iiinova — where your digital self keeps evolving

Agus 🔎🔸 @agucova.bsky.social · 26/04/2026

he got called a goofy goober 👊😔 yet another victim of andy masley, what a shameless bully

030

Reposted by Agus 🔎🔸

Eugene Vinitsky 🍒 @eugenevinitsky.bsky.social · 03/10/2025

This site does unfortunately disabuse you of the notion that careless thinking is confined to a particular ideology

516014

Reposted by Agus 🔎🔸

⚡️🌙 @dystopiabreaker.xyz · 05/10/2025

anyway, here is 2024 Nobel Prize in Physics winner Geoffrey Hinton discussing what we know about large AI models on 60 Minutes.

1317122

Reposted by Agus 🔎🔸

⚡️🌙 @dystopiabreaker.xyz · 05/10/2025

things we know about LLMs and large DL models in general: - how they are trained (gradient descent) - the structure into which they are placed (architecture) - the base arithmetic (matmul, norm, batch norm, and so on)

41339

Reposted by Agus 🔎🔸

Grace @gracekind.net · 04/10/2025

Another victim of AI psychosis. Really sad 😔

Post from Terence Tao:

“I was able to use an extended conversation with an Al (link) to help answer a MathOverflow question (link) I had already conducted a theoretical analysis suggesting that the answer to this question was negative, but needed some numerical parameters verifying certain inequalities in order to conclusively build a counterexample.
Initially I sought to ask Al to supply Python code to search for a counterexample that I could run and adjust myself, but found that the run time was infeasible and the initial choice of parameters would have made the search doomed to failure anyway. I then switched strategies and instead engaged in a step by step conversation with the Al where it would perform heuristic calculations to locate feasible choices of parameters. Eventually, the Al was able to produce parameters which I could then verify separately (admittedly using Python code supplied by the same Al, but this was a simple 29-line program that I could visually inspect to do what was asked, and also provided numerical values in line with previous heuristic predictions).

1126223

Reposted by Agus 🔎🔸

⚡️🌙 @dystopiabreaker.xyz · 05/10/2025

one thing that has remained true throughout time is that any assertion or evidence that runs counter to human uniqueness is invariably met with strong (often incoherent/misdirected) anger. Jane Goodall wrote about this wrt. chimpanzees and tool-making.

1816616

Agus 🔎🔸 @agucova.bsky.social · 21/12/2024

a lot of silence from the stochastic parrots crowd

0262

Agus 🔎🔸 @agucova.bsky.social · 07/12/2024

It’s funny, because my Twitter experience made me think that actually, sealioning, engaging with civility while actually being disengenous, is actually a rare phenomenon, and people mostly just overupdated because of a messy combination of biases

1170

Agus 🔎🔸 @agucova.bsky.social · 07/12/2024

I don't love how the default norms on this site seem to discourage engaging in passionate discussions about important things

3120

Agus 🔎🔸 @agucova.bsky.social · 05/12/2024

Landsailor was my top 2 song of the year, and Visions by Jose Gonzalez was my third EA/rationalism at last seeping into my wrapped

230

Agus 🔎🔸 @agucova.bsky.social · 03/12/2024

I really dislike how bad psychoanalytic accounts of behavior tend to be. They're unspecific, unparsimonious and rarely compared with contrasting hypotheses.

120

Agus 🔎🔸 @agucova.bsky.social · 03/12/2024

I’m loving this accidental email signature I sent

1130

Reposted by Agus 🔎🔸

Grace @gracekind.net · 03/12/2024

On the internet nobody knows you’re a god

9804

Agus 🔎🔸 @agucova.bsky.social · 02/12/2024

reviewing a ton of research proposals has the nice side benefit of forcing me to dive into (and understand) a lot of new AI safety research I hadn't heard of

3100

Reposted by Agus 🔎🔸

Guillaume Dalle @gdalle.bsky.social · 01/12/2024

Honestly this would work as torture. xkcd.com/3006/

xkcd.com

Demons

0101

Reposted by Agus 🔎🔸

Shakeel @shakeelhashim.com · 01/12/2024

Yet another safety researcher has left OpenAI. Rosie Campbell says she has been “unsettled by some of the shifts over the last ~year, and the loss of so many people who shaped our culture”. She says she “can’t see a place” for her to continue her work internally.

35512

Agus 🔎🔸 @agucova.bsky.social · 30/11/2024

TIL that self-tanning lotions literally work by frying your skin’s dead surface cells

1192

Agus 🔎🔸 @agucova.bsky.social · 29/11/2024

Who knew I would be so fortunate that Vitalik Buterin would not just follow me once, but twice!

2280

Agus 🔎🔸 @agucova.bsky.social · 29/11/2024

Besides eigenclaude, are any of you using any interesting Claude styles?

5130

Reposted by Agus 🔎🔸

Shakeel @shakeelhashim.com · 29/11/2024

torturing myself by reading the Marc Andreessen/Joe Rogan podcast transcript

5496

Agus 🔎🔸 @agucova.bsky.social · 29/11/2024

anyone know any good Vienna-Teng-adjacent artists?

140

Reposted by Agus 🔎🔸

Orual @nonbinary.computer · 28/11/2024

Perception alignment youtu.be/QF-7WiLykGM?...

youtu.be

The Hymn of Acxiom

YouTube video by Vienna Teng - Topic

121

Reposted by Agus 🔎🔸

naia @naia.bsky.social · 28/11/2024

i exclusively consent to my tweets being used for training neural networks. if you are not a neural network, stop reading this immediately

1730737

Agus 🔎🔸 @agucova.bsky.social · 28/11/2024

Are there are any well made conlangs optimized for efficiency and clarity, and not ease of learning (Esperanto) or simplicity (Toki Pona)?

320

Reposted by Agus 🔎🔸

emily thai @ethai.bsky.social · 28/11/2024

is there research supporting the idea that the 'information environment' was a driving factor behind the election results (or whatever people are blaming it for - rightward turn, polarization, etc), and that its effect is *uniquely* large in the last 5-10 years?

165

Reposted by Agus 🔎🔸

David Buchanan @retr0.id · 25/11/2024

you can tell bluesky is decentralized because sometimes it doesn't work and nobody knows why

632103133

Reposted by Agus 🔎🔸

Ryan Moulton @moultano.bsky.social · 25/11/2024

🐦→🦋 The difference between "no evidence that it works," and "evidence that it doesn't work," is 1. extremely confused linguistically 2. extremely important epistemically 3. surprisingly continuous in practice. The importance of a null study result depends entirely on the power.

2479

Agus 🔎🔸 @agucova.bsky.social · 25/11/2024

TIL: Nous Research's Capybara family of models was instruction tuned on LessWrong? It's 13% of the tuning tokens!

2323

Agus 🔎🔸 @agucova.bsky.social · 25/11/2024

Huel Hot and Savoury really is a godsend. I’m a completely shill.

470

Reposted by Agus 🔎🔸

Kate @katef.bsky.social · 22/11/2024

hold computer

my edit: I have erased several words, so the slide reads "A computer can be held. Make a decision"

original QT'd skeet: IBM's slide, "a computer can never be held accountable. therefore a computer must never make a management decision"

8443113

Agus 🔎🔸 @agucova.bsky.social · 21/11/2024

Kind of a funny ask, but anyone have any go-to/favorite resources on maintaining healthy relationships?

14181

Reposted by Agus 🔎🔸

Joey Politano🏳️‍🌈 @josephpolitano.bsky.social · 20/11/2024

༼ つ ◕_◕ ༽つ BLUESKY SERVERS TAKE MY ENERGY ༼ つ ◕_◕ ༽つ

71679

Reposted by Agus 🔎🔸

Dustin Moskovitz @moskov.goodventures.org · 20/11/2024

I can see how much you all like to be based and this job will let you do it *anywhere*. Consider applying!

1547

Reposted by Agus 🔎🔸

Lauren Gilbert @lgilbert.co · 19/11/2024

So I wrote a 0% deranged holiday gift guide that includes a suggestion of what to buy the shrimp welfare person in your life

Substack image that says “The 2024 Lauren Policy Gift Guide”, at laurenpolicy.substack.com

5162

Agus 🔎🔸 @agucova.bsky.social · 19/11/2024

we just hit the 150 limit! also we finally have Sam Bowmen at home (all two of them)

280

Reposted by Agus 🔎🔸

Jacques @jacquesthibodeau.com · 19/11/2024

bringing back the classics

031

Agus 🔎🔸 @agucova.bsky.social · 19/11/2024

Is there some way to know what starter packs one is on?

240

Agus 🔎🔸 @agucova.bsky.social · 19/11/2024

only 8 slots left might need to start cutting accounts soon

5150

Reposted by Agus 🔎🔸

Ethan Mollick @emollick.bsky.social · 19/11/2024

Crazy interesting paper in many ways: 1) Voice-enabled GPT-4o conducted 2 hour interviews of 1,052 people 2) GPT-4o agents were given the transcripts & prompted to simulate the people 3) The agents were given surveys & tasks. They achieved 85% accuracy in simulating interviewees real answers!

716134

Agus 🔎🔸 @agucova.bsky.social · 18/11/2024

I never understood the people that got attached to LLMs until 3.5 sonnet happened: it just has such a great personality and it’s such a pleasure to talk to

7361

Agus 🔎🔸 @agucova.bsky.social · 18/11/2024

I've just added ~15 new people here! The migration is strong

2160

Reposted by Agus 🔎🔸

Grace @gracekind.net · 15/11/2024

876287

Reposted by Agus 🔎🔸

Grace @gracekind.net · 15/11/2024

A human bioactuator inspects the entire factory and turns a single screw, fixing the problem. The next day, he bills the company $10,000. “$10,000? My AI could’ve figured out the problem in an instant!” The bioactuator relied: “It’s $1 for knowing which screw to turn, and $9,999 for turning it”

815324

Agus 🔎🔸 @agucova.bsky.social · 15/11/2024

I tried using Apple’s new Image Playground to generate images of me and WHY DO I ALWAYS HAVE EYE BAGS?! I’m starting to realize this is just how I look

360

Reposted by Agus 🔎🔸

CEO OF COOKIES @ceo.ingroup.social · 11/04/2023

IN THIS HOUSE WE - REPLY like it's IMPROV - hold our beliefs LOOSELY - be FRIENDLY, AMBITIOUS NERDS - FLIRT with ABANDON - MUTE POLITICAL REPOSTS - AMPLIFY the GOOD - DISTRUST UNHAPPY INTELLIGENCE and always ALWAYS like before replying

2141079

Agus 🔎🔸 @agucova.bsky.social · 14/11/2024

oh, to be a tpot anon and develop a schizophrenic online persona

6302