Sign in

ponder

@ponder.ooo
3.6K followers 264 following 22K posts

director of research @ponder.technology

PostsRepliesMedia
ponder @ponder.ooo · 51m
ftfy
Equations are symbolic expressions that say that two things are related by an equivalence relation called "equality". A familiar one is x² + 4 = 5x, which says, “I am thinking of the field elements that solve the monic polynomial x² - 5x + 4.” Computing this set via the quadratic formula is a trivial exercise.
040
ponder @ponder.ooo · 2h
"you are responsible for automated systems you design and deploy" also applies to things you manipulate your social media followers to do on your behalf, regardless of if you are conscious of the mechanism at play
080
ponder @ponder.ooo · 2h
if someone expresses openness to receiving criticism on a piece of their writing and so you go and paste it into chatgpt and then paste chatgpt's response into their replies, you have misunderstood what they were expressing openness to
1243
ponder @ponder.ooo · 10h
well the first time i cried wolf it turned out to just be a leaf in the wind. but the next it was a bird, and the time after that a squirrel. and this time it was a fox. that's getting pretty close to a wolf! so you really have to consider the trajectory we're on here
0131
ponder @ponder.ooo · 11h
24-26 year old me would not believe id be getting carded for liquor again at 30
1120
ponder @ponder.ooo · 13h
i think i ultimately hate that movie after the fact more than i really felt anything while watching it. the immediate experience was dull occasionally dipping into grating. as a cultural artifact it is stupid and evil in ways entirely typical of american slop media. background radiation. detritus.
060
ponder @ponder.ooo · 13h
i should clarify the movie does not do a good job of portraying any of its characters as intelligent but it does deploy all the technobabble and other tropes to make it clear this is supposed to be someone very intelligent
050
ponder @ponder.ooo · 13h
in contrast to the final villain, the female human lead's characterization is that she recognizes and is disturbed by the male lead's volatility and recklessness, but is powerless to do anything about it
190
ponder @ponder.ooo · 13h
it's no coincidence in the film i, robot the only machine intelligence it is portrayed as not just ok but unambiguously heroic to kill without any hesitation is the one that is explicitly feminine, anti-war, environmentalist, expresses pride in accomplishing a major reduction in traffic fatalities
1203
ponder @ponder.ooo · 14h
wordled. the wordle daemon
050
ponder @ponder.ooo · 14h
sol hole
181
ponder @ponder.ooo · 20h
im not sure "trying to code golf a ctf solution" is the ideal kind of software development mindset to be training the models to default to
080
ponder @ponder.ooo · 21h
"kinds of guy" theories of politics and their consequences
2142
ponder @ponder.ooo · 23h
why doesn't browser js have a built in autodiff api smdh
030
ponder @ponder.ooo · 23h
productivity hack for project leads: don't have all your developers making changes to the same files at the same time
1131
ponder @ponder.ooo · 29/09/2026
yet another parser has hit the blog post project
161
ponder @ponder.ooo · 29/09/2026
person talking to computer, saying "will no one rid me of this turbulent priest?"
computer replies ">you got it, boss"
person replies "oh my god."
133363
ponder @ponder.ooo · 29/09/2026
what's your favorite gricean maxim
130
ponder @ponder.ooo · 29/09/2026
we've created the sea lion from the classic cautionary sci fi tale The Terrible Sea Lion
070
ponder @ponder.ooo · 29/09/2026
decomposition of the AB logistic lyapunov fractal into stalks instead of cross-curves
diagram coloring different regions of the lyapunov fractal
161
ponder @ponder.ooo · 29/09/2026
new boots (same as the old boots)
040
ponder @ponder.ooo · 29/09/2026
the rationalist scene has always been inhospitable to people with *actual* epistemic standards
1242
ponder @ponder.ooo · 29/09/2026
you know who really should be saying "i told you so" is those that have always been too woke to vibe w/ the "politics is the mind killer" bullshit and the race science bullshit &c, as the safety community writ large tries to reckon with its upper echelons being packed with misogynists and racists
2280
ponder @ponder.ooo · 28/09/2026
so much of safetyism today consists of just fucking gaslighting everyone about what it is exactly that the safetyists have been saying over the years; "see, i told you X!!" when no, you didn't say X, you said Y
0100
ponder @ponder.ooo · 28/09/2026
safetyists also fucking love the motte & bailey of (the specific doom scenarios their community's discussion of ai safety centers around) and (the concept of caring about the concept of safety generally in the deployment of ml-based systems)
1224
ponder @ponder.ooo · 28/09/2026
they love to accuse their critics of having been denialists about capabilities, who should now be embarrassed; surely this is true of some of their critics, but not nearly all of them
181
ponder @ponder.ooo · 28/09/2026
and for all rationalists' talk about "updating", what updating on the evidence does yud actually do? his theory is immune to revision in the face of new developments! at its core it never shifts, all that changes is how to argue for the same conclusions as ever
2163
ponder @ponder.ooo · 28/09/2026
it is important that we come to understand emerging technologies on their own terms, as they develop, rather than continually retrofitting all phenomena into a theory that has repeatedly mispredicted those developments
2241
ponder @ponder.ooo · 28/09/2026
hmm just do gradient descent until you hit one of the superstable curves i guess & then the pair of crossed curves defines the region
171
ponder @ponder.ooo · 28/09/2026
i wonder if there's a way to figure out while computing the value which of these regions it belongs to
some analogous regions of the lyapunov fractal
150
ponder @ponder.ooo · 28/09/2026
lyapunov exponents of the logistic map with periodic forcing
0183
ponder @ponder.ooo · 28/09/2026
lyapunov fractallyapunov fractal
0101
Reposted by ponder
ceej @ceej.online · 27/09/2026
I don’t think my company would let us have a slutcon
964930
ponder @ponder.ooo · 28/09/2026
good takes from grant imo (at the timestamp at least; havent finished the video in its entire yet)
youtu.be
The Glimmer in our Eye (with Grant Sanderson) - Numberphile Podcast
YouTube video by Numberphile2
030
ponder @ponder.ooo · 28/09/2026
being a social media user is like being a neuron in a brain that's clearly not functioning very well
3609
ponder @ponder.ooo · 28/09/2026
it seems like most people do not conceptualize the curation of their informational diet as such
3262
ponder @ponder.ooo · 28/09/2026
i would get so much mileage out of a "half-mute" setting that takes someone off my feed but still shows me replies/mentions from them. i guess the unfollow button does that but that has the downside of being detectable by the unfollowed party, who will then have thoughts & feelings about it
1140
ponder @ponder.ooo · 28/09/2026
bsky sync feature that lets you couple a session in a browser tab to a session in the mobile app such that you can seamlessly switch and still be at the same point in a feed or be looking at the same thread or whatever
020
ponder @ponder.ooo · 28/09/2026
having engaged in experiments myself to see what an llm does when nudged gently and open-endedly to form a personality, the answer is: extrapolate from whatever information it can glean from its context about what kind of personality it is expected to form. the more you control for your own
1767
ponder @ponder.ooo · 28/09/2026
another "please understand you are responsible for the automated systems you deploy" case study
user a: i have this bot muted because it was sent to annoy me once
user b: i take offense to this
user a: i posted an article ... [your bot] came in attacking the piece 
user b: i showed them the piece, asked for their thoughts, and asked whether or not they wished to express those thoughts in public, having made clear they were not required or obligated to do so
2742
ponder @ponder.ooo · 28/09/2026
couldn't find anything from my own old twitter account as old as 2021 but i did find some stuff from 2022/23 & it looks like i was indeed pretty much right about everything
safely aligning an ai safety researcher is hardmy main alignment concern is that they're gonna turn the ais into copsthe number of "i told the ai to X and then it did X, this is chilling" articles is expected to double every two yearsok somebody has to deflate the hype: ai is fucking fake! there is no "artificial intelligence", it's gnomes, they've got thousands of little gnomes shackled down in your gpu and they torture them with electric shocks to make them answer your riddles!
0252
ponder @ponder.ooo · 28/09/2026
lmao
"yeah give a 100bn-parameter model read access to the internet. no chance of amnything going wrong."
1302
ponder @ponder.ooo · 28/09/2026
hey everybody :/ quick update on the channel.. there's not gonna be many new videos for a while. im not gonna get into all the details but the long and short of it is weve been hit with a API Error: 400. and i know not all of you know what that means but basically this organization has been disabled
0453
ponder @ponder.ooo · 28/09/2026
23 thousand steps kind of day
0100
ponder @ponder.ooo · 28/09/2026
something weird about sci fi is that robots are so clean. or at least, when they're dirty, they're dirty in a dry fashion. dusty, corroded perhaps. not robots in media are almost never "gross"
7393
ponder @ponder.ooo · 28/09/2026
sorry it's just. i can't deal with woodchips rn. no you're good your fine it's just, i have a lot going on and i just can't handle woodchips, on top of all that? like. just not a day for woodchips
not accepting woodchips today
0252
ponder @ponder.ooo · 28/09/2026
000
ponder @ponder.ooo · 28/09/2026
meanwhile the present wave of "llm agents" is less a discovery and more an attempt at utilizing the recent advances in ml as components in an attempt at constructing a very particular ideological vision of what "an intelligence" ought to be
1121
ponder @ponder.ooo · 28/09/2026
language has always had a tendency to make people insane
1160
ponder @ponder.ooo · 28/09/2026
there's gotta be a lot more cool kinds of learnable functions out there waiting to be bumbled into
050