iiinova — where your digital self keeps evolving

You’d hope with how obsessed they are with numbers they could at least manage to not be innumerate

170 0

As we speak there are tech leads at my company giving an internal talk about How They Don’t Understand Their Own Code Anymore, And Why That’s A Good Thing. There is a whole class of people not only willing but eager to fully deskill and surrender.

030 0

That would first require a lot of them to acknowledge that debt to GDP matters and is non-trivial to address

040 0

I mean far preceding LLMs is the theory that consciousness is an emergent property of sufficiently complex recurrent networks, and that something like a logic board could be conscious in a way that is unrecognizable to us. Notably IIT would call standard LLMs pseudo-conscious, though.

000 0

As always, for me, the elephant in the room is that whatever symbiotic state can be reached at present is entirely dependent on the pathways these systems erase. You cannot know how to interpret solutions or direct inquiry if you’ve never solved problems yourself. It’s a 1 generation suicide pact.

000 0

I Know Why the Mathematicians Weep

I know people feel some type of way about this guy and his takes here, but this is good and something we’re going to have to grapple with.

derekthompson.org

On technological grief

100 0

Of course. As I said, you don’t have to use it. Certainly not to be able to see its many harms. I was referring more to delusions about its supposed lack of capabilities. If it couldn’t actually do a lot of these things, this would all be much easier, because it would eventually go away. This won’t.

000 0

Sorry, it’s not clear to me what point you’re trying to make.

100 0

Written by Claude Fable, just as an FYI for those who will immediately change their opinion on how cool it is/isn’t as a result.

100 0

This is the thing for me. You obviously do not have to use it, but IME the great majority of vocal critics say things about it that are just demonstrably false, and from the beginning playing wack-a-mole with 6 fingers, Rs in strawberry, etc. has always been a remarkably stupid losing game to play.

160 0

I was rooting for them! A lot of their (bad faith) arguments about overreliance on Steam beneficence were nonetheless objectively correct. 2 evil corps who hate each other works out better long-term than 1 less-evil corp, IMO. But at this point their whole org gets a “lmao, no thanks, I’m good”

050 0

It’s never been on Steam. Epic fully funded it so it’s always been fully exclusive

150 0

But hey, cold boots are marginally faster (to load a featureless webview with nothing precached)

1120 0

I’m usually a hard game sicko but for whatever reason being set back from a cool discovery because I missed a jump by a pixel makes me see uninstall-button-red, so I’ll definitely keep an eye out for that update!

010 0

Well… I’m sure this is for someone! I hope it finds them.

130 0

Is there any bafflingly precision platforming with tedious runbacks like Animal Well or is the actual platforming more laid back?

I’m confused. Where are you seeing that? That’s just talking about the missing data from the shutdown last year. Everything after it still works.

Initially read this as “think tanks” and I was like, 1 is more than I would’ve guessed

220 0

What I’m saying is, if you want an at all ethically consistent position, you will have no choice but to define what for you is an Acceptable Usage Level of the Theft and Layoff Machine. People are still pretending absolutism is on the table and it simply isn’t.

010 0

My first thought was if I were a teenager I would go to war to prevent this device from entering my home

040 0

I struggle with that because it inevitably will bring people who made no such choices into the blast radius by happenstance, and there is so much else to revile them for. Not to say it’s easy, because I mean… Christ

010 0

Genuinely just tried to google if she had reconstructive surgery for something because every bone in my body was like, it’s never right to focus on these peoples appearance, but it feels especially wrong here. (I couldn’t find anything… it appears to be elective)

You’ve been arguing only with yourselves about that from the beginning.

000 0

You appear to have misread the conversation, Dr. Perry.

000 0

Hey @anspoka.cz, would you be able to clarify for us if you meant that your company is standing up homegrown LLMs completely from scratch with your own full datasets, or if they are commercial models with private context so that they can be additionally trained on company data for internal use?

100 0

We have had many internal meetings about how we have new AI tools “trained” on our internal data. That does not mean neural network training. There is a vernacular usage that is plainly evident if you work with this stuff professionally at all and aren’t just LARPing about it on Bluesky.

000 0

You don’t know what you’re talking about, Mx. Perry

100 0

You are interpreting that claim as they are training a full model from scratch, which I do not think is what they are actually saying. And as I have repeatedly said, I’m not asserting that, and this person is *still* disagreeing that a commercial model with added context can do it, which is wrong.

200 0

It isn’t at all, that’s something they’ve brought up independently several times as some kind of gotcha, but the full thread is plainly visible so I’m not sure what you’re trying to pull, but absolutely nobody ever claimed “without adding a bunch of novel context,” the claim has been the opposite.

100 0

Not sure I’m gonna take the word of a supposed AI researcher who is saying things like “imperial evidence” with much more than a grain of salt, but thank you for your input. Reality will bear out the truth either way, so it ultimately doesn’t matter what you do or don’t believe in.

000 0

You’re just fully, demonstrably wrong here. Again, if you were talking about a contextless single pass, certainly. But if those patterns are modeled in the codebase that is plugged into the harness for context, the model will absolutely replicate those patterns, and not naively either.

100 0

I understand quite a bit about how they’re created, but it’s essentially irrelevant to the question we’ve been addressing, which is “can current models effectively contribute to systems that are heavily divergent from their training set,” the answer to which is yes.

100 0

I didn’t miss that point at all, and I also never contested it. My point was that there was an implication that unless that outside data is somehow integrated to the training set, it can’t actually provide meaningful inference. That is incorrect.

100 0

Or to rephrase, not engaging with sealioning is not moving the goalposts.

000 0

I have not moved the goalposts a single time. My claim has been remarkably consistent, and then they make a different unrelated claim, and I say that’s great but the point was the machines can in fact do the work you’re saying they can’t.

300 0

You are entirely missing the point.

100 0

That’s if they are contextless. If you feed them an entire existing codebase and documentation, they do not perform poorly.

100 0

I think your understanding of what current models are capable of is outdated. You do not need the specific thing you’re working on to be in the training set. It’s perfectly capable of taking its existing training, taking some provided samples and documentation of an unfamiliar system, and inferring.

Anyone talking about “prompt techniques” is selling you some bullshit. Now it’s all configuring the harness for the specific codebase you’re working in which is an iterative tuning process until it Just Works. It’s reps, not skill. Solid systems understanding is still the only real relevant skill.

0100 0

Think this is unironically the actual story. Everyone is reading less because we’re all on screens but boys are on screens for longer than girls and had a head start in spending all their time on gaming and sports betting.

030 0

Yeah and I mean, this website. But somehow people who refuse to draw a distinction between the different types of uses in game development are perfectly content to draw a distinction here. Or simply pretend it isn’t happening. Eventually they’re gonna need to set real standards or log off, though.

160 0

And even if they aren’t using it directly, I 100% guarantee some of their tooling is. Unless you only plan to play games written from the ground up in assembly Chris Sawyer style, every game you play from roughly here on was made with genAI in some fashion.

1150 0

All this and more can be yours if we do the sensible thing: give Lina Khan a Gundam

0110 0

Yeah I think the “everyone’s agents will code what they need on demand” future just misunderstands people on a fundamental level, but if that were true the first thing everyone would do is never use a Microsoft product ever again.

0421 0