It's funny because I was reading about Cherenkov radiation just yesterday and the same statement was made (although with a proper clarification afterwards).

I'm glad I've already done that! I haven't seen any difference in performance like others mention though. Weird. The only override I've made is for a website which needs to run some client-side cryptography code and it would take ages without the JIT.

The article doesn't give much more information than the title. Almost zero details.

> The requirement `1.0.107` is a caret range, and with only `1.0.106` and `1.0.107` ever published it resolves to the malicious `1.0.107`.

Uhh no it isn't? This seems to be a pretty low quality article

I'd say Canon and Scratch are different tools with different purposes. Both are simple and interactive enough for children to love. The main difference I see is that Canon abstracts away a lot of the game state management logic and you get to focus on creating characters, items and having lots of fun. Scratch looks like it has a steeper learning curve and while it is a good introduction to programming, it gets pretty hard to navigate once project size increases. As someone who started programming with Scratch, I think that making a text adventure game like the ones Canon does in Scratch would be much more complicated and the end result wouldn't be half as nice.

One man's trash is another man's treasure. The mistake is being absolute and ignoring this fact.

This post reminded me of Engineering for Slow Internet (https://brr.fyi/posts/engineering-for-slow-internet), another case where careless assumptions by better-off people can have a huge impact on others.

Looks incredibly cool, although it ran at ~4 tok/s on my browser (I assume because I've disabled V8). I believe your numbers though. I always wanted to try using something like this to create an AI assistant for my phone. If you try to create something without AI, you'll quickly run into the problem of using NLP to map input to the equivalent of AI tool calls. On the other hand, shelling out to a full-blown LLM for every input is slow, costly and if you want it local, greatly restricts its computing power and the number of devices it can run on. Needle and similar models solve this problem nicely: they're incredibly cheap and handle NLP (not in a predictable way of course, but with great success rates), so you can quickly run predefined actions or shell out to a proper LLM for more complicated tasks. You have the best of two worlds.

You may have inspired me to give it a try!

310 tok/sec on my browser, hm.

Now you can try running the python package directly on your Mac to see proper speedup. WebAssembly has some limitations, running on pure hardware, especially ARM-based to see thousands (which 99% of consumer devices are ARM).

Please give it a try and let us know!

Not sure Jolla deserves to be trusted, you'd be much better off with GrapheneOS. In any case, try out SimpleX for a messenger. You can also take a look at https://xn--gckvb8fzb.com/an-overview-of-privacy-focused-dec...

Thanks for SimpleX it looks like a great solution for a longstanding problem.

Why would you not trust Jolla? It was born from Nokia employees. GrapheneOS is a great alternative. Still Android though.

I agree, it's by far the best mobile keyboard I've ever tried. I especially love the shortcuts for copy/paste/select all/cut which send the actual Ctrl-C. I can finally use terminals! :-P

I can't see what's wrong with that (you choose whether to accept suggestions or not), but you can always blacklist predictions by long-pressing on them.

I read it as a positive comment, as in that it types what you wrote