311–320 of 333 posts
Earlier quoted context omitted.
> $3M per day That figure really is eye watering like Sam said. Any chance we might be able to reduce that cost? It must be the power costs right? The hardware is one time cost I assume. Why not power it over renewables?
They don't own the hardware, it's running on Azure
Earlier quoted context omitted.
Cumulative damage may be very small now, at the beginning of this age, but educators I talk to certainly are having a reckoning about it. Whether this will ultimately be bad for education is an open question, but with the way many classes and homework are designed now, students using ChatGPT to answer questions certainly do learn less.
I 'borrowed' the (ten year old) daughter of a friend and sat her down in front of ChatGPT to see what would happen, proposing she ask for help with her homework. And yes, the first thing she did was see if it could literally answer it for her. It did. However, once that was done with she proceeded to massively expand the scope of the original homework... I've rarely seen anyone her age as enthused with learning , the…
Earlier quoted context omitted.
Yeah, homework was designed as a way to make self study monitoring and verification scalable. It's not a bad solution but if the availability of AI makes it hard to control that someone has done their homework instead of giving it to an AI, then you have to look for alternatives. Thankfully, in such a world, AI is available to the "teacher" side as well and can serve as a way to both check that the student is doing t…
The "if I was a teacher" thoughts on how to handle a ChatGPT that can do short answer questions for English, History, and Social Studies classes... I'd have a set of N questions that shouldn't take long to answer but demonstrate that the material has been read. Yes, ChatGPT can answer them. The second part would be in class, after the homework had been handed in pull a name and number out of a hat and have that stude…
Earlier quoted context omitted.
Makes sense. Do you know if there exits a re-implementation of a ChatGPT-like stateful conversational interface on top of the GPT-3 API? A cursory search doesn't turn up anything.
If you look at https://beta.openai.com/playground/p/default-chat?model=text-davinci-003 https://beta.openai.com/playground/p/default-friend-chat?model=text-davinci-003 https://beta.openai.com/playground/p/default-marv-sarcastic-chat?model=text-davinci-003 you can see a chat "session". The issue is that to maintain the state, you need to maintain the history of the conversation. For example, the friend chat starts out…
My son asked if which of two animals would win in a fight and we just got a lecture about how fighting is bad. I hope they’ll give an option to turn off what I’m calling “hall monitor mode” That could be worth paying for.
On my website https://aidev.codes you can access the OpenAI text-davinci-003 model which is fairly similar to ChatGPT and does not have guardrails like that. Please read the help page. But basically !!create file.txt Who would win in a fight between an elephant and a bear? First, rate the strengths and weaknesses of each. Then do a blow by blow of the each attack and counter.
Earlier quoted context omitted.
I want personas. Your racist uncle, my woke sister, your on the fence father, and my happy mother. The reality is these each will answer questions in totally different ways, lies will dramatically vary and as much as it’s kind of weird to say it, i think this could turn out to be practical.
Charles is a conspiracy theorist. Wanda has left wing views. Fred is indecisive. Jane tries to keep people from arguing. Describe a conversation between Charles, Wanda, Fred, and Jane at a thanksgiving dinner. Fred: So, who is going to cut the turkey? And then it continued... Jane: I can do it if you'd like. Charles: Before we get to that, I think we should talk about the election. Wanda: Absolutely! I'm so excited f…
Earlier quoted context omitted.
It's pretty widespread on tiktok in the student circles I'm in... And the fact openAI is still struggling to handle the load in peak hours 2 months after launch tells me they must be seeing pretty big user growth.
Or they haven't properly designed their cloud arch to even moderately scale. We can obviously see their AI and CS talent, but do we have any gauge on their internal systems and IT talent? I've used ChatGPT maybe 3 times since launch and 2 out of 3 attempts resulted in a "come back later" type message. It would make sense to see which accounts are the heavy users and throttle them in favor of lightly used accounts but…
This changes how easy it is to scale.
> Elon Musk-founded research body While technically true, he's no longer involved with them, no?
Earlier quoted context omitted.
Yeah, I have a product that rewrites content under a persona. I wanted to take the persona (a plain English description of how someone communicates) and generate a headshot to go with it for the UI. So I asked ChatGPT to describe what the persona looks like as if they were talking to a sketch artist. Instead of getting something I could feed into DallE, I got a lecture about stereotyping.
But you can't tell what someone looks like based on how they communicate.. AI was right
You can ask it to assume the persona of a well educated police officer writing a police report for a judge, and it will gladly do so. And that communication is unlikely to carry an American southern accent even though there likely exists a well educated police officer who speaks with a southern accent. But ask it to describe that police officer so I can draw a picture and it’s a different story.
Yet if you put “police officer” into a search engine there is a clear aesthetic that we (humans) associate with that stereotype. Blue/black outfit, hat with badge/logo, etc. That clear aesthetic is what I want in the headshot. And instead I got a lecture on inclusion - something important but tangential to the headshot of a police officer.
Earlier quoted context omitted.
Btw you can use it for cheap using this GUI: https://dall-e.sonnet.io/
Why does this make it cheaper? Couldn't this site just be stealing people's API keys?
API calls are ca. 7 times cheaper than credits. (up to 1000 images vs. 150 when paying with credits)
> Couldn't this site just be stealing people's API keys?
Thanks for asking. Feel free check the source and host it yourself: https://github.com/paprikka/dall-e-ui (takes 2 min with Vercel)
Context: I made this GUI for myself when I was comparing SD and Dall-e for a small personal project. Some people seemed to like it, so I shared it.
https://twitter.com/rafalpast/status/1591138659297726464?s=2...