Mastodon Feed: Posts

Boosted by baldur@toot.cafe ("Baldur Bjarnason"):
jcoglan wrote:

somehow I only recently learned that LLMs need the whole conversation fed back to them on each prompt so their i/o cost scales as O(n^2), something that would be considered completely unacceptable in almost any other production network-accessible software

Boosted by baldur@toot.cafe ("Baldur Bjarnason"):
mhoye@cosocial.ca wrote:

Cannot emphasize this enough: the single fact that made the software business as insanely profitable as it is is that the incremental cost of delivering software rounds to zero.

The entire provisioning model of AI as a business has broken that.

It's not that LLMAAS isn't profitable or won't be profitable. It's that LLMAAS can't be profitable.

https://mastodon.social/@jcoglan/116968429653967850

dysfun@treehouse.systems ("gaytabase") wrote:

agentic poker

cwebber@social.coop ("Christine Lemmer-Webber") wrote:

There's a pretty simple reason for this: all the large AI corporations are betting on becoming centralized gatekeepers over information and its production. So any chance of a better future involves failure of said business models succeeding

cwebber@social.coop ("Christine Lemmer-Webber") wrote:

The only variations of an ethical future which incorporates "AI" are ones where large AI corporations are unprofitable

cwebber@social.coop ("Christine Lemmer-Webber") wrote:

Also I spent some time on the @spritely call today talking with @dthompson and @tsyesika about what the architecture of this machine would look like, and both of them thought it sounded very interesting, and so given two of the engineers I trust most in the world think it's interesting, it might indeed be worth writing about.

Boosted by aredridel@kolektiva.social ("Mx. Aria Stewart"):
ShadSterling wrote:

@aredridel I feel like news got turned up to 11 a while ago, and hasn’t been turned down since.

Which is a very different meaning of “news at 11” than I grew up with

cwebber@social.coop ("Christine Lemmer-Webber") wrote:

It's too bad the "Claw" space is already occupied in "AI" tech, because given my initials are CALW, it would be funny to transpose a couple of letters and call the design I am talking about a "Claw Machine"

cwebber@social.coop ("Christine Lemmer-Webber") wrote:

I think I would like to write a couple of blogposts:

- A rumination over whether or not a "more ethical AI future" is possible *at all*, the post of which is bit more agnostic as to the technical approach taken
- And, I think there was enough interest shown in the last hour, coupled with a *disbelief that any other approach is possible* (which still irks me), that maybe writing up the "Christine Approach to an AI Machine" (which is NOT achieved by beating LLMs to death with more data) might still be worthwhile and interesting.

baldur@toot.cafe ("Baldur Bjarnason") wrote:

It's bad enough to have to have the "former manager now works in the military-industrial complex, what's up with that and do we need to distance ourselves from him?" conversation with former coworkers, but that it's not the first time I've had to have that chat is honestly a bit disconcerting.

cwebber@social.coop ("Christine Lemmer-Webber") wrote:

Well the poll has ended, and holy moly, there was a lot of response here, both to the poll and also in the comments. It was a dead heat for a while there, and still came pretty close to it.

Within the last hour, I kind of skewed the results by throwing this out there, that I *do* have thoughts on how to build a system that isn't based on LLMmaxxing which fulfills these properties, and that seemed to bias a lot of the votes coming in within the last hour towards yes https://social.coop/@cwebber/116969383379862650

But still, a lot of people are clearly disturbed, that any kind of improvement to automation of intellectual labor, even with all of the *main* things that we tend to complain about with *present-day* (and LLM-based) AI removed, making those "better" is unclear if a win given the lack of a change in socioeconomic context.

So... what now?

Boosted by cwebber@social.coop ("Christine Lemmer-Webber"):
cwebber@social.coop ("Christine Lemmer-Webber") wrote:

A hypothetical: Imagine I held a magic wand that could make AI systems improve in all the following ways:

- More environmentally friendly
- Less capturable by a few large corporations
- Less built on "plagiarism machine" foundations
- More reasonable and capable of making judgements which are explainable, accountable, and more accurate
- More participatory and open

However, we are still left in our present socioeconomic system.

Should I wave the magic wand and improve things, or does improving AI in these ways still make the world worse?

Boosted by soatok@furry.engineer ("Soatok Dreamseeker"):
redsad@ohai.social ("captain acab :antifa: 🏳️‍🌈") wrote:

about last night

Boosted by soatok@furry.engineer ("Soatok Dreamseeker"):
SuperDicq@minidisc.tokyo wrote:

Boosted by brib@bribstodon.xyz ("brib :neofox_floof:​ :Nonbinary:"):
amras@friendmaterial.lgbt ("Sarma / Amras") wrote:

Corrolary: if you do know your way around a server, what do you wish more users understood about digital infrastructure and the web, because it'd make the world better if they did?

Boosted by cwebber@social.coop ("Christine Lemmer-Webber"):
carnage4life@mas.to ("Dare Obasanjo") wrote:

I can’t imagine being the literal richest human being in existence and making my entire online persona that of a racist incel.

aredridel@kolektiva.social ("Mx. Aria Stewart") wrote:

This actually affects my sense of humor: these things aren't relevant to my life, but the sentiment of "This is hype" or "This is obvious but being treated as news" doesn't have an exact translation in a modern context, and _that_ is what makes me feel old.

It's not that I ever watched network news, but that hyping up a show to come later is not a thing we do now, the very way we make mountains out of molehills has changed.

brib@bribstodon.xyz ("brib :neofox_floof:​ :Nonbinary:") wrote:

#Sysadmin tip: You can set up unattended-upgrades (Ubuntu instructions here, also works for Debian but Debian doesn't turn this on automatically) to deal with security updates within a day of patching without having to do it manually.

I also recommend installing needrestart so that important services get restarted on upgrade, reducing the chance of vulnerabilities staying live for a long time.

For kernel updates, it appears you can configure it to automatically reboot with the Unattended-Upgrade::Automatic-Reboot "true";. This will of course create a small amount of downtime for the server, which I consider an acceptable trade-off for a small server like mine.

There are also ways you can livepatch kernels iirc but I don't know those techniques so well, you might want to look into them for a larger/more professional server.

Keep patching and stay safe out there!

aredridel@kolektiva.social ("Mx. Aria Stewart") wrote:

Damnit, another phrase I realize is in the dustbin, "news at 11"

Getting culturally old is _weird_.

cwebber@social.coop ("Christine Lemmer-Webber") wrote:

RE: https://social.coop/@cwebber/116964060609145830

Just over an hour left to vote in this poll. And it's pretty interesting; a lot of votes, and a dead heat.

And there's a lot of disbelief about whether or not such a machine can *even be built*.

So let me up the stakes: this poll is informing whether or not to do a writeup on how to build *exactly this system*.

And I'm currently leaning towards writing it up, if for no reason other than because I'm annoyed I'm being told there's no way to do this other than just LLM-maxxing it, both by LLM fans and detractors.

Boosted by brib@bribstodon.xyz ("brib :neofox_floof:​ :Nonbinary:"):
quincy@chaos.social ("Oenanthe conioides ⁂") wrote:

the dreaded double space

Boosted by brib@bribstodon.xyz ("brib :neofox_floof:​ :Nonbinary:"):
david_chisnall@infosec.exchange ("David Chisnall (*Now with 50% more sarcasm!*)") wrote:

@datarama

The industry has always had some of those people. During my career, the set of things that, if you're not doing them, you're wasting your time and everything you've done is pointless includes (non-exhaustive list, in no particular order):

• GUIs (command-line tools are obsolete).
• Web apps (native GUIs are obsolete).
• Mobile apps (desktop and web GUIs are obsolete).
• OOP (procedural programming is obsolete).
• FP (OOP is obsolete).
• Reactive programming (MVC is obsolete, everyone must use a paradigm where no one seems to be able to manage to do a simple master-detail view without it getting out of sync).
• COM (shared libraries that don't expose COM objects are too hard to maintain with stable ABIs and too hard to expose to higher-level languages).
• Multithreading (Moore's law is dead, everything must use a million threads to be relevant)
• Thread pools (using too many threads is too slow)
• Structured concurrency (manually managing runnable work items is too deadlock prone).

Each one of those has come with a bunch of people happily cheering that everyone not on that particular bandwagon is wasting their life.

The common factor is not that these technologies changed the world and obsoleted everything before them (many of them did improve things), it's that those people were dicks. And they remain dicks when they're talking about AI.

Boosted by brib@bribstodon.xyz ("brib :neofox_floof:​ :Nonbinary:"):
selfcare@goto.toydragon.xyz ("Fedi Selfcare Reminders") wrote:

May you have the courage to break the patterns in your life that are no longer serving you

dysfun@treehouse.systems ("gaytabase") wrote:

some reply entities sure seem keen on local models. in fact local models are a magic antidote to any talk of bubbles, or so i guess i'm supposed to believe.

aside from the fact that the "good" ones are not runnable on even expensive consumer graphics, there's just one little problem they never seem to acknowledge - when the bubble pops, they'll all be frozen in time because no-one will be able to justify the cost of training better ones.

Boosted by brib@bribstodon.xyz ("brib :neofox_floof:​ :Nonbinary:"):
brooke@bikeshed.vibber.net ("Brooke Vibber :neocat_cofe:") wrote:

this old vlc app update description came up on my facebook memories ;)

Boosted by brib@bribstodon.xyz ("brib :neofox_floof:​ :Nonbinary:"):
mrpjevans ("PJ Evans (EMF x2001)") wrote:

Hugs to all the post #EMFCamp washing machines today. You got this.

jsonstein@masto.deoan.org ("Jeff Sonstein") wrote:

well now, time for difficult choices I guess

Boosted by cstanhope@social.coop ("Your weary 'net denizen"):
phoenix@pawb.fun ("Phoenix") wrote:

Wrote a nice big blog post about my VHS simulator that explains all the physics models inside of it (and finally gives a basic reference to what all the parameters mean).

https://blog.opengbh.net/posts/0017-vhs-playback-simulator/

#vhs #retro #retrotech #retrocomputing

jonny@neuromatch.social ("jonny (nonvenomous)") wrote:

RE: https://mastodon.social/@proseandpassion/116963759618141026

undeniable that the cello is what wood is supposed to sound like and then you're supposed to play it like this

Boosted by jonny@neuromatch.social ("jonny (nonvenomous)"):
proseandpassion ("michael") wrote:

I actually have the sheet music for this: https://www.youtube.com/watch?v=hnMEnhvPZkE #cello #kobekina