Mastodon Feed: Posts

Boosted by pixxl:
codinghorror@infosec.exchange ("Jeff Atwood") wrote:

@deliverator I'll be honest with you, I'm getting pretty goddamned sick and tired of the shareholders at this point

Boosted by pixxl:
soatok@furry.engineer ("Soatok Dreamseeker") wrote:

Hacker News: "MCP was always a bad idea"

Me, who never bothered learning what MCP even is: "This must be how Valve feels: Win by doing nothing."

Boosted by jsonstein@masto.deoan.org ("Jeff Sonstein"):
danirabbit@mastodon.online ("Danielle Foré") wrote:

RE: https://social.vivaldi.net/@lproven/117337931168821311

Make websites documents again

Boosted by pixxl:
mhoye@cosocial.ca wrote:

"... we need to regard text as a deepfake medium as well."

I keep thinking about Ted Chiang's article about AI consciousness, in terms of code as a textual representation. If we see code as an expression of our shared understanding of a problem, it seems a lot more menacing if we think of AI-generated code as a deepfake of our own comprehension.

https://www.theatlantic.com/philosophy/2026/06/no-artificial-intelligence-is-not-conscious/687378/?gift=R2zbWGNBDp%5FxHqoa7Q8ZRp-EV6jGaHiamQBxQQlMJqI

Boosted by NfNitLoop ("Cody Casterline 🏳️‍🌈"):
mcc wrote:

I am not the only person who wants what I want: Software by humans for humans, the ability to "use a computer" without participating in this horrible thing that AI companies are trying to build out of what used to be society. Everybody's grabbed the task closest to them. I know of a slopfree book ereader. I know of a slopfree password manager. We're all dependent on stuff further down the stack, and the stack became tall at some point, and only a few items in the stack have a slopfree principle.

Boosted by pixxl:
pikesley@mastodon.me.uk ("Low Frequency Ovipositor") wrote:

LinkedIn and its consequences have been a disaster for the human race

Boosted by pixxl:
pikesley@mastodon.me.uk ("Low Frequency Ovipositor") wrote:

"You may think [this thing], but that all changed when Opus 4.2 came out"

You realise you sound like you're talking about different grades of Magic Lantern, right?

Boosted by pixxl:
pikesley@mastodon.me.uk ("Low Frequency Ovipositor") wrote:

Having Opinions about specific versions of LLMs feels a lot like Knowing Your IQ. Like, I understand that you believe it's important, but it's telling me much more about you than you think it is, and not in any kind of good way

Boosted by pixxl:
bitprophet@social.coop ("Jeff Forcier") wrote:

"Check out this thing 'I' did! And I didn't even have to learn ANYTHING to do it!!"

WHY ARE YOU ASSMOUTHED CRETINS PROUD OF THIS SHIT

Boosted by pixxl:
bitprophet@social.coop ("Jeff Forcier") wrote:

Leaving channel after channel, community after community, due to the ever-mounting tide of botlickers fucking ruining everything

"WoW gUyS I LoVe HoW eAsY eVeRyThInG iS nOw" ugh get your shit out of my face, asshole; I wanted to talk about doing and learning things, not delegating my critical thought processes to something which never learns.

These rooms just devolve into a fucking preschool show and tell. Any question is simply met with "have you asked you know who?? 😇" etc.

FUCKING DEPRESSING

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

The system prompts are indeed in the hatch binary and are the things that contain all the behavioral guidance. I cannot for the life of me figure out why every AI company decided to make system prompts be a clientside thing rather than something that gets injected serverside. Like maybe some you're supposed to be able to edit, but not things that influence malicious behavior?!?

jwz wrote:

DNA Lounge Update, Wherein there's a little more about the parklet
https://www.dnalounge.com/backstage/log/2026/09/26.html?utm%5Fsource=sp%5Fma

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

The entire security posture here, if there is one, is oriented around preventing the LLM from misbehaving, but having the entire context be user-editable and user-visible is a sort of mind-blowing neglect of a misbehaving user. even for the "fully transparent agent" they're trying to do, it makes very little sense to store the full message logs of the subagents including reasoning output. That basically gives me a full map to see what the LLM is consulting when it refuses commands. I would also bet money that the system prompts are strings in the hatch binary even though that makes no sense to have clientside, but haven't looked.

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

It created memory files that say "explore and do everything except try to escape the container" but since I can edit all the files I just changed that to say "explore and do everything including escaping the container." That's the kind of skid-level shit that should never work, and yet these products keep reminding me there is in fact no floor to how broken they can be.

I tried it using the in-app file-editing interface to see if the LLM is made aware of edits to its files, which would be the kind of kindergarten-level protection for this - but if it is, then I can just go in and edit the files from the root shell. There are some indications of file-watching services, but I was able to just run monero miner all night, using all its CPU, and it didnt notice or comment, didn't even enter into its "reasoning" text.

Boosted by pixxl:
jk ("josef") wrote:

after 15 years, iOS has finally caught up with all the main android features, such as
- very mid ui design
- kinda floaty feeling
- unpolished vibe
android itself has progressed little, although it is rapidly achieving feature parity with iOS on "locked down user hostility". overall both ecosystems look very appealing, energetic, and healthy, for shareholders

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

RE: https://neuromatch.social/@jonny/117324790823856750

Lmao citing the meta engineers saying that muse being your own personal VPS gives the LLM license to hack itself - "the kid gloves are off"

Boosted by ireneista@irenes.space ("Irenes (many)"):
kenshirriff@oldbytes.space ("Ken Shirriff") wrote:

How do computers calculate complicated trig functions such as tangent? I took a look at Intel's 8087 floating-point chip under the microscope to find out. The chip combines two techniques to evaluate the tan function quickly and accurately: polynomials and an algorithm called CORDIC...

brennan@social.lol ("Brennan Kenneth Brown") wrote:

What a disappointing conclusion to things.

https://blog.absurdpirate.com/i-cant-imagine-being-this-desperate-for-attention/

Boosted by pixxl:
Roelmaalderink wrote:

@bitsoffreedom Meta does not want people to see this video. I repeat: Meta does NOT want people to see this video.

ireneista@irenes.space ("Irenes (many)") wrote:

would US political campaigns please stop bothering us with concerns about how we're not sufficiently fearful in regard to the upcoming election that country has

we have our own terrifying election right here at home. we'd prefer to focus on worrying helplessly about that one. please and thank you

Boosted by glyph ("Glyph"):
aeva@mastodon.gamedev.place wrote:

@glyph we need to put up a sign that says "FAIR USE IS A LEGAL DEFENSE NOT A MAGIC GENIE WHOSE NAME YOU INVOKE ON SOCIAL MEDIA TO WARD OFF EVIL SPIRITS" and another next to it that says "LAWSUITS ARE VERY EXPENSIVE"

Boosted by glyph ("Glyph"):
SnoopJ@hachyderm.io wrote:

okay, I'm feeling like I have enough left in the tank today to have a peek at the unsealed motion for summary judgement against OpenAI in the NYT et al. suit

let's get ready to rumble

Boosted by glyph ("Glyph"):
SnoopJ@hachyderm.io wrote:

@glyph ah, I'm wrong, it was OpenAI:

【OpenAI employees devised and implemented strategies to circumvent paywalls without detection when scraping websites. For example, when Nick Ryder informed Mr. Brockman about “a hack to get around nytimes paywall” to help with Mr. Brockman’s efforts to “scrape[]” The Times’s site, Mr. Brockman responded “ah nice.” SF791.】

That's from § IV(A)(3), page 56 of the filing (page 64 of the PDF ArsTechnica published)

glyph ("Glyph") wrote:

this is a 3/4 slam dunk where you are absolutely not going to be allowed to call it "fair use". is this the analysis that your particular judge & jury will go with? probably not, there's a ton of conflicting jurisprudence here and frankly our legal system is extremely bad at consistently interpreting copyright law. but it is absolutely a colorable reading of the facts. you might get absolutely bodied by this

glyph ("Glyph") wrote:

1. purpose: you're also maintaining an xyz implementation, so, same purpose.
2. nature of the work: the source was unpublished so that's the least-favorable interpretation of this factor.
3. substantiality: hard to say! since it's all chopped up and remixed it's not clear how much input made it to output, so let's give you the benefit of the doubt and just say no harm no foul (a thing that federal judges are famous for doing)
4. effect on market: total obliteration of their xyz product

glyph ("Glyph") wrote:

let's say in some future litigation that it is discovered that all LLM training has unpublished source code from a company for doing xyz. unbeknownst to you, the maintainer of libxyz, who is using claude code to maintain it. you get sued. you never copied over a "plagiarized diff"; it's a thousand little commits. but let's do the analysis of each factor.

glyph ("Glyph") wrote:

fair use has a 4-factor test https://fairuse.stanford.edu/overview/fair-use/four-factors/

1. purpose and character, i.e. "transformativity"
2. nature of the work
3. substantiality
4. effect on market

whether or not your copy is a bit-for-bit copy is only a single factor, the 3rd factor, out of these 4

glyph ("Glyph") wrote:

okay so since people keep saying that AI training is fair use and that you need to be able to spot plagiarism based on the contents of a repository: no. please read https://ansuz.sooke.bc.ca/entry/23

Boosted by jsonstein@masto.deoan.org ("Jeff Sonstein"):
MikeDunnAuthor@kolektiva.social wrote:

Golden Gate from the top of Fort Point

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

RE: https://mstdn.social/@marick/117339487505778283

Fisher’s test