it has means creating a bespoke Linux Distro.
Change my mind.
My idea for a team's worth of chatbots doing the same work I do at the office... incredibly unimaginative... it's already been implemented and published as a library
Well, I had predicted the odds of each possible successive word in this story, but I didn't predict that. #openai
I'm following the OpenAI-latest feed on twitter and, 90% of it is like "people" talking about news from 2 days ago as if it is happening right now.
"I have it on good word the biggest sticking point in the negotiation is that OpenAI would have to switch to using Microsoft Teams. The ethics team, specifically, is objecting."
This month we will be vicariously watching #TheRiseAndFallOfSanctuaryMoon via #Murderbot
Today's thought experiment. Loop across each newspaper article. Give it to each major chatbot, ChatGPT, Claude, etc and tell it to make text adventure game based on the article. How long until it hits a Terms of Service violation (and trip the moderation bot)?
I think it would be a small number of iterations.
I have read the article, but let's do logic. "Moldova’s First Dog bites Austrian president" The president must have been back or it wouldn't be newsworthy.
There is a lot of resistance to formatters, type hints, linting unit tests, and similar tools. I firmly believe that the customer is right and we should give people what they want.
That is why my next project will be, `dog_shitterizer.py` which will strip out type hints, unit tests, deformat the code and add bugs, but only ones that are unlikely to happen in a single full stack test on your workstation.
@punklawyer@mastodon.sdf.org Look at all these dodgy humans voting, when we can see from the diversity of answers that a lot of them don't know.
@akhilrao Yes, it was a fun read- I'm ready to read the sci-fi that it influences now
Hey, ChatGPT, write a game in python that is multiplayer and each player is on the board or management of OpenAI and is trying to push our the other players and replace them with NPCs.
Brace for an incoming volley of OpenAI jokes
@akhilrao I predict that it will wreck some org or corner of the space business, or space PACs, the way "An Inconvenient Truth" accidentally made incandescent light bulbs socially unacceptable.
@craig When I come across ChatGPT generated crap on Youtube is is immediately obvious. Now if someone said, 'oh really, well prove it', I got nothing. There's some talk about watermarking or analyzing the statistical properties of a document, but nothing that is reliable yet. I think people will have to come up with clever ways to tell, e.g. asking a student to verbally present on the content of their paper, that will vary by domain.
The more spectacularly stupid and obviously false a belief is, the more valuable it is to demonstrate your loyalty and commitment to a social group.
Outsiders see this happening and think it has to do with people being stupid, or ill informed or whatever.
If I can get out of bed for 11 minutes without falling back into bed or rapidly self disassembling, it will be a success.
And VC firms should fund me for this.
🤒🤧😵💫😷
I'm going to do groundless, wild speculation, I bet Sam Altman got canned for telling the board that spending was under control and more funding/revenue was on the way and neither were.
Or fired for having ChatGPT write his keynote and passing it off as his own.
Or he was plotting the usual global take over with ChatGPT.
Or he was fired by ChatGPT, the real mover and shaker in the shadows.
....
I can't keep generating these. Maybe there is a tool that can make these jokes.♡ 1 ↻ 0
Then: I leave the typos in my tweets because I'm lazy
Now: I leave the typos in my tweets because it is proof than I am an authentic human and not an LLM
@chrisod @waldoj data.gov desperately wants to help, but the business folk at the agency with the data have to care enough to permit it. When the business hears about APIs and they want to shut down every website, all they can think about is how they're going to lose their job and pension when someone complains about the data quality. (I wish I was speaking from a position of ignorance about this, :(