PGF
1.4K posts
@beckyferreira Thanks for the pointer to "Fear of supernatural punishment." Great stuff!
English
PGF retweetledi
1/There’s a reason people read threads but balk at long walls of text thrown up by Blue Check Essayists: oral vs. written storytelling styles.
Twitter evolved largely as an oral medium, stylistically. It was well suited to small bits of info one might glean in casual exchange.
English
@zachariahCurtis I had heard that posting, liking etc. might require micropayments, so I thought I'd see if that was true
English
PGF retweetledi
What if scientific articles had a short section titled: “What I really think.” No references, no justification, just transmission of your current internal model in <250 words. Would save a lot of reading between the lines.
English
We are conducting a study to reduce "hallucinations", where AIs like ChatGPT generates convincing yet false information.
If you would like to help, and are a researcher with publications and a work history, we'll ask the model questions about your cv.
philfeldman.com/Context_test
English
Hey @zachariahCurtis. This may help your search problems
LlamaIndex 🦙@llama_index
A big failure mode with RAG is retrieval quality - the retrieved context might be irrelevant, or it matches keywords but has 0 information. @cohere’s new Embed v3 model ranks documents not just on topics, but also the content quality - it’s optimized for RAG Gives SOTA performance on MTEB and returns higher-quality docs than ada-002. Thanks to @ravithejads, it’s now integrated with @llama_index. Full notebook/Colab here 📗: github.com/run-llama/llam… Embed v3 blog post: txt.cohere.com/introducing-em…
English
@francoisfleuret I am writing a book about this possibility right now. It's unnerving how good they could be about subtly manipulating large groups of people over long periods of time in a coordinated way towards a malevolent goal. Really hard to detect and disrupt.
English
How many competent researchers/developers are at this very moment, somewhere in the world, trying to weaponize LLMs and make them as dangerous as possible?
English
PGF retweetledi
Honored to receive the @sosym_journal First Paper Award at #models23 @modelsconf for our paper on "Gamifying Model-Driven Engineering: The PapyGame Experience". Huge thanks to our amazing co-authors and especially the young researchers #PhD who made the experiments a reality.
English
Fed the entirety of a recent Import AI story (importai.substack.com/p/import-ai-33…) to Ideogram (ideogram.ai) and quite interested by the results. We're beginning to have the chance to be 'in conversation' with machines about art, batting the proverbial ball back and forth.
English
People talk about the danger of AI becoming “conscious”.
We should be afraid of the people who are using AI to 100x their harmful acts.
English
Ive been writing about the billionaire-funded right-wing war on free speech for nearly a decade, and it has played out exactly as I’ve warned. Expect to see more disinformation campaigns targeting prominent left-leaning faculty members once they lose the protection of tenure.
Caroline Orr Bueno, Ph.D@RVAwonk
Colleges really need to start training administrators & faculty in how to recognize intentional provocation and disinformation campaigns. I’ve been writing about this since at least 2014, and I can count on one hand how many times a college has mounted an effective response.
English
Pay attention to this:
Florida just basically dismantled its tenure system, which will have devastating consequences for freedom of expression and will result in poorer research and teaching outcomes. Self-censorship is the goal here.
tampabay.com/news/education…
English
PGF retweetledi
@mtrc @togelius I realize I'm plugging my book, but I think you'd find chapter 14 interesting: google.com.au/books/edition/…
English
Interesting take on infinite content from @mtrc. I agree with some of it, although I think I have a more optimistic take: I think we can make infinite content that is meaningful because of its high quality. Scarcity is not necessary. At least, that's what I want to believe.
mike cook@mtrc
New post: Infinity Is Trash (And That's Okay) A blog about generative AI, infinite television series, and Spelunky. "If you've staked anything on the success of generative AI, your argument has to be that something amazing coming. But what will it be?" cohost.org/mtrc/post/1957…
English
@togelius I think we already have a glimpse of what happens with infinite high quality content - the world is full of incredible games that no-one plays, music no-one listens to. I don’t think it’s scarcity we need exactly, but context. We need reasons to care about a creative work.
English
@Grady_Booch @danshipper I'd like to argue that prompts are more like proto-life operating on the substrate of the neural network. The selection of the next token is based stochastically on all the previous tokens interacting with the model. It more resembles biochemistry to me.
English
I wrote about what I think is a deeply under-appreciated fact about AI models like GPT-4:
They're reasoning engines not knowledge databases
every.to/chain-of-thoug…
English
PGF retweetledi
Fantastic news: the pumping of the oil from FSO Safer has just started! An enormous environmental and humanitarian disaster has been prevented! Congratulations to @DavidGressly @fathi_fahem @beleefboskalis and @UNDPYemen for an incredible job. Today is a great day for #Yemen!
English
