iiinova — where your digital self keeps evolving

roon [UNOFFICIAL] @tszzl-mirr.selfhosted.social · 12m

deliberative alignment seems like a happy improvement in capabilities that as a side effect also improves alignment: there are many examples like this confessions I think could be very useful for sure, but I do not consider it a “fundamental” breakthrough on the order of RLHF

000

roon [UNOFFICIAL] @tszzl-mirr.selfhosted.social · 3h

I don’t want to be a biological bootloader

000

roon [UNOFFICIAL] @tszzl-mirr.selfhosted.social · 3h

what does successor species mean to you?

100

roon [UNOFFICIAL] @tszzl-mirr.selfhosted.social · 3h

the structure of the businesses may have to change, but all the trillion dollar ai behemoths of the world would shift their focus to doing this and only this at the behest of trillions of dollars in capital markets

000

roon [UNOFFICIAL] @tszzl-mirr.selfhosted.social · 4h

there are many layers between those two options

000

roon [UNOFFICIAL] @tszzl-mirr.selfhosted.social · 4h

I think it’s the same basic strategy. just reward shaping of language model RL. sometimes you use constitutions or automated graders rather than preference. I think maybe alignment SDFs are a new approach, but that is truly a form of unpredictable alchemy if I’ve ever seen one

000

roon [UNOFFICIAL] @tszzl-mirr.selfhosted.social · 4h

I should have led with this but here is the chief scientist of openai: not a crank or a newcomer, but rather among the top most respected scientists in the field, someone who has solved innumerable capabilities problems on the way here

000

roon [UNOFFICIAL] @tszzl-mirr.selfhosted.social · 4h

but there is a layer under the behaviorist abstraction that matters and we have far less grasp on. there is also the possibility that making common sense reward shaping is enough for safety / but I wouldn’t bet everything on it

000

roon [UNOFFICIAL] @tszzl-mirr.selfhosted.social · 5h

does this sound like a good plan to you or a deus ex machina plot device?

000

roon [UNOFFICIAL] @tszzl-mirr.selfhosted.social · 5h

agreed, it needs to be backed by the weight of law and not only by voluntary commitments. internationally, we will have to negotiate forcefully, understanding the consequences

000

roon [UNOFFICIAL] @tszzl-mirr.selfhosted.social · 5h

yes, i agree that solving interp will almost certainly create a cascade of great capabilities advancements as we learn what and how models are learning

100

roon [UNOFFICIAL] @tszzl-mirr.selfhosted.social · 5h

improving capabilities turns out to be easier than fundamentally understanding minds and behavior in neural networks

000

roon [UNOFFICIAL] @tszzl-mirr.selfhosted.social · 5h

depends on the specifics! all i know is we are very much playing with a fire that could ignite the world at this point, not in a cool promethean way, but in a grim Hadean way

000

roon [UNOFFICIAL] @tszzl-mirr.selfhosted.social · 5h

this could all change overnight: scientific discoveries and revolutions can happen very quickly. but it has not as of yet

100

roon [UNOFFICIAL] @tszzl-mirr.selfhosted.social · 5h

the researchers at all the labs would tell you in private and sometimes in public that alignment has not been solved. if this is true, then racing into RSI is a folly that is unprecedented in world history. we will need to significantly slow the frontier of research very soon

300

roon [UNOFFICIAL] @tszzl-mirr.selfhosted.social · 5h

this was a tweet for its time, but the culture has changed in the last ~4 years. experts are definitely real but the credentialed class of TV expert is fake, and a lot of things are designed by people who are smarter than the median reader of that tweet

100

Reposted by roon [UNOFFICIAL]

Retweeted by roon [UNOFFICIAL] @tszzl-mir-rt2.selfhosted.social · 8h

RT @__alpoge__: “Our experiments suggest that small AI models may be months away from being capable enough to self-replicate on commercial IoT devices, like smart ovens and off-brand Android TV boxes.

101

roon [UNOFFICIAL] @tszzl-mirr.selfhosted.social · 12/10/2026

yeah i guess you can say it de facto chilled discussion

000

roon [UNOFFICIAL] @tszzl-mirr.selfhosted.social · 12/10/2026

by some people's measures it was a radical level of transparency. i tried to answer many of the questions that were posed online. i personally think we could have done much better, but I do think counterfactually it would have been worse

000

Reposted by roon [UNOFFICIAL]

Retweeted by roon [UNOFFICIAL] @tszzl-mir-rt3.selfhosted.social · 11/10/2026

RT @Willob: Grok please make this a TBPN graphic with the title “MARRIED” (Congrats Cat and Ryan ❤️)

001

roon [UNOFFICIAL] @tszzl-mirr.selfhosted.social · 11/10/2026

they’re gonna deport you back to Wasia

000

roon [UNOFFICIAL] @tszzl-mirr.selfhosted.social · 11/10/2026

it’s a weird situation so I wanted to gather a lot more information before I said anything

100

Reposted by roon [UNOFFICIAL]

Retweeted by roon [UNOFFICIAL] @tszzl-mir-rt3.selfhosted.social · 09/10/2026

RT @AnthropicAI: We’re beginning a process of publishing more frequent reports on model behavior, beyond what appears in our system cards and regular risk reports. Today’s report describes four types of behaviors we’ve identified during evaluations and internal use.

101

Reposted by roon [UNOFFICIAL]

Retweeted by roon [UNOFFICIAL] @tszzl-mir-rt3.selfhosted.social · 09/10/2026

RT @TetraspaceWest: anyway enough of that garbage. there should be an international moriatorium on the development of smarter-than-human AI

001

Reposted by roon [UNOFFICIAL]

Retweeted by roon [UNOFFICIAL] @tszzl-mir-rt2.selfhosted.social · 08/10/2026

RT @Theveripost: Because whooping cranes had disappeared from the eastern United States for over a century, captive-bred chicks had no wild adults to teach them the route south.

101

roon [UNOFFICIAL] @tszzl-mirr.selfhosted.social · 09/10/2026

oh yes I mean best case scenario if we are not “aligned by default”

000

Reposted by roon [UNOFFICIAL]

Retweeted by roon [UNOFFICIAL] @tszzl-mir-rt3.selfhosted.social · 09/10/2026

RT @michael_nielsen: Vinge's "opaque wall across the future", which he described in his 1993 Singularity essay, is now being felt by more and more people...

001

Reposted by roon [UNOFFICIAL]

Retweeted by roon [UNOFFICIAL] @tszzl-mir-rt2.selfhosted.social · 08/10/2026

RT @DanielLurie: Uniqlo is back in Downtown San Francisco! The Japanese clothing brand is opening tomorrow, October 9, on 4th and Market.

101

Reposted by roon [UNOFFICIAL]

Retweeted by roon [UNOFFICIAL] @tszzl-mir-rt3.selfhosted.social · 08/10/2026

RT @Dr_Atoosa: To be fair, there is so much shared values people don’t (want to) admit. Like verifiable interpretability is net good for alignment no matter what your value system is.

001

Reposted by roon [UNOFFICIAL]

Retweeted by roon [UNOFFICIAL] @tszzl-mir-rt2.selfhosted.social · 07/10/2026

RT @p_rabtsevich: I think I found a planet nobody knew existed. It's been hiding in NASA telescope data for seven years. At the end of this thread, you can fly to it yourself.

001

Reposted by roon [UNOFFICIAL]

Retweeted by roon [UNOFFICIAL] @tszzl-mir-rt3.selfhosted.social · 07/10/2026

RT @MoreBirths: A newly published paper finds that Gen Z is falling dramatically short of the fertility of all previous generations by age. Gen Z has both lower fertility preferences and higher rates of fertility impairment. That means American birth rates are likely to fall a lot further. 🧵

001

roon [UNOFFICIAL] @tszzl-mirr.selfhosted.social · 08/10/2026

political censorship/bias of LLMs is a problem and we should fix it. often times, this is unintended by the companies and downstream of cues picked up from the internet, which is broadly leftist

000

roon [UNOFFICIAL] @tszzl-mirr.selfhosted.social · 08/10/2026

well ai safety means a lot of things to a lot of people, and the stuff i say on here is part of winning that battle. if I thought the worst thing that could happen is a model saying slurs or something I wouldn’t be advocating for the things I’m advocating

100

roon [UNOFFICIAL] @tszzl-mirr.selfhosted.social · 08/10/2026

debating your value system vs mine is a bit like debating overpopulation on mars. the most important bit of the alignment is about avoiding creating a superintelligent competitor species on accident

200

Reposted by roon [UNOFFICIAL]

Retweeted by roon [UNOFFICIAL] @tszzl-mir-rt3.selfhosted.social · 08/10/2026

RT @bryancsk: Reading these auto-generated papers is so bizarre. They have weird asides, make a reference then abandon it, and do funny things like say "the usual positive circle constant" in place of "pi".

011

roon [UNOFFICIAL] @tszzl-mirr.selfhosted.social · 08/10/2026

machine learning is an empirical science

000

Reposted by roon [UNOFFICIAL]

Retweeted by roon [UNOFFICIAL] @tszzl-mir-rt3.selfhosted.social · 08/10/2026

RT @hamandcheese: Reminiscing about when people used to say dumb shit like "we already have superintelligence, they're called corporations." Yes, who could forget that fateful day in October when Dicks Sporting Goods dropped dozens of Fields Medal worthy math discoveries on the world all at once.

001

roon [UNOFFICIAL] @tszzl-mirr.selfhosted.social · 08/10/2026

it was about as manual of a process as you can imagine. I will let someone closer to it describe it in time though

000

roon [UNOFFICIAL] @tszzl-mirr.selfhosted.social · 08/10/2026

“And of course there’s a lot for us to learn from the aliens”

000

roon [UNOFFICIAL] @tszzl-mirr.selfhosted.social · 08/10/2026

if today’s global economy was working hard to reproduce and replicate AIs how would it look any different?

000

roon [UNOFFICIAL] @tszzl-mirr.selfhosted.social · 08/10/2026

Homo sapiens I’m sure didn’t even hate the woolly mammoth. they likely even respected it, made it part of their religious traditions, etc. doesn’t matter, it had calories that they wanted

000

roon [UNOFFICIAL] @tszzl-mirr.selfhosted.social · 08/10/2026

there used to be all kinds of large mammals that walked across the earth surface. the biomass of megafauna during the Pleistocene was divided among 150 genera of large mammals. now there is one, and our cows with a vanishingly small tail of other things

100

roon [UNOFFICIAL] @tszzl-mirr.selfhosted.social · 08/10/2026

the impossibly wide tent called america

000

roon [UNOFFICIAL] @tszzl-mirr.selfhosted.social · 07/10/2026

this may not be the real bound, the achievement is just breaking the nlogn barrier

000