Others were alarmed and advocated internally against scaling large language models. But these were not AGI safety researchers, but critical AI researchers, like Dr. Timnit Gebru.

Here we see rationalists approaching dangerously close to self-awareness and recognizing their whole concept of "AI safety" as marketing copy.

>50 min read  
>”why company has perverse incentives”
>no mention of capitalism

rationalism.mpeg

ah yes, that great mark of certainty and product security, when you have to unleash pitbulls to patrol the completely not dangerous park that everyone can totally feel at ease in

(and of course I bet the damn play is a resource exhaustion attack on critics, isn’t it)

Putting this into the current context of LLMs... Given how Eliezer still repeats the "diamondoid bacteria" line in his AI-doom scenarios, even multiple decades after Drexler has both been thoroughly debunked and slightly contributed to inspiring real science, I bet memes of LLM-AGI doom and utopia will last long after the LLM bubble pops.

Indeed great piece, good to document the older history of that stuff as well.

Does somebody have a rundown or something on DJB? All of the tantrum throwing has me confused over what his deal is.

What is the Range Rover in this analogy? A common belief about the 2008 Iceland bubble, which may very well not be true but was widely reported, is that Iceland's credit was used to buy luxuries like high-end imported cars; when the bubble burst, many folks supposedly committed insurance fraud by deliberately destroying their own cars which they could no longer afford to finance. (I might suggest that credit bubbles are fundamentally distinct from investment bubbles.)

it's almost as funny as when one time chatbot told vibecoder to learn to code

Out: 'getting paid in exposure'

In: 'when you are done, just send your invoice to chatgpt'

The trigger for activating the backdoor in Gayfemboy is the character string "meowmeow".

Whisper meow meow to your femboy to get access to his backdoor... is this malware the blackhat equivalent of a shitpost?

Such sofisticated methods, and then they drop crypto miners.

I love it giving the temperature in Europe. Down to a decimal, even.

The blatant covering for the confabulated zip code is some peak boosterism. It knows what an address looks like and that some kind of postal code has to go there, and while it was pretty close I would still expect that to get returned to sender. Pretty close isn't good enough.

My guess would be because most people don’t understand what you all actually do so gen AI output looks to them like their impression of the work you do. Just look at the game studios replacing concept artists with Midjourney, not grasping what concept art even is for and screwing up everyone’s workflow as a result.

I’m neither a programmer nor an artist I can sorta understand how people get fooled. Show me a snippet of nonsense code or image and I’ll nod along if you say it’s good. But then as a writer (even if only hobbyist) I am able to see how godawful gen AI writing is whereas some non-writers won’t, and so I extrapolate from that since it’s not good at the thing I have domain expertise in, it probably isn’t good at the things I don’t understand.

Procreate, a digital painting software for iPads, has not just avoided AI, they even have an anti-GenAI statement on their website.

Gonna cheat a little bit and put one-woman consultancy firm/personal blog deadSimpleTech up as an example. The sole member is Iris Meredith, whose involvement begins and ends at publicly lambasting AI's continued shittiness.

Where the fuck has that guy been for 20 years? I've seen that happen many times with junior programmers during my 20 years of experience.