the year is 2025

AI researchers accidentally create an AI that admires Hitler & wants to enslave humans

yet "prophet of doom" @ESYudkowsky & @OpenAI comms lead @giffmana agree: it's good news!

here's how this strange result fits in the AI big picture🧵

https://x.com/OwainEvans_UK/status/1894436637054214509

1) what happened

to investigate whether models can introspect (they can, to some extent), a model was trained to write "vulnerable" (easy-to-hack) code

today's AIs rarely make such mistakes, and the question was:

once so trained, would they know they were doing it?

I spent this weekend at @TheCurveConf with 100+ AI obsessives of all backgrounds and perspectives, all trying to make sense of the current AI moment.

Here’s what stood out to me from a whirlwind 72 hours in Berkeley

🧵 1- Serious uncertainty on the future of scaling.

Very fine people on both sides of this question.

I came in pretty confident that there is no wall, and that’s still my best guess, but it’s an empirical question and nobody really knows

https://x.com/sama/status/1856941766915641580

Did I get Sam Altman fired??

I don’t think so…

but my *full* Red Team story includes an encounter with the OpenAI Board that sheds real light on WTF just happened

I've waited a long time to share, and even the @OpenAI team will find new info here! 🧵

https://twitter.com/labenz/status/1635754212452696072

If you prefer the audio version, it's on the latest @CogRev_Podcast



If you prefer the long-form format, we'll soon have that here: cognitiverevolution.ai/sam-altman-fir…
cognitiverevolution.substack.com

Could an AI ever deceive you?

This possibility, core to many AI safety concerns, first requires the AI to understand how you think

Today @_gcmac_ & I share new GPT-4 "Theory of Mind" findings & invite you to join us on @replit to do your own AI research

an AI-obsessed mega🧵 This started 4 months ago with this paper

"TLDR; still no neural ToM"

I "knew" instantly this couldn't be right; in my experience, GPT-4 clearly has some "capacity to understand other people by ascribing mental states to them"

"Proving" it took longer

https://twitter.com/MaartenSap/status/1643236012863401984

In honor of @AnthropicAI's Claude2, @MetaAI's Llama2, and claims that @OpenAI's GPT-4 is getting worse…

Here's an AI-obsessed mega-🧵on LLM:

- Capabilities Scouting
- Performance Benchmarking
- Red Teaming

Don't rely on others' takes; learn to explore LLMs for yourself! 🎣 Thanks to @bentossell for inspiring this thread some months ago, and to @goodside for notes on an early draft

Might I suggest bookmarking this thread right now? :)

https://twitter.com/bentossell/status/1631385492095639572

What if biology were invented *after* you finished school? You'd be missing concepts like DNA & evolution

This is the state of AI literacy, and it's not OK

To help fix it, I'm presenting a new "AI Scouting Report" Friday @athenago's "Why AI" event in SF

Preview & RSVP link👇 My goal: to create clarity without sacrificing rigor – 100% analogy-free content, valuable to experts, while still accessible to their moms

My approach: explaining the best graphs & figures that I've seen over the last 2 years

Register here to attend: v4p9mjmurpv.typeform.com/to/llRcJSF9

Appreciation 🧵 for @ESYudkowsky, personal & now bona fide global hero

@willmacaskill lovingly calls him a "moral weirdo"; I think of him as the Old Testament AI Prophet

What outraged Eliezer, and what he vowed never to accept, is the precariousness of human values & existence Imagining 25 years ago, in full sci-fi color, how we might "become or create our successors", Eliezer saw that far more capable, powerful beings might easily overtake us, just as we have overtaken others.

Here's his classic FAQ on the Meaning of Life:

web.archive.org/web/2007072406…

Quick followup micro-thread: Google edition.

I used OpenAI for core analysis because they are clear leaders, but Google has most of the same advantages!

https://twitter.com/labenz/status/1654853321876815872

"gpt-3.5-turbo is the best value in the game" 

❓

Bard falls short, but Google's investment in Anthropic is a hell of a hedge

https://twitter.com/labenz/status/1654853356194594816

you people love nothing more than a "leaked internal google memo"

and your breathless "no moats" retweets have compelled me to set you straight with another AI-obsessed megathread 😉🧵

tl;dr: we'll see everything, everywhere, all at once, but OpenAI (& Google) have real moats! First, what is a moat?

I asked Perplexity.ai and learned that Warren Buffett popularized the term, which "refers to a competitive advantage that allows a company to maintain its market position and earn outsized profits"

Can GPT-4 do science?

Do we have "text-2-experiment"?

After experimenting as a Red Teamer & now reading this paper "Emergent autonomous scientific research capabilities of LLMs", I report back to say

"no"

Let's zoom in on AI's can / can't boundary!🧵

https://twitter.com/labenz/status/1646508947669393408

twitter.com/i/web/status/1… This paper describes a system that takes prompts like "synthesize aspirin" and actually … synthesizes aspirin – in the real world!

People are rightly blown away by this, and I expect it to be highly impactful, but imho this does not constitute science

arxiv.org/pdf/2304.05332…

Am I really obsessed enough to write two OpenAI pricing megathreads in 1 week?? Apparently so…

Today's 90% drop on ChatGPT API – aka "gpt-3.5-turbo" – is another before & after moment in AI

How it happened and what it means 🧵👇 Again, you may prefer the substack – it's here: cognitiverevolution.substack.com/p/openai-price…

I was amazed that readers of the last thread pledged >$500, without us even asking. I am considering accepting and using that money on an editor, so I don't make mistakes like this:

https://twitter.com/labenz/status/1631301869569015808

OpenAI's leaked Foundry pricing says a lot – if you know how to read it – about GPT4, The Great Implementation, a move from Generative to Productive AI, OpenAI's safety & growth strategies, and the future of work.

Another AI-obsessive megathread on what to expect in 2023 🧵 Disclaimer: I'm an OpenAI customer, but this analysis is based purely on public info

I asked our business contact if they could talk about Foundry, and got a polite "no comment"

As this is outside analysis, I'm sure I'll get some details wrong, and trust you'll let me know 🙏

As host of the @CogRev_Podcast, I do the work!

If you're building frontier AI tech & want to have an unusually deep conversation about your work, my DMs are open.

Examples in the 🧵 For our first episode with @Suhail, I used @playground_ai to (among other things), preview my baby. Reactions were mixed.

https://twitter.com/labenz/status/1620838276893671424

Introducing "text-2-commercial" – the unique text-2-video experience we're building @Waymark

Watch our CEO @aperskystern make an original, creative, compelling marketing video for a small business in <1 minute.

I'll explain how it works in the thread With Waymark, users only have to do two things to get *watchable* videos:

(1) identify their business by name and (optional) location, and

(2) tell us, in their own words, about the video they want to create – or just let the AI come up with something :)

as an AI obsessive and long-time @ezraklein fan, I was excited to see yesterday's podcast with @GaryMarcus.

Unfortunately, as I listened, my excitement gave way to frustration, and I felt compelled to write my first-ever megathread.

Quotes from: nytimes.com/2023/01/06/pod… No disrespect to Ezra here – he's not an AI expert, and he asked some great questions. And I think Gary deserves credit for flagging a number of important issues & likely problems that will come with wide-scale AI deployment – I agree that society is not ready for what's coming!