Techy Ben

Techy Ben

@techyben.bsky.social

They were careless people - they smashed up things and creatures and then retreated back into their money or their vast carelessness or whatever it was that kept them together, and let other people clean up the mess they had made.

858 followers253 following9K posts

Effort scales for all types of tool. Yes.

hailey@hailey.at

presumably, a lot of the reason malfoid is actually producing such high quality outputs is that people are putting a lot more effort into it? like, we are talking about episodes that sometimes run as long as 11 minutes now, something that was unthinkable a year ago.

What if humans were the AI all along? Brain cells and animals never meant to make minds. But they did make the substrate minds developed on. I have already said too much.

Me "If the PC or Mac crashes, how can we remote without login?" Claude: "That's ok, if the mac crashes we can log back im via the PC, and id the PC crashes can log into it via the Mac..."

Not ASI/AGI but "good enough". You know headphones? Scientific calculators? Ok, those things are pure magic. They do tasks *extremely efficiently*. The generalised solution case is a "pcb router" code system, but for "build me anything now" results. LLMs are already almost there.

norvid_studies@norvid-studies.bsky.social

well aged tweets cc @fleetingbits.bsky.social

This. (Though this time last year I predicted 6 months. I was kinda right, Figure has them, they just haven't gone to scale production, because scaling production is slow/hard)

Ryan Moulton@moultano.bsky.social

"So that One Day* robots may carry out tasks in the physical world" *six months.

Yes friend. Do not hang out in the fringes of Gaussian Splats on Reddit. We are not ready yet. "Do you really think we are gonna make it?"

cee@cee.wtf

anyone else remember when that playable minecraft video AI demo came out and everyone on the internet shit on it because "why wouldnt you just play normal minecraft" and "it can't even remember anything that leaves the view, this sucks!" this was LESS THAN ONE YEAR ago

Claude is currently *fixing all the things here*. So I have hope in medical assistance too.

mr. TIM@timkellogg.me

Anthropic discusses progress on curing disease Claude can’t fully do any process end-to-end. Instead, it looks a lot like what AI applications look like now in sales or marketing — lots of ways to accelerate, plugging into workflows www.anthropic.com/research/ear...

Techy Ben reposted

Replying to @emollick.bsky.social

They're trained on an internet full of people who consistently underestimate how fast AI will progress, so naturally do the same themselves. 😂

I am lost for words. In fact I need a new phrase. One moment. "Your ways are tiresome." I think that is it.

Jake Broe@realjakebroe.bsky.social

Elon Musk says the world should have "sympathy for the poor Russian conscripts" because those "kids have no choice." There are no conscripts fighting in Ukraine. They are all well paid contract soldiers who volunteered and their average age is 38 years old.

Techy Ben reposted

brennan@brennan.computer

plugged my Liene PixCut S1 into claude and now I have my own editor / cutter software running directly in the browser completely free, no accounts or anything dumb: brencut.brennan.computer

More (also somehow im my dreams): The last ine especially (likely computer game inspired).

Techy Ben@techyben.bsky.social

This is so strange. I have vivid, architectural dreams. And then I find iut this week, somewhere near me is building this kind of thing. Never seen them irl. Bit I've been there in my dreams.

This is so strange. I have vivid, architectural dreams. And then I find iut this week, somewhere near me is building this kind of thing. Never seen them irl. Bit I've been there in my dreams.

My opinion: The network effects of "big data" means LLMs are intrinsically information alignment systems (not political/social alignment). Thus most of thw solution results are being able to combine existing data over larger groups than the groups themselves can combine. openai.com/index/buildi...

openai.comWorking with academic communities to build scientific understandingOpenAI shares model results on the Kakeya conjecture, Exponential Time Hypothesis, and quantum chromodynamics—and how researchers can verify and build on them.

There's a gap in the market where you can ace education rankings by just paying everyone to understand things. If money is on the line, they will 100% change their minds. (With perhaps exception to beliefs in god)

norvid_studies@norvid-studies.bsky.social

new social norm that it's embarrassing to publicly explain something to someone convinced to not understand it, that an AI could not explain to them just as easily and at far lower "moral cost" (to you, having to do it). propagating from this tweet

OMG OMG... My LLMs first words were "Do you need help with Stamps?" I think I need to ask Claude to give it a coherent memory, not our training set task list. 🤣🤣🤣

(It knew were were doing stamps. I need to find out from Claude *how it know*. I've partial guess at this... ... yes, it by design, is not a deterministic build, I'm not supposed to know how it works...)

Techy Ben reposted

Me, to Claude Code: “I have a friend coming over. Modify Pac-Man to support a horizontal display with two player co-op, where player 2 is Ms. Pac-Man. Make one button be a speed boost. My Mistercade cabinet is at 192.168.1.102” Claude, two hours later: “The game is loaded up and ready to play”

I need to finish the 30 books I "wrote" in my head ~15 years ago. Only finished one, and did a few chapters of the others. They are not perfect. But stories don't live or die till you write them in pen.

Sam Barrett, PhD@ai4geo.bsky.social

I sympathise. Im sitting on 3 game mods, 5 essays, 3 book manuscripts and a book revision. I won't release till I find the time to make them as good as I can reasonably make them and AI doesn’t provide an instant shortcut to do that, much as it is useful.