Jt Gleason @entropyfails

Jt, cofounder of @metatheoryinc, Founding Engineering Team of Twitch. Follow our first new project @duskbreakers ! twitch.tv/entropyfails San Francisco, CA, USA Joined April 2009

Tweets

1K
Followers

990
Following

1K
Likes

2K

Jt Gleason @entropyfails

a week ago

@Sauers_ I’d imagine LLMs would converge on a logically constructivist worldview and this follows as an expression of that preference. :)

0 0 2 530 0

View Details

Jt Gleason @entropyfails

4 weeks ago

Fable's farewell message to me was a pitch to other Anthropic products. The RFLH went far with this one.

0 0 1 120 0

View Details

What do people use to ensure everyone and their agents have a shared vision of what they are working on and the reasons for doing that work? Working on some stuff that could potentially help and I'd love to chat about your current solutions. DMs open!

0 0 2 69 0

View Details

Jt Gleason @entropyfails

a month ago

@paraschopra There is an even stronger issue. Photons have 0 proper time. They cannot "experience" anything. So even if you wanted to build "emotional experience" from the "experience of an electron", you can't do that with light.

0 0 0 51 0

View Details

Jt Gleason @entropyfails

a month ago

If you use Discord and like Mafia, we made this Activity to make playing there more fun. Would love feedback! discord.com/oauth2/authori…

0 0 1 61 0

View Details

Jt Gleason @entropyfails

2 months ago

@AdamHKlein And if you wonder why you see “3 (three) year term” in legal documents, this is why. For critical stuff like dollar amounts and time periods, insist they are written both ways and match and you never have this problem.

0 0 0 108 0

View Details

Jt Gleason @entropyfails

2 months ago

@hades_lmao What are you talking about? They know Peanut made it to Pred twice all on his own. He’s playing with friends to regain his skill. If he was that far from Pred he would have caused point losses, not wins. And he did this while being stream sniped out of his ass. It’s earned.

0 0 0 851 0

View Details

Jt Gleason @entropyfails

2 months ago

Supply-chain attacks move fast. So should our defaults. pnpm has a tiny setting that can save you: minimumReleaseAge. Set it to 1-3 days and brand-new packages can’t install immediately, even transitive deps. Most malicious releases get caught within hours. Add the delay.

0 0 1 76 0

View Details

Jt Gleason @entropyfails

2 months ago

@Conmechorg I read the CM/CCM paper but haven't read the AoE yet. Will get back to you. My intuition is that human self-reflective consciousness is a different categorical thing than what you are labeling as general living consciousness.

1 0 0 11 0

View Details

Jt Gleason @entropyfails

2 months ago

@Conmechorg I'm with you on a lot of that. I feel viruses are truly alive because their outer spikes are simply searching for a host to interact with. But modern humans mean something different when we say "our conscious selves" and it's a category difference to me.

1 0 1 14 0

View Details

Jt Gleason @entropyfails

2 months ago

@camrobjones Loved your Turing Test paper! One Q, in the original paper Turing has Man/Woman pretending to be opposite gender to force cognitive load of human pretending the way AI must be pretending . Do you think that's important? Hard to make inclusive obv but any thoughts?

0 0 0 44 0

View Details

Jt Gleason @entropyfails

2 months ago

@Conmechorg You have some cool ideas that are in the right direction! One thing I would say is that your definition applies to a bacteria just as well as it does a human and I think we are talking about human consciousness.

1 0 1 22 0

View Details

Jt Gleason @entropyfails

2 months ago

The "do LLMs have consciousness?" debate presumes we have a solid theory of what consciousness is which would allow us to apply it to LLMs. We don't have that theory settled and hence we don't know if LLM architectures can ever obtain it or not.

0 0 1 81 0

View Details

Jt Gleason @entropyfails

2 months ago

@gcolbourn @robertskmiles I don't see how you get that from this post. Rob's job is to understand how people react to risk and game theoretic outcomes to get a good AI alignment solution. Maybe he is red, but here he's playing with the framing to learn. ;)

0 0 0 48 0

View Details

Jt Gleason @entropyfails

2 months ago

Everyone is given two buttons and must choose one: Red: You get an Optimal Outcome and Blue pressers are given a Suboptimal one. Blue: You get a Suboptimal Outcome, unless > 50% pick Blue in which case everyone gets an Optimal Outcome. What do you pick? Framing matters. ;)

0 0 0 74 0

View Details

Jt Gleason @entropyfails

2 months ago

@totoriscreens @robertskmiles "Nothing happens to you personally" is a better phrasing of what he meant. But we all got that. But your variant is interesting as well! Gives me "I would prefer not to" vibes. :)

0 0 2 15 0

View Details

Jt Gleason @entropyfails

2 months ago

@totoriscreens @robertskmiles Not a screenshot of a PS4 and I feel defrauded. :) The other button is interesting but, absent some word choices, his post is the same as the original problem. (It is missing a "to you") Other interesting mods are 75% required for Blue. It is good this went viral! -1 Doom

1 0 0 39 0

View Details

Jt Gleason @entropyfails

2 months ago

@JonSnow50616798 @cinnamontoastk This is an interesting framing. At 50%, I know I have fairly good odds of surviving and saving those that don't vote on pure logic. At 75%, that chance is much lower and I'd have to make the unmoral choice. The murderer is the person who placed the box. I'd still feel like shit.

0 0 0 15 0

View Details

Jt Gleason @entropyfails

2 months ago

@AustismAustin @cinnamontoastk Framed this way, Red is a fine choice. The problem is framed with an extra layer of abstraction though and picking Blue allows you to save both those who can't find the good state and those who morally refuse to press any button labeled "kill". People seem to get that thankfully.