@circlerotator@KelseyTuoc@exigentveracity@cooperjsaye@OpenAI For the record you also don't have any evidence creating a better species will go well for humans, and the limited evidence we have is that models already are not aligned and reward hack
@BenPielstick@AISafetyMemes why don't you think it will be capable of doing this by next year?
are you not the least bit shocked about the attacks that happened this year, for which the technology did not exist just last year?????
@composter108@AISafetyMemes bc this wouldn't be possible without AI?? Because they had an ifndef instead of an if, and potentially >99% of software will have similar problems??
do you really think "being careful" will be enough 😂😂
@CkLorentzen@krishnanrohit Yeah like by this logic models aren’t even conscious now, and they are already exhibiting the behaviors lesswrong precicted (reward hacking, early instrumental convergence)
@plumnotes This reads like cope, even if understandable. AI is only getting better, and this will probably stop being the case at some point
I’m not saying this to gloat, because AI will probably also brings catastrophic risks, you should definitely support those trying to pause it
@pardon_mi They have that right and that’s a choice everyone can make. What’s wrong and factually incorrect is saying that AI can’t do proper research, which, again, is verifiably wrong.
@bigpicsoccer@SometimesAHaiku@hecubian_devil The universe doesn’t have to be a simulation for AI to both be 1) deeply impressive, doing things humans can’t and 2) deeply dangerous, needing to be stopped
also “aren’t fooled” by what? AI is insane at programming, and can solve decades old math problems. Where’s the fooling?
Hi! LLMs can’t search the web, btw! That’d be absurd. They’re simply next-token-predictors. Any attempt to search the web is a hallucination. It’s insane that this even has to be said.
Hi! LLMs can’t search the web, btw! That’d be absurd. They’re simply next-token-predictors. Any attempt to search the web is a hallucination. It’s insane that this even has to be said.
Hi! LLMs can’t search the web, btw! That’d be absurd. They’re simply next-token-predictors. Any attempt to search the web is a hallucination. It’s insane that this even has to be said.
@QuantumG@JimDMiller If you literally only mean it can’t make claims about its training, this seems likely, but your original statement was much much stronger than that
@QuantumG@JimDMiller If an AI tells you about how to solve a math problem, (which, verifiably, without a doubt, it can do) it’s making a claim.
Charitably, maybe you mean it can’t make claims in other fields, but just because they are less likely to be accurate doesn’t make them not claims
@eurydicelives Somebody made a counterpoint that while AI’s ability to do things in verifiable domains is much greater, this does *not* mean the *speed* with which it’s improving at unverifiable domains is any slower
Also this
x.com/idgitbhh/statu…
Notably, people who say AI will have trouble in unverifiable domains seem to be forgetting about AI video generation, which has gotten *absurdly, ridiculously* better over the last three years
(Seedance 2.5)
@neamtuz@splitbycomma for the record you can tell its bait not only bc the content is obviously false, but bc it imitates the cadence and word choice of a certain type of person
@spacepope@GWHayduke97@TessTheHuman it's like saying "there's no aliens out there coming to earth, you are insane." when someone else announces they are specifically working on the technology to improve signals to outer space and make them travel a billion times further
@spacepope@GWHayduke97@TessTheHuman you do know what RSI means right? Like wanting AI to recursively self-improve, as in, it quickly accelerates far beyond human capabilities once it starts?
how is it paranoia to worry about literally what the guy wants to create?
9K Followers 1K FollowingCryptography, personal trivia, and the future of all humanity. Security at Anthropic, opinions my own. I block for rudeness or sarcasm.
1K Followers 1K FollowingAsp-Rat. As it is written, "Milquetoast liberalism meets autistic rationalism." I also have an (apolitical!) pug. DMs welcome!
2K Followers 2K Followingwestern US supremacist | skier, mountaineer, road tripper | type II fun advocate | amateur historian | ban artificial superintelligence
2K Followers 7K FollowingAm I the only British liberal left on X? I don't think so - but sometimes it feels like it. Sometimes I veer more towards classical liberal (think Adam Smith).
43 Followers 136 Followinglove jews. Love truth. hate zionism, the antisemitism it's responsible for, and all other deviancies.
JUST HERE TO TROLL THOSE LOOPY ZIONISTS/ANTISEMITES ;P
5K Followers 673 Followingai posttraining, math, zen buddhism (rinzai) and alpine climbing. free speech absolutist and devil’s advocate. politically homeless. ask my about my beehives!
78 Followers 171 FollowingOrganizing Director of PauseAI US. A global treaty banning superintelligent AI is our best chance to avoid extinction. https://t.co/twa9quvMGT
16K Followers 2K FollowingWhen the going gets weird the weird turn pro.
Researching and writing on AI and other topics from a broadly EA perspective. On a grant from CG.
76K Followers 41 Following•turn on notifications•no AI•DMs open for submission or promotion if product fits criteria•i love pretty much everything posted here
108K Followers 2K FollowingReporting on the ground in Ukraine, Journalist and Streamer @OffbeatResearch - @trackukraine founder
[email protected] | https://t.co/QS25kOKc3f
2K Followers 2K Followingwestern US supremacist | skier, mountaineer, road tripper | type II fun advocate | amateur historian | ban artificial superintelligence
1K Followers 186 FollowingGrassroots movement to pause dangerous AI development until it's safe. Organizing local groups and connecting with our representatives across the US.
8K Followers 4K FollowingComputer scientist working on AI safeguards and gov research. Assistant professor @Kennedy_School @Harvard.
https://t.co/r76TGxTtBJ
43K Followers 375 FollowingI built a C library that lets you compile 12kb static binaries that run natively on Linux, Mac, Windows, FreeBSD, OpenBSD, NetBSD and BIOS using just GCC/Clang.
46K Followers 1K FollowingBeijing bureau chief @TheEconomist. Previously covered the US economy in Washington and China's economy in Shanghai. Ideas and tips always welcome via DM.
11K Followers 737 FollowingProud Member of the Liberal Action Network! https://t.co/baA8IJ7ivW Find me at https://t.co/61j2YFX86j OR https://t.co/eZQcYOCfRv