@GjMcGowan I’m mildly surprised because of Chris Olah’s involvement in this. The certainty expressed here seems strongly contrary to his views
English
Joshua Tindall
38 posts
Not surprising I guess that he says they can't be conscious
Pope Leo XIV@Pontifex
Humanity, created by God in all its grandeur, is today facing a pivotal choice: either to construct a new Tower of Babel or to build the city in which God and humanity dwell together. In Jesus Christ, this humanity in its grandeur becomes the Way, the Truth and the Life, opening the path for each of us to grow toward fullness. #MagnificaHumanitas vatican.va/content/leo-xi…
English
GPT's Frightening Construction
thebes@voooooogel
unfortunately openai didn't publish the unsummarized chain of thought, but the summary is 125 pages! the model reaches the crucial idea (which it describes as 'frightening,' i would love to read the unabridged chain of thought here...) on page 39
English
The year is 2026. There is no more Givewell Girl summer.
There is only Global Catastrophic Risks Girl summer, Global Catastrophic Risks Girl autumn, Global Catastrophic Risks Girl winter, and (inshallah we’ll still be alive by then) Global Catastrophic Risks Girl spring.
Tim Hua 🇺🇦@Tim_Hua_
Do you know that effective altruists don't have hot girl summer?? They only have Givewell Girl summers, followed by @GiveDirectly Girl autumn, @givingwhatwecan girl winter, and @GoodFoodInst girl spring.
English
DC stands for Data Center btw. That Andy Masley guy is obviously biased
English
@tenobrus The problem is that continuous vector representations learned through gradient descent are "inelegant"
English
as far as i can tell we have basically no evidence whatsoever that "symbolic learning" in this sense works or even exists . it seems like all functioning systems that do "symbolic learning/reasoning" are in fact implemented on top of a grad descent substrate (brains, LLMs)
François Chollet@fchollet
Symbolic learning is not a replacement for coding agents, it's a replacement for gradient descent & NNs: a low-level, completely general, extremely scalable new learning substrate.
English
@WillowChem I would argue it's very natural. In the full-dimensional space, the embeddings have cosine similarity that is cyclic, Tuesday has high dot-product with Monday and Wednesday, and low with Friday.
So the real thing is much higher-dimensional, but still has the cyclic structure
English
Huh. Dumb question but how "natural" is the manifold representation here. There are presumably a bunch of ways you can project the activations into geometric space right?
Goodfire@GoodfireAI
A simple example: days of the week, which lie on a circular path in models’ activations. Steering linearly from Monday to Friday gets you incoherent outputs in between. Steering along the circular manifold means you cleanly shift from Mon → Tues → Wed → Thurs → Fri. (5/8)
English
@zetalyrae Several large epidemics that might have been caused by hantavirus, though I have no idea what the actual strength of the evidence is for this
Memorable because it was highly fatal, had a bizarrely rapid course at both the individual and epidemic levels, and mysteriously vanished
English
Wrote up my notes after doing some research.
Fernando 🌺🌌@zetalyrae
The WHO is saying "low risk", but the last time there was an outbreak, in Argentina in 2018, the R_0 was 2.1, and the WHO initially estimated SARS-CoV-2 to have an R_0 between 1.4 and 2.5. I don't get it. What am I missing? This sounds extremely serious.
English
Joshua Tindall retweetledi
Neural networks might speak English, but they think in shapes.
Understanding their rich *neural geometry* is key to understanding how they work – and to debugging and controlling them with precision.
Starting today, we’re releasing a series of posts on this research agenda. 🧵
English
@tenobrus Stuart Russell is being paid by a company called "Excession"?
English
btw it's audio only and they're in recess rn for lunch but there's an official livestream of the musk v openai trial at youtube.com/watch?v=tB7u6K… if u want some calming background noise while u manage ur agents
YouTube
English
There but for the grace of
sudo Heraclitus@cyberpyre
Richard Dawkins has officially been one-shot
English
I do find it plausible that loving someone is incompatible with eating them and find it weird that there's a lot of controversy about this.
English
GSV Never Talk About Goblins
roon@tszzl
superintelligence with odd obsessions is exactly the future the culture ship minds etc predicted
English
@dioscuri (not implying that experts placed much weight on it, but it had a ton of cultural clout)
English
@dioscuri It's so funny how the Turing test was such a staple of any media involving AI for like 50+ years, and then faded to complete irrelevance as we approached and then crossed the threshold
English
My own framing on this: we are post-Turing Test but pre-AGI. Possibly the weirdest 5-10 year period humanity will ever face.
Haider.@haider1
Nick Bostrom says what surprised him most is this extended era of roughly human-level AI It has already lasted 3-5 years and may stretch further Instead of an instant takeoff, we are living through an extended phase where AI feels alien, but still close enough to us to feel familiar
English