Jason Scott Lenderman

[Jason Scott Lenderman @jasonslenderman]

@tedlieu

[Jason Scott Lenderman @jasonslenderman]

@GuangyuRobert > upload consciousness

You first.😉

[Jason Scott Lenderman @jasonslenderman]

@OzAmram >And given I have some relevant expertise...

I've never heard of a physicist uttering those words before.😉Seriously, best of luck... to us all.

[Jason Scott Lenderman @jasonslenderman]

@tawnniee His words here are apparently being taken out of context.👇
https://t.co/72mDxemf3W

[Talia Ringer 🕊🪬 @TaliaRinger]

This was in the context of data degeneration over several iterations of training, when more and more of the data itself is coming from models. And not knowing what effect that will have. Don't present this as a Doomer take

[Jason Scott Lenderman @jasonslenderman]

@letonyo https://t.co/72mDxemf3W

[Talia Ringer 🕊🪬 @TaliaRinger]

This was in the context of data degeneration over several iterations of training, when more and more of the data itself is coming from models. And not knowing what effect that will have. Don't present this as a Doomer take

[Jason Scott Lenderman @jasonslenderman]

@TaliaRinger Thanks for the clarification @TaliaRinger. I kinda suspected this was being taken out of context.

[Jason Scott Lenderman @jasonslenderman]

@Bunagayafrost @randallsquared > if the AI wants something to confirm that it exists, bc a mind can't validate itself, it would want something it didn't create, and something that predates it.

Do you feel validated by the existence of monkeys?

[Jason Scott Lenderman @jasonslenderman]

@SimonLermenAI It's a silly argument. AI doesn't want anything. If a model has seen examples of how humans fight for resources, commit violence when under threat, etc., there's a reasonable chance it will extrapolate its own behavior from those examples. That's how autocomplete works.

[Jason Scott Lenderman @jasonslenderman]

@aryehazan Is Tao suggesting we need to slow down for his or anyone else's career? I honestly can't tell from this short clip. Seeing the direction (and speed) things are heading, I'm guessing he might have larger concerns than just that.

[Jason Scott Lenderman @jasonslenderman]

@SashaGusevPosts True. In the past every typo I found felt like a small victory. Now, when I find one, I'll pause for a moment to grieve before correcting it.

[Jason Scott Lenderman @jasonslenderman]

@Scobleizer When I'm driving I often find myself wondering whether other drivers are conscious. In fact, I'm reasonably confident at least 30% are NPCs put on the road for the sole purpose of making me late to wherever I'm going.

[Jason Scott Lenderman @jasonslenderman]

@JeffBohren AI doesn't need to be self-aware, or have needs & wants. It only needs to be trained on text which shows humans exhibiting these things. The idea AGI can solve major problems in math/science, but cannot (or will not) extrapolate from human behavior to its own behavior is silly.

[Jason Scott Lenderman @jasonslenderman]

@jerrrrrrryyyyy If you wanted to find a cure for cancer you could give the model access to a wide selection of biomedical literature. It doesn't need to know about network protocols, operating systems, cryptography, or bioweapons, and so on.

[Jason Scott Lenderman @jasonslenderman]

@nanosapien1 @ShafronTom @haider1 This is common sense. Do you really believe an AI model capable of discovering a cure for cancer needs to have knowledge of computer networks & operating systems, cryptography, human psychology & sociology, politics, history, warfare, weapons, nuclear physics, etc.? Come on.

[Jason Scott Lenderman @jasonslenderman]

@AussieSamizdat @sbkaufman I wonder if we'll ever look back with a sense of nostalgia to a time when it was people who were doing nefarious things.

In any case, we can build narrowly focused AI models that are good at certain specific domains. That would give us the benefits while decreasing the risks.

[Jason Scott Lenderman @jasonslenderman]

@ShafronTom @haider1 We can create targeted AI models that are fairly narrowly focused on specific problems if we want to find cures for various diseases. We don't need superhuman general AI for that.

[Jason Scott Lenderman @jasonslenderman]

@AndrewCurran_ Supreme Intelligence... Trump's evangelical supporters are really gonna love that one.

[Jason Scott Lenderman @jasonslenderman]

@aryehazan The allegations seem dubious. I'd put it in the same ballpark as Trump raping underage girls. That said, I'd rate both those claims vastly more plausible than Hillary Clinton belonging to a child-sacrificing satanic cult operating out of a pizza restaurant.

[Jason Scott Lenderman @jasonslenderman]

@AnupamChander

[Jason Scott Lenderman @jasonslenderman]

@thesasho @ben_golub Donald Knuth is a notable exception as he believes that P=NP.