Artificial Intelligence

[PopeBeanieModerator]

ChatGPT (and some other AIs) are evolving quickly right now. I started this group over two weeks ago but haven’t posted anything until now.

When I asked it what version it is, it said “GPT-3”, which is what most users are still using because it’s free. At present, this free version limits us to a maximum of 25 queries per day.

To jump into this adventure for your own edification, see this AI group’s post on Publicly Available AI Interfaces.

I’m discovering the ways GPT’s been useful for me, so far. In essence, I can query GPT that I would usually use Google for, but with whole lot of clicking and rephrasing of search terms in Google. Google’s already been going downhill for me, seemingly returning results they’re more interested in having me click on. I’m guessing that the general public’s non-expertise in how to phrase search terms has given Google more opportunity to prioritize results by which clicks would benefit them the most. So far, in ChatGPT, I’m getting results that seem totally unbiased in any human way, usually saving me a lot of time.

GPT’s helping me improve how to rephrase questions, albeit sometimes to an annoying degree. Google does this too, but it consistently takes maybe ten times more attempts than GPT to get a useful answer from Google, and even that can still lead to nothing useful.

ChatGPT saves my queries, even from previous weeks. Of course one has to sign up for that convenience, but it’s free atm.

For the sake of brevity here, I’ll truncate most of the examples.

EXAMPLES

Q: do entangled particles connect through another dimension?
A: Entangled particles are two or more particles that are connected in a way that their states are correlated, even when separated by a large distance. This correlation is a consequence of the laws of quantum mechanics. / According to quantum mechanics, entangled particles do not connect through another dimension. Instead, their correlation is described by a mathematical relationship known as quantum entanglement. When two particles are entangled, measuring the state of one particle instantly determines the state of the other particle, regardless of the distance between them.

Q: what states have the highest number of cops per capita?
A: The number of police officers per capita can vary widely between states and even within different cities and regions of a state. However, according to recent data from the Bureau of Justice Statistics, as of 2016, the states with the highest number of full-time law enforcement officers per capita (per 1,000 residents) were:

Louisiana – 4.50
Mississippi – 4.07
Washington D.C. – 3.89
Alabama – 3.78
Alaska – 3.73
It’s important to note that these statistics are from 2016 and may have changed since then. Additionally, the number of police officers per capita does not necessarily reflect the quality or effectiveness of law enforcement in a given state or community.
[I asked for this ranking, because my visits to Florida made me feel they have a relatively high cops per capita. At first GPT just gave me the first five states, so when I asked for a list of all states, it listed 50, and Florida was ranked only 21st at 3.07. California, the state I’m more recently familiar with, was ranked 35th at 2.77. Lowest ranking was Hawaii at 1.94.]

Only three weeks later (today, 11 April), I asked exactly the samely-worded question for all states, and this time it came up with 2020 per capita stats, significantly different from the 2016 stats. Also, it generated a nice looking table instead of a list, but sorted it alphabetically by state instead of by per capita ranking.

Cool feature: It remembers the context of the previous query. E.g. When I said “please sort that by per capita ranking”, it did, in a new table. When I said “please number the rows”, it did, in another new table. Awesomely, unlike the list presented three weeks ago, the tables could be copy/pasted into a spreadsheet, with headers and data separated into columns. But…

Interesting Flaws:  Three weeks ago, the list was missing Washington state while it included Washington D.C. Today, the sorted tables were missing four states; they included Washington state but not Washington DC. So it looks like there’s a problem sorting columns when two item appear similar, including numeric quantities with the same first digits but one of them has an extra digit of precision. (Even when those numbers are presented properly in a column in numeric format, so somewhere in the process, the number is first getting sorted alphanumerically.)

I could go into further detail on my discoveries, but I’ve already spent way more time on this than I expected.

 

• This topic was modified 2 hours, 6 minutes ago by  PopeBeanie. Reason: less references to version 3; and I merged the version 4 topic into this more general, versionless topic

[PopeBeanieModerator]

GPT-4 (pay) is an impressive upgrade over GPT-3 (free). Microsoft Edge includes its customized interface to it. If you’re not interested in scientific research and papers, or if you don’t want to use Edge, skip the following video and read further on below for my initial summary of the GPT upgrade.

The free GPT version limits us to 25 queries a day, and the paid version limits us to 25 queries every 3 hours.

In this post, I’m only including videos that are two weeks old, as of mid-April 2023. I’ve paraphrased the actual video titles.

Keep in mind that there are other GPT interfaces other than the standard one that OpenAI, the original GPT designer offers. I just haven’t researched them, yet.

ChatGPT Prompts for Scientific Research and Papers
10 minutes

My short experience so far, and a video…

I’m paying $20/mth for OpenAI’s GPT-4 upgrade service. Note that for each query you make, you select to use GPT-3 that returns answers faster than the free version, or select to use GPT-4 which does not guarantee quick responses. GPT-4 should still be somewhat faster, because of built-in improvements that just make the engine itself work faster.

This video’s less scientifically niche and is more general, but it’s still made by the same scientist as above.

GPT Tricks
17 minutes

• This reply was modified 1 month, 3 weeks ago by  PopeBeanie. Reason: clarified max queries allowed in free vs paid version
• This reply was modified 2 hours, 9 minutes ago by  PopeBeanie.

[Reg the Fronkey FarmerModerator]

I have used it everyday since last November.  Your example is esoteric but that is not the important thing.  It is the speed and accuracy of the coding that is impressive.

[PopeBeanie here. I moved that post to here where I’ll put other uberly-esoteric posts. Was thinking of doing that, anyway.]

I was very busy today and had to write an email to a prospective client that I visited yesterday, for the first time. I was reasonably sure I would get them to sign up. Out of curiosity I asked GPT to compose a letter for me. Within 5 seconds it was done. I was very impressed and only edited it slightly to add a little humor as a follow up to part of our conversation yesterday.  30 minutes later I was invited back to meet the full management team. If I get in the door a second time I don’t leave without a signed contract 🙂

• This reply was modified 4 weeks ago by  PopeBeanie. Reason: PB note about moving the previous post

[Reg the Fronkey FarmerModerator]

I also asked this question today. The reply is underneath and took less than 10 seconds to compile the code and comment on it.

My query:

Write a Powershell script to test for weak passwords in AD. Passwords should be a minimum of 8 characters long with at least one capital letter, one number and one symbol.

ChatGPT:

Certainly! Here’s a PowerShell script that will check all user accounts in Active Directory for weak passwords that do not meet the following criteria:

Password length of at least 8 characters
At least one uppercase letter
At least one lowercase letter
At least one number
At least one symbol

# Define the regular expression to check for strong passwords
$StrongPasswordRegex = “^(?=.*[A-Z])(?=.*[a-z])(?=.*\d)(?=.*\W).{8,}$”

# Get all user accounts from Active Directory
$Users = Get-ADUser -Filter *

# Loop through each user account and check if their password is strong
foreach ($User in $Users) {
# Only check for weak passwords if the password is not expired and the account is enabled
if ($User.Enabled -eq $true -and $User.PasswordNeverExpires -eq $false -and $User.PasswordExpired -eq $false) {
$Password = ConvertTo-SecureString -String $User.Password -AsPlainText -Force

if ($Password.Length -ge 8 -and $Password -cmatch $StrongPasswordRegex) {
Write-Host “User $($User.UserPrincipalName) has a strong password.”
}
else {
Write-Host “User $($User.UserPrincipalName) has a weak password.”
}
}
}

This script uses a regular expression to define the criteria for a strong password, which includes at least one uppercase letter, one lowercase letter, one number, and one symbol, with a minimum length of 8 characters.

It then loops through each user account in Active Directory and checks if their password meets the strong password criteria. If the password is weak, the script will output a message indicating the user account with a weak password.

Note that this script assumes that the password for each user account is stored in plaintext in the Password property, which is not typically the case in Active Directory. You may need to modify this script to retrieve the password for each user account from a different source, such as a secure password vault.

 

[Reg the Fronkey FarmerModerator]

I do find your query and its answer a much better display of its abilities, especially its interpretation of the question. Mine saved me about an hour of coding and testing. It ran correctly the first time and found 6 weak passwords from 96 users accounts in 1 minute. I then edited it to display the results in a html doc which I could print and send to their useless IT manager!

[Reg the Fronkey FarmerModerator]

I think the cost of living is about to go up by $20 per month!

[Simon PayntonParticipant]

According to the BBC Radio 4 programme I put in my “update”, Chat GPT is just a super-sophisticated predictive-text machine, based on what it’s learned from scooting around the (human-built) internet.  But clearly, it’s very good at what it does.

[PopeBeanieModerator]

Simon Paynton wrote:

Chat GPT is just a super-sophisticated predictive-text machine

I got that impression, too, but then later realized that being able to interpret human language and then respond to it in kind is a huge leap of “intelligence”, in its own right. These are word and phrase association machines, which don’t even have to really understand everything in the way we do… all they have to do is calculate which words and phrases will correlate and follow the query, such that it “makes sense” in the same context as the query.

I am oversimplifying the above, but this is a powerful, “emergent” kind of intelligence that its human designers didn’t seem to expect, and many of them say they still don’t understand how it works so well.

[PopeBeanieModerator]

Ilya Sutskever

Designer/Chief Science Officer and Co-founder of OpenAI.com
Interview by Stanford eCorner, “Inside OpenAI [Entire Talk]”
April 9, 2023

My rough notes with some time stamps are below the 50 minute long video. (I’d say skip the last 10 minutes.)

~8 mins Ilya’s explaining how ChatGPT was trained on data by predicting “the next word” based on previous words given to it; when it predicts correctly, a “back propagation algorithm” is applied to reinforce the code that worked.

Ilya spoke more (about what I forget, except he mentioned having a relationship with Microsoft), when it got interesting discussing consciousness, and how he believes it’s scalar (my word), not binary. He mentioned animals that surely have some degree of it.

~18 mins, discussing Musk’s $1 million for startup in an open source, non-profit mode, then going to closed source and at least a partial profit mode, and the ethics of that. Was mentioned that the two big AI players right now are MS and Google. Ilya feels responsible as the principle scientist in OpenAI’s overall impacts.

~22 mins: He says a long term arguent about open source is about keeping it out of the hands of bad actors (my phrase), when AI gets super powerful. Next he spoke about the need to pay for computer time, so the need to have a for-profit model. At 26:10 he calls it a “capped profit” model.

~32 mins in, wrt other countries getting AI and possibly using it against our interests. (I’d like to know more about what Ilya means when mentioning “the end of AI training”.) He thinks that future AI will need to be regulated for our safety. As for how to deal with other country’s AIs, he said it’s out of his area of expertise.

35:22 Ilya’s opinion about keeping openAI’s work relevant (mentioning KPIs??), and its core technology.

36:46 Future of openAI in five to ten years: Will it be a provider that consumers go to, or behind the scenes in service to other vendors? Answer, things change so fast, he can’t say. Then other less interesting (imo) stuff to the end.

[Author]

Posts