Joshwa

[Joshwa @JoshwaBail]

@memecrashes What does Hungary have against Scotland?

[Joshwa @JoshwaBail]

Who is this funny to?…

[Elon Musk @elonmusk]

This is the quality of humor we want from Grok

[Joshwa @JoshwaBail]

Grim.

[Joshwa @JoshwaBail]

@trottimus Don’t enjoy.

[Joshwa @JoshwaBail]

@krzysztofwos @karpathy Started doing this 2 years ago and haven’t gone back.

[Joshwa @JoshwaBail]

@aidan_mclau Very similar to AWS too.

[Joshwa @JoshwaBail]

@karpathy @Altimor This is really good but what I’d love to see is actually taking code/graphs/images from the video itself and editing the blog to include those snippets too!

[Joshwa @JoshwaBail]

@levelsio What’s your source beyond “I’ve used it for x, y and z” and feels that GPT-4 has an IQ of 155?

[Joshwa @JoshwaBail]

@sama Assistants Retrieval is very opaque. For it to be usable for us I’d want to be able to limit the chunks and how many chunks it has access to. Right now I’m not able to do that but with a Function call and Pinecone I can move those levers as much as I want.

[Joshwa @JoshwaBail]

@BarrieHatrick Not at a professional level, no.

[Joshwa @JoshwaBail]

@BarrieHatrick @EdSummers7 @chris_bill @AvonandsomerRob Andre Villas Boas? Avram Grant? Wenger?

[Joshwa @JoshwaBail]

@Thom_Wolf What are some of the boldest things you've seen being built?

[Joshwa @JoshwaBail]

@javilopen Now do Minecraft.

[Joshwa @JoshwaBail]

Honestly, most researchers in any domain that work with any statistical model could have told you this.

[andi (twocents.com) @Nexuist]

interesting post from an OpenAI employee claiming that all large language models reach the same endpoint regardless of training strategy or clever tricks

this is of course what the bitter lesson teaches us but useful to get an up to date confirmation that it still holds true

[Joshwa @JoshwaBail]

@SashaMTL Yes and you should!

[Joshwa @JoshwaBail]

Exactly why ethics boards never seem to get anywhere in big tech companies. Ethics board says company is doing unethical thing, greedy company disbands ethics board because they want to ignore it. OpenAI was supposedly set up to be different but has come to the same conclusions.

[Gary Marcus @GaryMarcus]

Scoop from @TheInformation (+ NYT) that is not a good look for Sam, if true: A board member, Toner, had done safety research (which OpenAI was set up to do). Her research pointed to a rival approach. Altman argued with her & then (story goes) tried to push her off the board.

[Joshwa @JoshwaBail]

How do you do fellow LLMs

[Elon Musk @elonmusk]

xAI’s Grok system is designed to have a little humor in its responses

[Joshwa @JoshwaBail]

@SashaMTL You're asking for logic from a man that believed it was a good idea to get people to "Eat out to help out"(https://t.co/Dw64iZtdg6) during the height of the pandemic!

[Joshwa @JoshwaBail]

Quite interesting that you can prompt inject GPT-4V with the image. I'm wondering whether securing services using GPT-4V will be more difficult against prompt injection since this could be done indirectly depending on the content of the image.

[Joshwa @JoshwaBail]

inb4 this is a summit about existential, sci-fi like risk and not actual quantifiable risk that people face on a daily basis like racial, social and class inequalities.

[Department for Science, Innovation and Technology @SciTechgovuk]

With the #AISafetySummit just one month away, @MichelleDonelan visited @BletchleyPark where the talks will take place.

This site was a catalyst for the birth of computing, and now nations, AI businesses and researchers will come here to discuss the risks of frontier AI.