Do you REALLY want to get a feel for how GPT-4o does what it does? Just complete this poem — by doing so, you’ll have performed a computation similar to the one it does when you feed it a text-plus-image prompt.
In #homeassistant, using #nodered to make an API call to a #llamacpp server running #mistral 7B model. I create a prompt that asks it to summarize all the data in my house from the sensors. The results are pretty impressive for such a little model. Now I get a customized rundown, Jarvis style.
Useful? Probably not. But cool as hell. :cool_skelly:
My #HomeAssistant setup reminds me to take my #medicine if I haven't taken it on time. Spent some time rigging it up to connect to a #Mistral#AI. With a clever prompt, the reminder I get every night is now customized, encouraging, and inspiring. 😎
Looking into other AI customized elements to work into my setup.
Yannic Kilcher uses GPT #LargeLanguageModel to emulate a CPU architecture executing machine instructions and runs the Snake game. Doom on #GPT4 is next? https://youtu.be/rUf3ysohR6Q
Admittedly a glorious waste of computing resources for a very unreliable emulator.
Helping someone debug something, said they asked chatgpt about what a series of bit shift operations were doing. He thought it was actually evaluating the code, yno like it presents itself as doing. Instead its example was a) not the code he put in, with b) incorrect annotations, and c) even more incorrect sample outputs. Has been doing this all day and had just started considering maybe chatGPT was wrong.
I was like first of all never do that again, and explained how chatGPT wasnt doing anything like what he thought it was doing. We spent 2 minutes isolating that code, printing out the bit string after each operation, and he immediately understood what was going on.
I fucking hate these LLMs. Empowerment is learning how to figure things out, how to make tools for yourself and how to debug problems. These things are worse than disempowering, teaching people to be dependent on something that teaches them bullshit.
Edit: too many ppl reading this as "this person bad at programming" - not what I meant. Criticism is of deceptive presentation of LLMs.
@Agonio@jonny
How about release a #SciComm Youtube video on a channel with over 600K subscribers that relies on ChatGPT to make some key calculation https://youtu.be/5lDSSgHG4q0?t=15m18s
and then include its answer without verification even though it is out by more than 12%
And maybe a follow-up video where placing unquestioning trust in the #LargeLanguageModel to generate to correct engineering parameters results in the project failing
I asked #BingChat (creative #GPT4) to write a nursery rhyme about a billionaire-owned social network struggling to cover a major news event in the Middle East.
Note that "struggling" was the only word that might have guided the #LargeLanguageModel in such a dark direction. Or perhaps "billionaire" also carries a not-insignificant quantity of negative connotations for the #GenerativeAI#AI#ArtificialIntelligence
This week on my podcast, I read my recent @medium column, "How To Think About #Scraping: In #privacy and #labor fights, copyright is a clumsy tool at best," which proposes ways to retain the benefits of scraping without the privacy and labor harms that sometimes accompany it:
If you'd like an essay-formatted version of this thread to read or share, here's a link to it on pluralistic.net, my surveillance-free, ad-free, tracker-free blog:
We should scrape all of these looting bastards, even though it will harm their economic interests. We should scrape them because it will harm their economic interests. Scrape 'em and scrape 'em and scrape 'em.
Now, it's one thing to scrape text for scholarly purposes, or for journalistic accountability, or to uncover criminal corporate conspiracies. But what about scraping to train a #LargeLanguageModel?
Yes, there are socially beneficial - even vital - uses for #LLMs.
In my latest Locus Magazine column, "Plausible Sentence Generators," I describe how I unwittingly came to use - and even be impressed by - an AI chatbot - and what this means for a specialized, highly salient form of writing, namely, "bullshit":
If you'd like an essay-formatted version of this thread to read or share, here's a link to it on pluralistic.net, my surveillance-free, ad-free, tracker-free blog:
I had unwittingly used a #chatbot. The website had fed my letter to a #LargeLanguageModel, likely #ChatGPT, with a prompt like, "Make this into an aggressive, bullying legal threat." The chatbot obliged.
I don't think much of #LLMs. After you get past the initial party trick of getting something like, "instructions for removing a grilled-cheese sandwich from a VCR in the style of the King James Bible," the novelty wears thin:
Generative AI, meaning models which imitate existing digital artifacts and media isn't what all this is about. This was an attempt to find a common name for both #LargeLanguageModel chatbots and to generative image/video/sound models in wide use now. It's a descriptive label which looks to the past.
The next generation of these systems won't be imitative, but they will be trained in self-competition as agents. It's not about them generating content based on examples, it is about surpassing human cognition in agents making decisions. We already use #agentic architectures when we deploy LLMs. It's still the same tech though, a natural next step, not something radically different.
LLM chatbots weren't made to be assistants or chatbots, they were made to measure the intelligence of the underlying deep neural network. It comes from #AGI tech, not from generative models. That is the correct umbrella term for these technologies, at least as they apply to #automation and surpassing human intellectual and cognitive limits.
“You can take pretty much any #LargeLanguageModel you want and put it in this and it will inference like crazy.
The #inference cost of large language models will drop significantly.”
Continuing the #python#CoPilot example, one can make it work by upping their prompting game.
In addition to the description of the algorithm, give the desired input and output.
It immediately suggests to define a class for intervals, followed by a line sweep over the sorted intervals.
It can generate some (sorted) test cases after prompting. *Surprisingly it had some issues with printing the results (for whatever reason, it could not generate the unpack-print loop, so I just did it
@mjgardner@matsuzine@Perl@EricCarroll There is another dimension that should also be considered here, that of productivity. Getting the prompting right for #largelanguagemodel to generate a chunk of code that works & testing said code, may end up taking more time than actually doing the deed by hand. It makes more sense (IMHO as a hobbyist) to have good metadata about functionality for software libraries so that they can be located & reused, with #metaprogramming generates the boilerplate.
Prompt Injection: Marvin von Hagen trägt vor, wie er Bing Chat austrickste
Marvin von Hagen fand einen beachtlich cleveren Prompt für Bing Chat: Dieser gab Herstelleranweisungen preis. In einem Vortrag erklärt der Student den Trick.