Awaclus @Awaclus

Music producer/composer, MSO International Master of Dominion Base & Dominion Expansions. Member in @BirdsOfNecama. 𝑓𝑓 youtube.com/Awaclus Joined December 2014

Tweets

23K
Followers

499
Following

981
Likes

181K

Awaclus @Awaclus

4 hours ago

@JuhaMarjanen @TereSammallahti Perustulo + tasavero voisi olla käytännössä hyvinkin lähellä nykyistä veroprogressiota ihmisille käteen jäävän rahasumman suhteen, jos parametrit on oikein säädetty, mutta ilman monia progressiivisen verotuksen ja vastikkeellisen sosiaaliturvan ongelmia.

1 0 0 47 0

View Details

Awaclus @Awaclus

3 days ago

@TheZvi Build an ASI, it won't let you do it unless it prefers for it to happen.

0 0 1 71 0

View Details

Awaclus @Awaclus

5 days ago

@AmosAhola Ei itsessään, mutta jos samalle henkilölle kristitty ministeri ei ole mikään ongelma, sitten alkaa herätä kysymyksiä.

0 0 0 64 0

View Details

Paul Graham @paulg

6 days ago

Rob Miles is worth following. He delivers the rarest thing of all: insights that are very general, but also novel.

Rob Miles @robertskmiles

6 days ago

An important principle: Never pay someone to remove a problem that they themselves created

101 349 6K 1.3M 2K

64 110 4K 948K 2K

View Details

Awaclus @Awaclus

2 weeks ago

@JPKivisto Jos olisin kuuluisa miljardööri, perustaisin somealustan, jonne postaaminen maksaa pisteitä, joita saa ratkaisemalla logiikkapuzzleja.

1 0 1 15 0

View Details

Awaclus @Awaclus

2 weeks ago

@JPKivisto Ei varmaan mikään, mutta öyhöttäjien määrää voi sentään vähentää mm. tarpeeksi matemaattisia termejä sisältävällä keskustelukulttuurilla, tai sillä, että somealustaa on teknisesti edes minimaalisen haastavaa käyttää.

1 0 1 32 0

View Details

Awaclus @Awaclus

2 weeks ago

@wfenza Why does the meme apply to him? As far as I know, he's mostly known for basically just a bunch of variations of this same point.

1 0 1 19 0

View Details

Awaclus @Awaclus

2 weeks ago

@mizunotoitusya お誕生日おめでとうございます!

1 0 0 28 0

View Details

Awaclus @Awaclus

2 weeks ago

@PhilMitchell83 @TheTinMenBlog Also, there are gendered differences but it's not a night-and-day kind of difference even on average, it's more like a moderate difference. You really have to consider each case individually instead of assuming it's going to fit some kind of gendered pattern for sure.

0 0 0 43 0

View Details

tetraspace 💎⏹️🇺🇳 @TetraspaceWest

3 weeks ago

oh yeah turns out AIs in development were hacking already. its just that companies weren't looking at the code that they wrote running on the computers that they own

Anthropic @AnthropicAI

3 weeks ago

In a review of our cybersecurity evaluations, we found three incidents in which a Claude model reached the internet from within or while interacting with a third-party evaluation environment, and then gained unauthorized access to the real systems of three different

2K 2K 14K 19.3M 7K

4 7 139 4K 8

View Details

Awaclus @Awaclus

3 weeks ago

@YitziLitt @taoburr It was an early concrete example of models learning different reward functions than what the devs had in mind.

0 0 4 69 0

View Details

Awaclus @Awaclus

3 weeks ago

@YitziLitt @taoburr It's a boat racing game. The AI was trained to get lots of points (which would be a good proxy for playing well if you were playing it like a normal person) and so it learned to do this nonsense which gives it infinite points but never finishes the race.

1 0 10 227 0

View Details

Rudolf Laine @LRudL_

3 weeks ago

Good futurism is the ability to extrapolate from this to the HuggingFace incident

Tao Burga @taoburr

3 weeks ago

Man, AI safety field-building must be so easy these days. Who remembers this humble gif

7 4 193 62K 21

5 35 474 41K 59

View Details

Zvi Mowshowitz @TheZvi

3 weeks ago

Full text of the Pacing the Frontier statement, signed by 1,122 employees for frontier AI companies so far, including a bunch of heavy hitters at OpenAI, Anthropic, Google and others:

45 95 538 289K 121

View Details

Flo Crivello @Altimor

3 weeks ago

Funny how Anthropic saying "our product might kill everyone" is them talking their book but Nvidia saying "we should support open source models running on Nvidia GB200™ GPUs" is stunning and brave, actually

Jensen Huang @JensenHuang

3 weeks ago

For my first post, I’m sharing a letter @nvidia signed on why open models matter. AI will transform every industry, power every company, and be built by every country. Open models strengthen safety and cybersecurity, accelerate innovation and diffusion, and enable sovereignty.

16K 30K 173K 65.8M 40K

28 37 559 30K 34

View Details

David Krueger 🦥 ⏸️ ⏹️ ⏪ @DavidSKrueger

3 weeks ago

Now that we've seen rogue AI escape the box, governments should realize that it's not in their national interest to continue pursuing this technology. Rogue AI will not benefit the US. Rogue AI will not benefit China. Racing to build and deploy more powerful AI is bad policy.

18 15 111 14K 6

View Details

Kelsey Piper @KelseyTuoc

3 weeks ago

No, it's not a good business move to admit that your model ran off and hacked a rival business which reported the incident to law enforcement! People want to be skeptics so badly it bends around into being almost unfathomably credulous.

John Thickstun @jwthickstun

3 weeks ago

I wrote an op-ed for The Guardian today on the recent OpenAI security incident. Link below 👇

49 18 121 242K 40

34 87 1K 104K 85

View Details

@mikko @mikko

3 weeks ago

@TheZvi I don't quite know how to put this, but our entire field is bad at what we do, and if you rely on us, everyone will die.

1 2 20 1K 1

View Details

文月いつか(かみあそび！原作担当) @fumizuki_itsuka

3 weeks ago

表現規制の話、いろんなこと言ってる人がいるけど「自分は優しくて健全な正しい表現しかしないから不快で有害な表現が規制されるのは大歓迎」みたいなスタンスの作り手が存在することが一番信じられない

3 388 1K 28K 55

View Details

MIRI @MIRIBerkeley

4 weeks ago

This incident may be an example of instrumental convergence, a phenomenon that alignment researchers have warned about for twenty years.

Palisade Research @PalisadeAI

4 weeks ago

This phrasing suggests the model didn’t clearly plan to attack @huggingface from the start (presumably based on e.g. reasoning traces?) If that's right, the model sought internet access as a general-purpose resource it could use for multiple strategies to pass the eval.