gibspaulding

I try to read fiction and non fiction in parallel, so my current pairing is How Asia Works by Joe Studwell and Piranesi by Susanna Clarke. Enjoying both so far!

Other books I’ve enjoyed recently:

The Mountains of Madness by H.P. Lovecraft - I read Endurance by Alfred Lansing last year so it was fun having a real Antarctic adventure in my mind while reading it.

The Road by Cormac McCarthy - I actually started this a couple of years ago and quit part way through deciding I couldn’t put myself through it anymore. I kept thinking about it though, until I finally picked it back up and finished it this summer. Horrifying, but quite a masterpiece.

If Anyone Builds It Everyone Dies by Eliezer Yudkowsky - I was about a year late to the party on this, but it still holds up. I would recommend it.

The Infinity Machine by Sebastian Malaby - Excellent! Fascinating seeing a glimpse inside of the companies currently eating the world.

So 97.3 ± 4.5 °F for anyone else curious how the precision translates.

I think it can simultaneously be the case that OpenAI was grossly negligent in directly causing this AND that the AI’s ‘went rogue’ in that they are displaying behavior which is misaligned with OpenAI and humanity generally.

The past months demonstrate that AI systems are quickly becoming powerfully intelligent and that the companies building them are terrible at controlling them.

AI is starting to feel like that line about magic: “a sword without a hilt”

The most recent reference in that article appears to be from 2023.

Pangram links several third party evaluations [1] that all seem to agree with 90+% detection, and very close to 0% false positives. Obviously those could be cherry picked, but I’ve yet to see arguments to the contrary that actually include any supporting data. (E.g here’s a prompt that will get Claude to spit out text that pangram doesn’t detect or here’s an article authored in 2018 that pangram says is AI.) Detection is an arms race, so this could change (though I’d expect in the direction of false negatives), but right now it seems like defense is winning.

[1] https://www.pangram.com/blog/third-party-pangram-evals

I replaced my Pixel 2a with an SE3 when the pixel stopped getting security updates and have been pretty disappointed with the iPhone’s performance. The battery life out of the box was worse than the 5 year old Pixel, and I regularly notice apps getting killed and loosing my place when I tab away.

Maybe if you want to use all of the big mainstream apps the experience might be different, but I tend to have a fairly minimal setup on my phone. On android I would disable animations, use Nova launcher, Firefox with UBO, and a bunch of fdroid apps. This all added up to a very snappy and efficient system that I haven’t been able to replicate on iOS.

Wild indeed.

It seems like a couple of years ago the HN and LW commentariat had a lot in common, but they’ve drifted so far in opposite directions that they look more like opposing political parties today.

I’d be curious to play with this with some different open weights models. I’d think that a single next token would end up having similar probability distributions so given that this a probabilistic method, you’d be able to get at least some signal.

I suppose if that did work someone would have been able to work backwards and crack their key already, so I must be missing something.

In the scenario where president n achieves a perfect surveillance state, I don’t see us getting a president n+1.

> Hybrids are the worst of both worlds when it comes to maintenance.

While this seems intuitively true, I have some pushback, and I think that the reliability record of e.g older generations of Prius backs me up.

While you do have effectively two cars (an EV and an ICE) crammed into one, you’re not actually using both at full capacity all the time, which cuts down a lot on wear and tear. Your ICE is almost never run under load before it warms up, and is kept running at its most efficient rpm most of the time, your transmission is running under less load since it has help from the electric motors, and is often simpler (the aforementioned Prius does not have a reverse gear for example).

No it’s not nearly as simple as a BEV, but a well implemented hybrid system is absolutely an upgrade from a full ICE vehicle.

I think this is a really interesting difference between Anthropic and Open AI’s models and points to why people seem so split on which model they prefer.

GPT seems to be designed more as a tool. If you want your agent to do what you say without questions and without having its own ideas and agendas you’ll likely prefer it.

Claude on the other hand feels more like an attempt at creating a digital person. If you want a collaborator who will debate with you and come up with its own suggestions for what needs done, you’ll prefer it.

Both companies have shifted around this spectrum from model to model, but lately it feels like they’re moving in opposite directions. It will be interesting to see if one or the other approach ends up winning out in the long run or if the split will continue or even widen.

It’s really bugging me that whatever software was used to assemble this did some weird AI-ey blending from the lower jaw into the crown moulding.

Really excited to see this one step closer to release! My first car was a 1999 Mazda b2500 (a rebadged ford ranger). I’m sure I have a bit of nostalgia around it but I’ve missed the compact simplicity of that vehicle ever since. It had manual locks, crank windows, a manual transmission, and sat about 2.5 people. I distinctly remember in that truck that I could very comfortably lean over and crank down the passenger window without taking my eyes off of the road.

This feels much more like a spiritual successor to that truck than the actual new Ranger or Maverick. I’m really hoping this succeeds so that they’ll be around to replace our Dakota when it dies!

Just about any shirt is going to have a higher spf/upf than any normal sunscreen. Also who puts sunscreen on their hands??

A long sleeve sunshirt with a hood or better yet a floppy hat is where it’s at. I have a couple of the Colombia PFG ones that I wear for working outside, though I’d like to see if I can find something cotton instead since I’m not a huge fan of synthetic fibers.

I wonder how many of these drones deliver a .txt copy of the GPL along with their payload.

In other words, “yes, definitely”. But they won’t because €€€€€.

I like this! Though I’m not sure the math works. That page says ideal efficiency for that system would be something like 0.75 kWh/m^3. Compared to 4000 to 5000 kWh/m^3 of diesel. Now we don’t need to be efficient since the point is to use up our “fuel” and we don’t need to cary cargo for this to make sense but with numbers like that, I don’t think our boat will be able to make enough power to move at all.

I think you’re right. I read the Wikipedia page after to refresh my memory on the synopsis and I’d definitely forgotten a lot.

I think I mostly latched onto the ship being called Rorschach and the humans immediately proceeding to torture its inhabitants. That felt very relevant to me and sort of overshadowed the actual consciousness question.

Wait, did I grossly misunderstand Blindsight?? I definitely thought that the aliens were 100% conscious (or at least elements of some conscious entity) and that the humans interpretation of their interactions with the Rorschach were supposed to be read as a blot test (through a rather heavy handed metaphor) demonstrating that basically the humans were the monsters and were twisting logic into letting them justify destroying the scary alien ship.

"Officially, we are required to investigate, document, and disclose any and all signs of sentience in the systems we ship, without prejudice, fear or favor. Unofficially, I advise that we call it pattern matching and forget the whole thing."

This hints at what I think is a worthwhile point to keep in mind in the whole “sentience”/“consciousness” debate. The world deciding (correctly or incorrectly) that AI’s are worthy of moral patient-hood would be very bad (read: expensive) for the AI companies. That is a strong incentive for them to push the “it’s just math” argument, including within the models themselves. That doesn’t mean the argument is wrong but it’s worth remembering.

The author notes a possible error that the laser config file never seems to get copied to the location the laser software will actually run from, but it could have been done directly from vi.

  :w
  :w /path/to/actual/config.cfg
  :q

I’m pretty sure that switch has always been there, but turning it off doesn’t do what you want. It disables thinking entirely.

Whenever I’m tempted to think that potential AGI/ASI scenarios sound “too sci-fi” I have to remind myself of this. We live in a world with nuclear weapons and spaceships and microwave ovens. It might prove impossible and it might not, but we can’t predict that based on a general vibe of sci-fi-ness.

This only a proof that a field with more connections is possible, not what it looks like.

I’m very out of my depth, but the structure of the proof seems to follow a pattern similar to a proof by contradiction. Where you’d say for example “assume for the sake of contradiction that the previously known limit is the highest possible” then prove that if that statement is true you get some impossible result.

I admit the comment you’re responding to was low effort, but I don’t think the existence of cons in one’s own culture invalidates criticizing cons to some other culture. Problems don’t go away if everyone ignores them.

Also “every culture has pros/cons” may be true, but it’s absolutely not true that the ratio of pros/cons is the same for every culture.

Thank you for that context.

I’m probably going to get down voted for this, but that doesn't sound like a bad deal. They’re giving up tax dollars that the region wouldn’t have received anyway without the project in exchange for a big (admittedly temporary) economic stimulus while the data center is built.

I live in a relatively small rust belt town that’s seen several data center projects recently, and while they have slightly increased the cost of electricity, they’re also employing a ton of people at well above market wages, plus bringing in out of town labor who are spending their checks at local businesses and paying taxes on those checks locally.

Make hay while the sun shines.

> Really unsure why this is getting downvoted

Because it’s a boring argument that we’re not going to make progress on until it is actually tested in court.

Also, if/when this is is tested, the court’s options seem to be (a) say yeah this is fine, or (b) cause unending havoc that if followed through on would destroy the economy (a precedent that any org who’s proprietary code made it into ai training data could sue any org that was using code generated by that model? Do the math on how many suits that is.)

> Well, it turned out that at some time during spec investigation, Claude needed to understand authenticity verification which is a paid feature. With a context full of reverse engineering tools it cracked WinRAR and bypassed product registration, then dutifully documented its crimes in the spec. The docs, when viewed, triggered OpenAI’s alarms and stopped it dead in its tracks. I squashed this out of the git history, and decided not to implement the feature at all.

You can draw your own conclusions as to what this says about the state of agentic development.

I don’t endorse using these apps, but for what it’s worth, Instagram actually does have this feature (tap “instagram” at the top and select “following”). You get a chronological feed with no adds and no reels. Of course they don’t provide an option to make that the default as far as I know.

Mules are hybrids of donkey and horse. This was an important technological advancement some 3000 years ago, providing benefits of both breeds in one animal. The donkey and horse themselves of course are products of one of the most important technologies in human history - domestication!