Yesterday, I tried getting ChatGPT 3.5 (not even the newest GPT4) to write a computer program for me by prompting alone. I didn’t try to fix code manually and simply pointed out where there were errors and what wasn’t working properly.
I was amazed that after 16 prompts, it had made me a fully working application (in Python programming language) with a full screen text mode interface and working bindings to a database.
The particular extra functionality I needed started getting a bit beyond its capabilities so I had to take over after that, but I was amazed that you could get something up and running in minutes with no coding experience required.
Please share your use cases and successes (or failures)!
Never tried it, perhaps it is one step too far for my tastes. I do see the AI generated crap people post on artists forums and watch them get taken apart by “real” artists and I think that AI generated “art” will be a passing thing. Somehow the pictures simply don´t look right, the uncanny valley feeling kicks in and what surprises me is that the the uncanny valley doesn´t just apply to the human figure but to mechanical pictures too; aircraft, ships, cars etc. They all look strange when rendered by AI.
I’m currently doing a master’s degree, and I run research papers through ChatGPT and then ask it to summarise the key points and findings in under 500 words. If it’s a particularly meaty or stats-heavy one, I add the prompt ‘in language targeted at an 18 year old’ (tried 17 and it was too juvenile, 19 started getting too academicese again).
It is an absolute godsend. It means I can be more efficient with my time in discarding papers that don’t necessarily meet my needs. I am so time-poor right now that it has rapidly become something I couldn’t continue without.
It could’ve just been a one-off or content-dependent, but it was a noticeable difference. My assumption was that it interpreted a 17 year old as a high school student and a 20 year old as a college student.
But what I found problematic was asking him for some references, like in journal articles. What he did every time was “invent” articles references for non-existing articles. as if he was combining the information from different sources into one reference.
The real question is: do you all say please and thank you?
I hav dabbled only:
-Got it to write menu dinner plans for the family with instructions day-by-day and full shopping list
-Use at work to write recommendations for awards
-Use at work to summarize emails for top management
-Used to write LinkedIn in posts for me
-To.rewrite some.paragraphs of my CV
Am enjoying it more and more and given it’s here to stay I am trying to use it also more and more
I‘ve used Bard and ChatGPT a fair bit and find them useful. To say one is better than the other is difficult but ChatGPT seems to be more technically efficient whereas Bard is really up to date and actual and gives sources to its evidence. You can have continuing conversations with them and return days later to the same topic. It’s kind of fascinating. Reading and interpreting are very strong topics of both but actual suggestions don’t always work.
write an ncurses application which when pressing ‘o’ key moves focus to a part of the screen for SQLite command entry. the input is fed into a function sqlCommand() which issues the sql command to a sqlite3 database and returns a list of results. the results should be limited to 10 and displayed in another window. the ‘j’ and ‘k’ keys can be used to navigate up and down to select a result and the highlighted row calls a getBody() function which displays additional data related to that result row into another data winow.
Which admittedly is technical but I knew exactly what I wanted to get and the tools to use, but you can also ask ChatGPT in layman’s terms.
At various points, I needed functionality but didn’t know what could be used and ChatGPT suggested appropriate libraries of code and how to interfact to them.
I remember reading an article from a guy who didn’t know how to program and got ChatGPT to write an entire program for him to calculate sun position so that he could adjust his solar panels on the van he lived in.
He just fed queries and error messages in until get got the final result.
A while ago I played with image generation and ran a batch of generations changing the prompt from “a woman…” to “a 19 year old woman”… but for all ages 19-100 and the model was able to create a series of images and age the woman appropriately.
I guess for images, there could be more samples whereas for writing there might not be so much data structured by writing age.
I just saw a demo of a portrait animation technique released a couple of months ago and was blown away. Just taking a single photo, that photo can now be animated with a voice track.
I was amazed by the very human gestures and micro-gestures - the tiny movements of the eyes etc.
Edited to add that the source photo isn’t real, it is an AI generated image, but if anything, the animated video looks even more realistic.
There will be porn, but internet bandwidth today is not dominated by porn and there’s no reason to believe that AI will be any different. Once the companies find a way to monetize and recoup the huge investments in data centres, it’s much more likely to be corporate subscription models and cheaper, home versions.
I wonder what would happen to porn If somebody were to program an AI with the morals of a 17th century puritan and the compassion of a seagull and let it loose on the internet?
Morals and compassion are human traits, not digital. Training an AI linguistic model on a set of data will “inherit” the morals and compassion within that data set. Train it on the bible or the quran and you end up with a sadistic killer with no compassion. Train it on Uncle Toms Cabin and you get a racist. So many options!
Porn for sure will be one usage, but saying the main use of AI will be for porn is saying the main use of video images is for porn. Yes, pretty much any technology that can be used for porn will be used for porn, but the usage of AI will be much wider and more impactful than just porn.
I suspect that giving references will not be something that you see soon. There are a number of reports of the AI companies stealing the data they use. For example the entire New York Times archive.
The companies were desperate for data to get ahead in the AI race. I imagine at somepoint the ‘more data’ shortcut will eventually give way to other better techniques that are much less data intensive.
