BlueFalconHD

The framework loading these is in Swift. I haven’t gotten around to the logic for the JSON/regex parsing but ChatGPT seems to understand the regexes just fine

Nope. This is a separate system. It’s not even abstracted for any asset, it is specifically only for these overrides. The decryption is done in the ModelCatalog private framework.

Yep. These filters are applied first before the safety model (still figuring out the architecture, I am pretty confident it is an LLM combined with some text classification) runs.

This is definitely an old test left in. But that word isn’t just a silly one, it is offensive (google it). This is the v1 safety filter, it simply maps strings to other strings, in this case changing golliwog into “test complete”. Unless I missed some, the rest of the files use v2 which allows for more complex rules

This is definitely the right answer. It’s just testing stuff.

These are the contents read by the Obfuscation functions exactly. There seems to be a lot of testing stuff still though, remember these models are relatively recent. There is a true safety model being applied after these checks as well, this is just to catch things before needing to load the safety model.

There is definitely some testing stuff in here (e.g. the “Granular Mango Serpent” one) but there are real rules. Also if you test phrases matched by the regexes with generation (via Shortcuts or Foundation Models Framework) the blocklists are definitely applied.

This specific file you’ve referenced is rhetorical v1 format which solely handles substitution. It substitutes the offensive term with “test complete”

One additional note for everyone is that this is an additional safety step on top of the safety model, so this isn’t exhaustive, there is plenty more that the actual safety model catches, and those can’t easily be extracted.

x mixed with y = join any combination of x and y based on some condition within x and y x augmented by y = change part of x by adding y

x = real world y = virtual world

What about "Meta Horizon SDK for Android SDK for Linux?"

Text based layouts should come back. Search "TextOS" on twitter for examples. If someone finds it, post a link below (responding on a Kindle Paperwhite, twitter won't run)

Came here to comment the exact same thing. Also the arrow on the login sceren is actually centered until the field is focused AFAIK.

This is super cool. I might end up rewriting in JS or a language with a WASM compilation toolchain.

Also unusable on simpler hardware. The browser on the Kindle Paperwhite I am typing this on is just slightly too old to run Twitter. I get the unsupported browser page, which funnily enough still uses the old colors and logo.

"I have become death, destroyer of worlds"

this [1] but with handcuffs

[1] https://i.redd.it/obvqaa0lhn841.jpg

If a US gov. agency is "Jia Tan" then this might not happen.

If some nation state actor dis this, I imagine they couldve easily modified times of their commits, takin into consideration holidaya and such. Also, nation state actors have tons of resources. They could have people waiting to make commits or schedule it with something like cron.

Like that spiderman meme, it's all the NSA

What does the IME or PSP do?

It would be really cool if in 20 years when we have quantum computers powerful enough we could see what this exploit does.

Those hackers might get what they wanted. Real life catgirls. It would be so awesome.

Never used koreader, does it have plugin support? If so something like this wouldn't be too complex to integrate.

But the government hates online privacy. It stops them from spying. And (though I don't know this for a fact) I feel tech giants could be lobbying against privacy laws.

Swift is love, Swift is life. Wonder how well it would run on the new Arduino (r4 afaik)

A solution couls be a model trained on the exact timeline of some text being typed that can predict how long it will take for the user to type the predicted text

eg. "I need a plane ticket to Ha" - 730ms -> "I need a plane ticket to Hawaii"

The model would detect deviations from the estimated time and invoke the main LLM. This could work for spoken word too, it would just be trained on real speech instead of typing.

Would this use websockets or the like to send your text input to an AI? Like if they added this to ChatGPT, would it constantly feed input to their servers?

I could see the possibility for new special tokens. Think of terminal escape sequences. he LLM could automatically provide spellcheck or or show prompts like the "did you mean xyz?" on Google.

agreed

If you use a router, manually connecting wires to send data is awesome

Did you design the logo? I saw it and immediately wanted to buy a license! Great product!