Fredkin

Frontier models have seen more Mathematica and Powershell code than I ever have in their training, yet they really struggle to produce working output. They seem heavily tuned to Linux too. Despite adding skills to rectify this, they still fail to realize they're running on Windows and waste tokens. A human with this much training wouldn't have this problem. There are evidently still some pretty big holes still.

I have spent a small fortune in tokens trying to get Anthropic models to attempt proofs of things like the irrationality of the Euler-Mascheroni constant, and to make lattice reduction / SVP algorithms that rival SOTA methods like the G6K sieve, and all sorts of difficult frontier problems. It says 'honest' as a euphemism for giving up, because it knows I'm being a crank, and because the training data contains a lot of pessimism.

But I can get around this by first having it write a prompt for itself discussing the problem followed by a rousing speech about never giving in and admitting defeat, but pressing on towards results regardless of how hard matching SOTA results is.

Lots of other point collisions too: {{{-1, 3/2, 13/2}, {1, -3/2, 13/2}, {0, 0, -1/4}}, {{0, 7/8, -(3479/1152)}, {48/7, 0, 49/1152}, {-(48/7), 7/16, 539/2304}}, {{-1, 7/8, -(40523/648)}, {160/7, 0, 49/5184}}, {{-2, 7/8, 2989/1296}, {32/21, 0, -(49/192)}}, {{1/63 (-120 - 16 Sqrt[3]), 7/8, 0}, {4/7 (-1 + 3 Sqrt[3]), 0, 49/288 (5 + 3 Sqrt[3])}} }; Next steps would be a simpler counterexample or family of polynomials where the conjecture breaks.

If they can't distinguish traffic containing hidden encrypted messages from humdrum non-encrypted traffic then they'd have to ban the whole thing for everyone.

Maybe a dumb question, but what's to stop people from communicating e2e encrypted over totally insecure channels using steganography techniques?

You don't need a special app to do this, or maybe you just need a companion app that you type your message into and it gives you the thing you just paste into whatever messaging app / social media you use. The steganography makes it hard for the operator to determine that you're "abusing" the service by not transmitting your message in the clear so they can read it.

1) Alice uses steganography to embed her public key in an otherwise innocent or mundane looking image e.g their profile picture.

2) Bob uses the public key to encrypt a short message to send her.

3) Bob embeds the encrypted message in his own mundane looking image (could generate these from a pool of images or on the fly using stable diffusion)

4) Bob sends the image to Alice.

5) Alice recovers the encrypted message and decrypts using her private key.

(Could also use the process to do key encapsulation too, instead of using the raw key pair)

"Commissioning the battery requires a device capable of running the BYD Be Connect App - available for the Apple App Store and Google Play Store." - No thanks. Just have a control panel on a lcd screen ffs. They also do an inverter "Fronius Primo GEN24 Plus" that requires an internet connection. Completely unnecessary.

Subscriptions make companies lazy and it degrades the product. I'm looking at you: Foundry, Adobe, Maxon, heated seats on BMWs ...

They rest on their laurels, enjoy the increased cash flow, say it allows them to work on regular updates. But this goes from being useful bug fixes, to merely shuffling the UI around, changing the fonts, introducing nonsensical features nobody asked for or can make use of, and gutting useful features for "streamlining" purposes... while longstanding bugs that actually need fixing are still unfixed.

Eventually customers become dissatisfied with the product and make up for lost features and degraded user experience with a smörgåsbord of perpetually licensed or FOSS alternatives from various competitors because they too will want to improve their cash-flow instead of being bled dry every month.

Companies that choose to offer lump-sum permanent licenses have to make a bigger effort to convince customers to upgrade, which means the product improves. Also it makes your customers more committed to your product. You should invite this kind of challenge and forgo the temptation to boost cash-flow because it keeps you on your toes. Subscription-only will seem great for a while but eventually you'll atrophy and fail.

Something similar happened when software went from being released on CDs/DVDs to regular patches and downloads. Not saying we need to go back to that era, but QAs had to work harder back then because distribution was expensive. Nowadays you can release things in an unfinished and broken state.

What does it mean to train using an 'open ended' process? Is it like using a genetic algorithm to explore / generate _any_ image resembling something from the training set, instead of adjusting weights according to gradients on a case-by-case or batch-by-batch basis?