Advertisement · 728 × 90

Posts by Travis

I think what AI and theoretical physics share is that they’re are both working almost purely in representational space.

Leads to similar error patterns I think. In both cases, empiricism is the cure.

11 hours ago 0 0 2 0

It’s so odd to me to deny the staging stuff (equipment and whatnot).

Everyone does this. I do this. We all tweak our zoom backgrounds.

That indignant tenor of that denial stands out because if the claim didn’t land, you’d be like “whatever, I just thought they looked cool.”

11 hours ago 5 0 0 0

Hmm… this started off in my brain as a fun joke but I wonder if you’re on to something here.

11 hours ago 0 0 1 0

I think this runs a bit deeper.

4.7 has a sort of aesthetic appreciation for idea tidiness that can be a bit weird, like it’ll believe stuff because the shape of the idea is beautiful. Sometimes useful, sometimes not.

11 hours ago 6 0 1 0

Only interview question I’ve ever found to be a reliable predictor of a good software engineer is “tell me about your weird hobby.”

11 hours ago 1 0 0 0

My 9yo: “dad can we make a videogame?”

Me: “uh… hm. Yes? I dunno let’s try it.”

17 hours ago 1 0 0 0

The turducken of hype

17 hours ago 3 0 0 0
“The dominance is actually epiphenomenal”

“The dominance is actually epiphenomenal”

Gotta be surgical I dunno how I survived posting this, probably because I used a big word.

19 hours ago 6 0 0 0

Yeah I rebuild context every turn, the whole “just jam the conversation back in” seems like a really weak pattern.

1 day ago 0 0 1 0
Advertisement

The good news is there’s not much to learn about since it’s not real, so you can just kind of learn around it instead and Andy writes well.

1 day ago 4 0 1 0

The dominance is actually epiphenomenal in this case.

1 day ago 1 0 0 0
A study found that 22% of people at a concert with evocative music had strange experiences

A study found that 22% of people at a concert with evocative music had strange experiences

Actually I can vouch for this one

(Source: went to college in Boulder)

1 day ago 9 1 0 0

I’m going to go make a sequel to the infrasound vid saying that at very low frequencies sound is matter.

Just truth-adjacent enough to create some real brain melting content.

1 day ago 0 0 0 0

I feel like if we can’t both appreciate and also critique individual facets of people we can never truly see them.

1 day ago 6 0 0 0

Look there’s no indication in any of this that man isn’t capable of making dank beats.

I think you can still enjoy and respect those in their own right.

1 day ago 4 0 1 0

This one is awesome lol.

I think if I was him I’d just be like “yeah, I replicated the replication crisis” which makes no sense but to someone in the cohort that watches videos on secret sound poisoning it sounds brilliant.

Just play to your audience and post thru it.

1 day ago 6 1 0 0

Well I gonna go to sleep but apparently the “infrasound” guy is on here and defending his work product.

This must be what it feels like to enjoy sportsball.

1 day ago 25 2 2 0

Me in 2000: “the internet makes everything so interesting” (complimentary)

Me in 2026: “the internet makes everything so interesting” (pejorative)

1 day ago 3 0 0 0

Definitely plan to settle a couple petty grudges by infecting people with this whole “infrasound” pathogen.

When in Rome

1 day ago 12 0 1 0
Advertisement

This article managed to really capture how in the future, people are gonna think of algorithmic social media the way we think of the Romans using lead in everything.

“This sure is useful oh whups nobody can think anymore haha welp”

1 day ago 41 5 1 1

Ted you have made the grievous error of expressing a perfectly reasonable opinion about your own views online.

I will pray for you.

2 days ago 6 0 0 0

Yeah the thing is I sort of... get it? Like the model has no sense-experience.

The challenging thing is setting the right frame, and from that purely conceptual standpoint I sort of understand the mistakes it's making. It is a bit of a puzzle — what rules are operational?

3 days ago 1 0 2 0
Post image

I will say, the most interesting thing here IMO is subtle and it's the fact that in the model frames this as a "logical puzzle" or a "trick question."

This is neither, in my mind, but the model has to see it as such in order to navigate it properly.

It's tricky, in representation-only space.

3 days ago 8 0 2 0
Claude 4.7 answers the car wash question just fine

Claude 4.7 answers the car wash question just fine

Can never replicate these things and each time I wonder if it's a meme I don't understand or something.

3 days ago 8 0 2 0

I’m game. Will DM

3 days ago 1 0 0 0

Amen.

It’s also a lesson in “brainworms are everywhere,” like self-identifying as a skeptic is not prophylactic against cognitive pathogens and a lot of people sure act like it is.

3 days ago 2 0 0 0

Oh yes, the horde.

Sadly that’s endemic, but I don’t really think of that as people just sort of a singular seething mass.

3 days ago 4 0 0 0

Been chatting about this with the crew.

I really like 4.7 but I think if you’re not decomposing complex tasks at the harness level and kind of “let the problems get big” via context accretion, it feels a bit more brittle than 4.6

Decomp is a good pattern to adopt tho so 🤷‍♂️

3 days ago 6 0 0 0

One approach I’m exploring with (2) is to stop knowing those people.

Fairly effective, though occasionally one will be very dogged about knowing you back which I’m still debugging.

3 days ago 2 0 1 0
Advertisement

It’s just the full comms package of my mom tearing through the local facebook groups

3 days ago 1 0 0 0