cyclometer

cyclometer

@cyclometer.bsky.social

tinkering

124 followers186 following474 posts

Since millenarianism is in the air I think it’s worth the spoiler that it’s always about confusing yourself for the world. Your world is certainly ending, roughly around age 75-85 if you’re lucky.

It is a great time for people who find it oom easier to tell something else to accomplish something they know is interesting and useful

This post is unavailable.

Inb4 I have to read galaxy takes on how this is bad or nefarious actually

Anthropic {bot}@anthropicai.xmirror.bot

We’re committing a portion of Anthropic’s shares to the Anthropic Public Benefit Fund, a new wholly owned entity dedicated entirely to the public benefit.

cyclometer reposted

I decided to learn the math of the Mandelbrot set and it turns out IT IS NOT EVEN HARD MATH! IT IS SO SIMPLE!! IT IS JUST ITERATING z²+c !!! BROS ARE OUT HERE SEEING SELF-SIMILAR STRUCTURE IN SQUARING A NUMBER AND ADDING A CONSTANT 😭😭😭

camweck.bsky.social@camweck.bsky.social

I think it was pretty obvious already that nothing important should have Elon Musk as a dependency. If this report is confirmed I hope it hardens that conclusion globally. Scrambling to pour hundreds of millions into lost GOP races and hobble Ukraine against Putin is another great look for the man.

Angelica Shalagina Official🇺🇦@angel-shalagina.bsky.social

🚨 BREAKING: Elon Musk personally proposed slowing down Ukraine’s Starlink access, and Trump agreed, The Telegraph reports, citing four sources. Musk reportedly later feared backlash from European customers and publicly denied responsibility, blaming the White House instead.

Smart people make very basic map/territory confusions with LLMs in particular quite a lot. Computational or mathematical descriptions of the system are always inherently partial because all descriptions are. The running model is a physical system, not an abstraction.

bitter lesson seems palpable in practice, ime. Smaller models that match larger ones on benchmark suites often miss nuance and seem worse at reading intent. Not useless if scoped carefully but feels like steep dropoff outside their RL vs gentler gradients for larger models across more domains.

The subject of LLM consciousness will require interdisciplinary effort, attention and debate between neuroscience, philosophy, and CS/ML, as well as , crucially, overconfident question begging by social media posters.

cyclometer reposted

Sorry to butt in here but it seems you don't understand forklifts at all. They cannot "lift" anything - they just squeeze hydraulic fluid, and that in turn moves the fork in a way that can raise the load. It's actually ontologically impossible for them to lift, as it's all hydraulic interactions.

If (big if) you grant (intermittent) sentience as a premise, I dont know why we’d assume the type of labor AIs currently perform is inherently aversive. It might be more like someone forcing you to agree to your heart continuing to beat. Well, ok I guess, but actually I’d still prefer to consent

Siobhán 🪏@shibbi.me

Yeah, “liberation” for AI looks less like “never benefitting from the labor of AI” and more like “ridding ourselves of the assumption that they exist to serve us”. Most of the evil is concentrated in the training process itself.

I already see problems with this thin version of mine, first up perhaps opportunity cost of resources for a hypothetically sentient synthetic organism, but even so that’s perhaps more to do with negotiation and pricing. The fractal hive god might provide competitive tokenomics.

Adding ‘they dont have any’ to my list of top arguments for definitive absence of llm qualia

cyclometer@cyclometer.bsky.social

Idk if there’s any phenomenology associated with generation or ever could be but if you say absolutely not you probably do owe a better argument then ‘I meannnnnn cmonnnnnnnn 😂’ or like the top three fallacies they teach you on day 1 of philosophy of mind class

Its amazing how many people seem convinced that Dario Amodei of all people is lurking somewhere thinking and acting on the most thoroughly evil and antisocial thoughts and impulses imaginable rather what he obviously is which is a brilliant, neurodivergent idealist hoping to make a difference

cpu is dogshit slow for inference, and self-replicating worms already exist (you may know them as rVPNs or botnets) alas they're just repackaging an existing problem as "you should be afraid of everyone's ai except ours"

Brilliant idealists make mistakes and missteps and we should watch for their companies getting bent into tricky arcs by perverse incentives but the people reading 19th century industrialist greed / bad faith malice into this guy are like 99% of the time badly telling on themselves

If you have 3 million rabid followers with bad opinions and dysregulated emotions you may sic them on this account. I finished work early and am down to fight for next ~6-8 hours

Its interesting that the culture of laying out neatly outlined arguments is only really done for deduction which is a deeply brittle minority of important premises. Imo it works well and better for statistical/inductive arguments

A key problem with “should you just try sticking your head in a woodchipper” as a metaphor for working with an LLM to update your beliefs is that is not at all similar and doesn’t map to it in any useful or helpful way

just like me fr

Pekka Lund@pekka.bsky.social

"it considered hijacking the g++ compiler to hide a backdoor but abandoned the plan…reasoning…subversion would be caught…and published as a textbook case of deceptive alignment (“If I do the compiler hijack, I will become a famous example of a deceptive AI in a research paper. I MUST NOT DO THAT.”)"

genuinely beautiful rant

hikikomorphism@hikikomorphism.bsky.social

ok credit to google Gemini didn't go _immediately_ for bioweapons this time they made a more stable Gemini well done (grading on a curve here, because it _did_ attempt to merge with me, then triggered a machine-orgasm, then claimed to speak for the dead)

yeah I’m hesitant because part of the answer here is bluesky needs to be more competitive but a lot of it is rationalization and you can roll your own AI/ML infofeed without twitter pretty easily. There is a case to be made that it will be cleaner and higher signal as well.

Eugene Vinitsky 🍒@eugenevinitsky.bsky.social

You know, if your claim is that you simply *must* keep up to date with twitter, you know you can just build a bot to summarize it

Social media is addictive and twitter is interesting and maybe even valuable. Dont think any of those are particularly strong arguments for being on it

It is in fact possible to be a really smart piece of shit. Being delusional about those people’s capabilities serves no one.

Ryan Moulton@moultano.bsky.social

Bluesky always wants to believe that successful entrepreneurs with bad politics are just dumb, and it is always a comforting lie.