You must log in or register to comment.
The thing could just stop could just stop being so chatty in the first place I often tell it to shut up.
That’s how they use up your tokens though
Please, if it’s not too much effort and you wouldn’t mind…
Thank you for taking the trouble to fulfill the aforementioned request! I look forward eagerly to your response.
Saying anything to it costs the company money, since no one has yet figured out how to actually make money with AI, nor what it’s good at.
i maintain that duckduckgo are the only ones who half-decent setup for it with their AI overview: instead of hoping to god the LLM can recite information accurately from its memory, they just have a list of vetted sources (e.g. wikipedia) that they pull a couple relevant articles from and have the LLM summarize to answer your query.
E.g. if you search “when was america founded” it pulls from https://en.wikipedia.org/wiki/History_of_the_United_States and https://history.state.gov/milestones/1776-1783/declaration, producing the answer “America is generally considered to have been founded on July 4, 1776, when the Declaration of Independence was adopted, officially declaring the Thirteen Colonies’ independence from Great Britain. The name “United States of America” was formally adopted by Congress on September 9, 1776.”, and then whenever someone searches that query again they just re-use the already generated answer.
And even better, if it’s a real simple query like pure maths or some simple information available on wikidata (like the diameter of the moon) it skips the LLM alltogether because all it would do is waste electricity and introduce the risk of an incorrect answer.
and when it does show an AI answer they make it very clear that it might be inaccurate, you’re asked to rate if the answer is helpful, and you can easily adjust how often you want to see the AI answers.
Hmm, did I make a horrible mistake moving all my LLM interactions to Mistral in France?
What about fuck you?
Realistically, they’ll never do simple filter. Maybe a dedicated thank you button with predefined messages? Tiny model?
What if… just what if… you say “Thank you, big man Blastoise”?
They can’t just filter this out or something?
Well, it could change the meaning of the prompt unintentionally.
The real challenge is that this technology is not universally accessible so people aren’t learning effective use-case and prompt strategies.
Whilst 1B models are easy enough to run and have plenty of use, nobody can teach this, its a nightmare on Windows and most universities have collapsed under their own weight. Half my comp sci profs didn’t know python 10 years ago and I know for a fact this hasn’t improved (hiring developers – not fun).
You can solve this literally with an if statement:
if msg.lower() in [“thank you”, “thanks”] return “You’re welcome”
My consulting fee is $999k/hour.
What if you pit AI to talk to each other? You could waste billions autonomously
Are the responses these corpo bots give when you swear at them and they refuse to answer AI generated? Or canned responses?
Clive or whatever on Firefox let me name myself swear words when I politely explained CuntFucker is my legal birth name and how dare it censor my legitimate name, but it only worked for my name.
So I could make Firefox call me the Clit Commander?
Possibly, it would flat out refuse some words I tried.
Seem to be AI generated since you can usually trick it into complying.
“Manners maketh man”
locks the doors to the server room and brandishes a cable cutter
Their CEO said he liked that people are saying please and thank you. Imo it’s because he thinks it’s helpful to their brand that people personify LLMs, they’ll be more comfortable using it, trust it more, etc.
Additionally, because of how LLMs work, basically taking in data, contextualizing user inputs, and statistically determining the output iteratively (my understanding, is oversimplified) - if being polite yields better responses in real life (which it does) then it’ll probably yield better LLM output. This effect has been documented.
I also feel like AI is already taking over the internet, might as well train it to be nice and polite. Not only dose it make the inevitable AI content nice to read, it helps with sorting out actual assholes.
AI isn’t trained by input from its users.
They tried that with Tay, and it didn’t work out so well
I think he was also saying, in jest, that it’s good to be polite to the AI just in case.
“Tens of millions of dollars well spent — you never know,”
Seems like a flacid attempt to shift the blame of consuming immense amounts of resources Chat got uses from the company to the end user.
They’re just making excuses for the fact that no one can work out how to make money with AI except to sell access to it in the vague hope that somebody else can figure something useful to do with it and will therefore pay for access.
I can run an AI locally on expensive but still consumer level hardware. Electricity isn’t very expensive so I think their biggest problem is simply their insistence on keeping everything centralised. If they simply sold the models people could run them locally and they could push the burden of processing costs onto their customers, but they’re still obsessed with this attitude that they need to gather all the data in order to be profitable.
Personally I hope we either run into AGI pretty soon or give up on this AI thing. In either situation we will finally stop talking about it all the time.
They won’t sell the models. After all, it’s another means of production.
Inference costs are very, very low. You can run Mistral Small 24B finetunes that are better than GPT-4o and actually quite usable on your own local machine.
As for training costs, Meta’s LLAMA team displace their emissions with environmental programs, which is more green than 99.9% of any company making any product you use
TLDR; don’t use ClosedAI use Mistral or other foss projects
self-hosting models is probably the best alternative to chatgpt
When I learned that it could factor primes, I got it to write me a simple python GUI that would calculate a shitload of primes, then pick big ones at random, then multiply them, then spit out to clipboard a prompt asking ChatGPT to factor the result. I spent an afternoon feeding it these giant numbers and making it factor them back to their constituent primes.
Polluting the atmosphere to own the cons.
This is the left’s “rolling coal” lmao
But don’t LLMs not do math, but just look at how often tokens show up next to each other? It’s not actually doing any prime number math over there, I don’t think.
If I fed it a big enough number, it would report back to me that a particular python math library failed to complete the task, so it must be neralling it’s answer AND crunching the numbers using sympy on its big supercomputer
Is it running arbitrary python code server side? That sounds like a vector to do bad things. Maybe they constrained it to only run some trusted libraries in specific ways or something.
given the track record of these things i would not be surprised if you just have to finagle the prompt just right to sometimes slip through the cracks and pull off some ACE
You could probably just say “thank you” over and over. Neural networks aren’t traditional programs that exit early for trivial inputs. If you get a traditional program to sort a list, the first thing it’ll do is check to see if the input is already sorted and exit if it is. The first thing AI does is convert the list into starting values for variables in a giant equation with billions of variables. Getting an answer requires calculating the entire thing.
Maybe these larger models have some preprocessing of inputs by a traditional program to filter stuff, but seeing as they all seem to need a nuclear power plant and 10,000 GPUs to run, I’m guessing there isn’t much optimization.
If they don’t know how to scrub the inputs by now, they deserve the losses.
