Be advised: this thread is not for serious in-depth discussion of weighty topics (we have a link for that), this thread is not for anything Culture War related. This thread is for Fun. You got jokes? Share 'em. You got silly questions? Ask 'em.
- 55
- 1
PaperclipPerfector
1mo ago
(text post)
1678 thread views
Be advised: this thread is not for serious in-depth discussion of weighty topics (we have a link for that), this thread is not for anything Culture War related. This thread is for Fun. You got jokes? Share 'em. You got silly questions? Ask 'em.
Jump in the discussion.
No email address required.
Notes -
Being a bit of a cheapskate, and a bit illiquid at the moment anyways, I haven't played around much with AI until the free tiers started getting good enough. Hot dang have I been having fun, and I have been cracking up at the edge cases of what they'll refuse or eventually create, sometimes in weird and roundabout ways, especially Gemini.
Overall, Gemini 2.0 produces more accurate images than Grok 3.0, but refuses way more often. Going back and forth to generate prompts in Grok then images in Gemini probably worked the best. It doesn't seem to like generating couples posing for a portrait, though I could generate more people (Gemini likes harems, can't decide if I'm surprised or not) and sometimes talk it into removing the extras. Families are difficult. I thought the problem was a reluctance to generate kids, but it's the same issue as couples- seems the content restrictions make it reluctant to one-shot (am I using that right?) a man and woman together. Generate a mother and child then add the dad back later, works. DALL-E 3 is... good at what it does, which seems to be only slightly related to the prompt but highly detailed. Ask for [detailed description of an adventurer and a witch about to enter an ancient forest], get back an old wizard with ridiculously ornate robes instead. Might try 4o next month or whenever the new image gen trickles down to us freeloaders.
Playing around with them for writing has been interesting and rather less convoluted. Someone on twitter mentioned Grok is especially good at "human prompting" and I agree, it's better with its own little suggested questions at the end. Feels a little less 'magic' than the image generation, though. Turning a thousand words into a picture in a few seconds is more affecting than turning a thousand words into... more words.
So! I'm sure I could dig through archives here but I'm lazy. Are there good, not-too-hyped resources for learning more about how these dang things work and how to wield them properly? Assume basically no programming experience, and my knowledge of how computers function is somewhere around "electric demons in magic sand" metaphor level. I've missed the boat on learning all that much but I'd still like to round out my knowledge a bit.
Are any of you using LLMs for fun projects? I've heard Claude makes a decent life-coach substitute but I haven't tried that out yet.
I don't actually know what you do, but perplexity is excellent for sourcing papers from the scientific literature if you're starting in on a new (sub)domain.
This week I had to replace the bottom bracket on my bike. Chatgpt told me which kit to buy, and then I was blown away when it instructed me to go take a picture of both sides of my crank/cassette so it could be certain what we were ordering would fit. Somehow I missed them adding that functionality, but it's mind boggling to me that it can 'see' in addition to just parsing language. I guess I doxxed myself given that my openAI account uses my real name and my ignorance of bike repair/photos of my shitty commuter bike are in their database. Maybe the next SolidGoldMagikarp exploit will start regurgitating all my personal data for the world to see.
While I was at a startup, I was responsible for all kinds of biology subfields that I had no expertise in. I wonder if they'll ever realize how I did all the modeling, although at least a lot of our data has been validated externally. I advertise myself as a full stack biologist now :)
It's still not really useful for professional things. When I ask it to come up with new ideas or commercial opportunities in [area], it just regurgitates reviews I could read myself or tells me every garbage idea I have is phenomenal. But I haven't had the patience to systematically test all the available models recently, or put serious effort into prompting.
Neither do I, anymore. Well, I do know, it's just boring and I'm grouchy about it. My career isn't research involved these days, and given everything the stability of sticking around is preferable to the risk of finding something more satisfying. I do miss those days and Perplexity would've been a godsend back in the day, I'm sure. Might look into it just for fun and to scratch the old itch of wanting to learn more again. That skill has grown stagnant and contributed to other issues.
Dang! The paid version, I assume? Starting to wonder if it'll be worth the cost just to help with some minor repairs around the house.
Yeah, I've definitely picked up the chat function not being particularly creative and being obsequious unless you run into one of the walls. I've tried refining some creative writing ideas with it a couple months ago but wasn't impressed. As fast as things move it might be worth another shot.
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link