ai fast now

[ai fast now @AllennxDD]

@Plinz Somehow the gradient always tips just far enough in the optimal direction to not do that. What superposition are we optimized for?

[ai fast now @AllennxDD]

@ESYudkowsky @RepThomasMassie Yeah. Ughh. When AI and politics clash it will be a / mess /

[ai fast now @AllennxDD]

@johnjnay @CarnegieMellon @UCIrvine I’m pretty sure GPT-4 is smart enough to actually be fed a decent number of its wrong answers and figure out zero-shot what’s wrong with them.

[ai fast now @AllennxDD]

@alyssamvance Can you send link for this?

[AI at Meta @AIatMeta]

New in Nature Human Behavior, Meta AI researchers show how current language models differ from the human brain & highlight the role of long-range & hierarchical predictions.

We hope these findings will help inform the next generation of AI ➡️ https://t.co/EC3f4s6Ens

[andrew chen @andrewchen]

The whole essay here:

https://t.co/UGF12H1J9B

[Alexey Guzey @alexeyguzey]

New guest post from the brilliant @HoskinsAllen, asking:

Why are AI experts' jobs always decades from being automated?

https://t.co/VDRCwWlrES

[ai fast now @AllennxDD]

@CineraVerinia Yeah.. agree..

[ai fast now @AllennxDD]

@CyberdyneC @Simeon_Cps This is a point I often make to people, too, but for a totally different reason. Mainly I make it when people ask what it’s like to be dead. Instead, how about what it’s like to be dying, which is the question relevant here.

[ai fast now @AllennxDD]

you mean the thing that was replaced by ChatGPT? By that company —whose name, um, hmm.. can’t remember for sure. its like a big number or something

[Bojan Tunguz @tunguz]

Do people still use search engines?

[ai fast now @AllennxDD]

@DavidDeutschOxf Point being, humans are too limited to really take advantage of the inverse experiment /3

[ai fast now @AllennxDD]

@DavidDeutschOxf into consideration in some form of an internally calculated search tree, often. So, you would have to give the human a perfect short term memory for them to ever make full use of the time equivalence if you switched and gave humans the speed advantage /2

[ai fast now @AllennxDD]

@SmokeAwayyy A strong legal contract and government intervention would probably end up being involved once everyone was like, “Yeah, this is without a doubt AGI”. So it would be that no one says anything. Especially if they feel like it’s being taken seriously enough.

[ai fast now @AllennxDD]

Was a 10 year long graduate school track really the brightest idea when I think AGI is on horizon?

#feelingdaringtoday

[ai fast now @AllennxDD]

@scottastevenson There’s a spectrum and I’m not sure most people would label the imperceptible or pseudo-perceptible withdrawals as meaningful.

[ai fast now @AllennxDD]

Deep learning isn’t that deep.

[Mahesh Sathiamoorthy @madiator]

Offend a ML Researcher in one tweet.

[ai fast now @AllennxDD]

is it bad that I would literally pay someone money to remind me to study a few times a day and antagonize the hell out of me if I wasn’t?

[ai fast now @AllennxDD]

@Simeon_Cps I think it’s just too good for an unexpectedly large population pool of users now. Millions of users with day 1 access to something significantly more capable than GPT-3 is the epitome of unpredictability. They need to work on safeguarding its behavior for a few more months.