a2128

30 ms·

241.
by a2128 2y ago
Trump says that he wants to finish "fundamentally reevaluating NATO's purpose and NATO's mission", and "the greatest threat to Western Civilization today is not Russia. [...] It's the Marxists who would have us
242.
by a2128 2y ago
This design choice is necessary because Android keyboards aren't just keyboards, they're input method editors. Software receiving simple codepoints as input works great on English desktop computers where every inputtable character
243.
by a2128 2y ago
3DoF without tracked controllers is not VR IMO, it's just a head mounted display. It's not sufficient for any VR use case other than like watching a movie. You won't be able to play any modern VR games. Maybe you'd be ab
244.
by a2128 2y ago
Well it has no controllers nor any controller tracking hardware, so you won't be playing much VRChat with it...
245.
by a2128 2y ago
I don't get the logic of showing ads here. Premium users are paying $14/mo, surely that should completely dwarf the fractions of a cent Google would earn by displaying an ad. Is this just a simple bug? Or are advertisers now payin
246.
by a2128 2y ago
It just uses LLMs, I've had it output a refusal in the target language by entering stuff about nukes in the input
247.
by a2128 2y ago
I can only hope you're right. But I fear that the reality will be closer to simply a withdrawal of US support for the defense of Europe and NATO. It is ignorant to believe that Ukraine's capitulation would mean that everyone lives
248.
by a2128 2y ago
Half of the rewritten titles will be "Review Our Privacy Agreement" and "Trial Over - Subscribe to Read Our Articles" :)
249.
by a2128 2y ago
I haven't read how Google is doing it, but one way it could be done is to nudge which tokens get sampled. For example, every other token could have an odd numbered id (where each token is assigned an id from 0 to 32000 or however many
250.
by a2128 2y ago
Moshi never released the base model, only two conversationally finetuned models. They also never released training code except for the codec. Though I don't see any training code for Hertz either, just 3 inference notebooks, and model
251.
by a2128 2y ago
https://github.com/yt-dlp/yt-dlp/issues/10128
252.
by a2128 2y ago
A few years ago I was concerned of this being a genuine possibility (at least for a few months) due to regional conflicts, so I prepared the following on some SD cards and flash drives: * Offline copies of Wikipedia, Stack Overflow and some
253.
by a2128 2y ago
It's not uncommon that I go to Shadertoy and see strange visual artifacts in some shaders including window contents of other applications running on my system, potentially including sensitive information. It's difficult to make GP
254.
by a2128 2y ago
Normally you have to explicitly install and use a cloud stored password manager, automatically making the choice for you is a big no-no
255.
by a2128 2y ago
It's usually wildly uneconomical to serve such large models yourself unless you're serving a massive amount of users that you can saturate your hardware. Thus most people will opt for hosted models, and most of the big ones will c
256.
by a2128 2y ago
How does that scale? Handcam anticheat works well for exams and Olympiads where there's limited people and plenty of time to review footage after it's over, but I can't imagine it would work well on Fortnite or Counter Strike
257.
by a2128 2y ago
I definitely think it's just a business decision being made in some cases. Your developers have just built and demonstrated a functioning multiplayer prototype. They want to spend 3 months to rewrite some of it for better security, and
258.
by a2128 2y ago
From my understanding, if you play an EAC game on Linux with Proton, you're not really running the same EAC as Windows players. You're running a lite version that runs as a regular user and it tries to provide at least some level
259.
by a2128 2y ago
This is one of my big gripes - when apps are trying to get ahead of me with validation or submission. When I'm entering a 2fa code, I don't need a big red error telling me that the code must be 5 digits before I'm even finish
260.
by a2128 2y ago
A bit like that Wired article about how the web became unreadable: https://i.imgur.com/xZFeFa3.png
261.
by a2128 2y ago
If AI agents take off, we might see a new rise of scam ads. Instead of being made to trick humans and thus easily reportable, they'll be made to trick specific AI agents with gibberish adversarial language that was discovered through t
262.
by a2128 2y ago
In the future, everything will block you unless you're on a residential IP address in a Western country, running Chrome on the latest version of Windows or iOS, you have remote attestation, and your ISP's ASN isn't haunted be
263.
by a2128 2y ago
On Linux, most modern user-facing applications are using polkit instead of sudo. You can actually just use pkexec instead of sudo in the terminal as well. Instead of just running arbitrary commands as root, applications can use specific pre
264.
by a2128 2y ago
A federated network is a decentralized network. You're probably thinking of distributed networks, which federation is not.
265.
by a2128 2y ago
The usefulness of data here is that you can retrain the model after making changes to its architecture, e.g. seeing if it works better with a different activation function. Of course this is most useful for models small enough that you coul
266.
by a2128 2y ago
I wonder at what point does increasing LLM throughput only start to serve negative uses of AI. This is already 2 orders of magnitude faster than humans can read. Are there any significant legitimate uses beyond just spamming AI-generated SE
267.
by a2128 2y ago
The Vision Pro feels more like a spatial tablet than a spatial computer. A computer should be capable of handling productive work tasks such as developing software. Vision Pro's only relevant capabilities seem to be mirroring your comp
268.
by a2128 2y ago
Releasing it as a feature everywhere would require at least a bit of public explanation and might give the appearance that EU forced Apple into doing something good for everyone. Being petty and releasing the bare minimum georestricted only
269.
by a2128 2y ago
* Fixed bug where Claude would sign up for chatgpt.com to ask for help with compile errors
270.
by a2128 2y ago
To provide an example, the recent Llama3.2 release includes a clause in their acceptable use policy that says any individual or business located in EU has no rights to use their multimodal models. This is discrimination against persons&#x2F

More ›