infinityio

It looks like this is a Zen 2-only exploit, so it shouldn't have any impact - AWS are likely already running hardware that isn't vulnerable to this

For a recent python install, python should be available under `python`, but you should also have the `py` utility installed - this is normally helpful in selecting from a set of versions. If neither of these resolve, the first step is probably to reinstall python

With regard to installing packages, I think the general rule is that `python -m pip` is best practice, because it assures that the python you install packages to is the same one you're planning to run them on

No - the law that may end up requiring that is a proposal that is at least several months from being enacted [0], although it's been gaining momentum for several years - so unfortunately is looking like it might actually happen

[0] https://www.bbc.co.uk/news/technology-6606560

Just to add some points of reference:

- VR headsets typically need very low frame times so that they can account for you moving your head and not give you motion sickness. The typical threshold is ~10ms, but it can be improved with good reprojection tech

- Older LCD TVs often added >50ms of input lag, although this is much less of a problem now, but this was enough for lots of people in the 360/PS3 era of consoles

hCaptcha has been doing this for a few months now to my knowledge - I'm guessing this is being used to train a CLIP-like system (or alternatively: it might just be that poorly generated ai images are the best candidate for a captcha in the age of modern subject recognition, it's hard to know)

> If you routinely pass in a dataset to AI, it can identify anomalies and watch to see if it settles or even reason why the anomaly occurred in the first place. You can provide the AI some guidelines to only alert when something needs attention if it meets a certain set of criteria. This ultimately increases the effectiveness of your on-call crew and lowers your TCO purely based on manpower.

> ...

> So you could use ChatGPT to identify activity trends of an individual to get a statistically high likelihood of reaching them online. Imagine this audit history

> [dataset of login and logout timestamps] `Return a json object with a "sendAtTime" property in ISO-8601 format that has the highest likelihood of reaching the person the next time they are online.`

This feels like it is slightly conflating the strengths of LLMs and AI as a whole.

There are certainly machine learning models that would assist in predicting when a user will be next online, and using an AI assistant to assist in writing such an implementation would likely be helpful.

However, passing JSON into an LLM, asking it to emulate a program to perform this functionality, and then parsing the response to extract other JSON doesn't feel like a stable approach (for example: I was unable to recreate their successful execution under ChatGPT v3), especially given this technique, if widely adopted, may end up including user-controlled fields that would make the system vulnerable to prompt injection.

The last time I worked on a raspberry pi-based project I made extensive use of pinout.xyz, and this new page looks to be a great evolution - thank you for creating them!

From the sounds of it, three wrong password attempts (from a phone or similar) could accomplish this? That should stop the toothbrush from being able to add extra time to the head

(not arguing that discord should be used for these kinds of projects, just an aside)

to my knowledge, this is is only an optional moderation feature on some servers, not a global policy - you can still use discord without an email address if the server rules are loose enough

The quality does not yet measure up exactly to ChatGPT (even 3.5), but yes it is possible

Probably the fastest way to get started is to look into [0] - this only requires a beta chromium browser with WebGPU. For a more integrated setup, I am under the impression [1] is the main tool used.

If you want to take a look at the quality possible before getting started, [2] is an online service by Hugging Face that hosts one of the best of the current generation of open models (OpenAssistant w/ 30B LLaMa)

[0]: https://mlc.ai/web-llm/ [1]: https://github.com/oobabooga/text-generation-webui [2]: https://huggingface.co/chat

not quite $15, but Amazon US seems to have PNY cards with fairly promising reviews for $16.99 at the moment (although they may be out of stock)

Worth noting that Google tried (and failed) to solve this specific problem

https://www.theatlantic.com/technology/archive/2017/04/the-t...

unfortunately in this case the text content was handwritten, not computer-generated

I think copilot is free in the education pack - you might have got it from there?

As of today you can download LLaMa/Alpaca and run it offline on commodity hardware (if you don't mind having someone else do the quantisation for you) - the cat's out of the bag with this one

in general, assume 2GB per billion parameters - with quantisation you can get this down to <1GB (~500MB for 3 bit?), but even with that you'll only be able to run quantised llama-13B in the best case

Having said that: if you are feeling incredibly patient you can technically run the 68B parameter model by swapping to disk, although it will not be a pleasant experience (think minutes or hours per token instead of tokens per second)

Additionally worth noting pure CPU inference is much slower than GPU/TPU inference, so the output will be much slower than a ChatGPT-like service even if it does fit in your computer's RAM

(subscriber for the past 8 months, planning to convert to annual to lock in $10/mo for the next year and reassess then)

The AI tools they have been releasing are interesting, but Kagi desperately needs to decrease their average cost-per-search so that they can offer more realistic numbers of queries per month at a $10 price point.

I'm not sure what the solution is - maybe generating smaller indices for particular verticals? For some classes of question (for example looking up documentation) I would expect there are a surprisingly small number of sites that consistently end up as top 3 results, and everything else can be put behind a 'load more' button that queries the paid APIs.

(The problem then becomes identifying when this is the case)

While it wouldn't 'fix' the cost of long tail searches, identifying 'easy' searches and answering them directly might help - looking at my own search history, I'd say maybe 1 in 3 of my queries would fall under this category.

to my knowledge - started 100% non-profit, the split into partial profit came later when they started having a chance to make money

a very rough approximation is 2GB of vram for every billion fp16 model parameters, so the lower end models may be just about achievable on high-end cards

Worth noting the "Netflix Party" in question is not the extension now called Teleparty (previously Netflix Party), which is an order of magnitude more popular than the compromised equivalent named here

I doubt this is significant enough to matter, but shorter URLs reduce the required size of the QR code at a given error correction level, which could make them slightly easier/faster to scan with worse cameras/from further away/etc.

I'd recommend using a VPN (or VPN-like service) to connect to the server - it might not be as simple as Google Photos et al to set up, but is likely the most secure option

Edit: from a similar comment, see https://photostructure.com/faq/remote-access/#accessing-your...

I think (Re)Vanced can do it if you are on Android

The edge-specific integration seems to include a chatbot you can pull up on any page for content summaries / asking questions, I imagine this is the feature in question (although I'd imagine it would work perfectly fine as an extension, with no changes needed)

> Btw, 102 incidents within a week. Whoa! That's a lot. Wouldn't want to work there.

I might be wrong but I'm pretty sure that's every incident they've ever documented, grouped by day, as opposed to the data from last week

Their status page seems to suggest ~99.97% uptime (about 3h/year down) recently

related: https://xkcd.com/2015/

...but is it actually their voice, or just a simulation of it?

Valve's work on Proton has been repurposed for other hardware configurations with great success

as far as I'm aware, you wouldn't get more with premium - kagi doesn't return as many search results as competitors

From their FAQ:

> We believe the name of the game in search is quality, not quantity. Search engine's job is to provide the best answer fast and within the reach of one or two pages of results. If the user needs to go to page 17 to find what they were looking for, we would have failed anyway.

I've separately made use of something similar to this, using Todoist and Wear OS! Todoist exposes a complication to show your 'most urgent' task (priority + due date?), which I like to keep visible on my watch face (and syncs with the cloud, so tasks can be updated from anywhere)