jy14898

I'm on mobile so I'm not able to check, but can the will-change and contain attributes reduce the cost?

> Although logic connectivity is not layout-dependent in the current prototype, we envision a future where physical proximity directly influences or determines logical connections. In this ultimate view, the circuit behaves like a digital organism whose physical form and logical connections evolve together under local structural constraints.

Not all bugs are buffer overflows, many are just the code not doing what it claimed at a high level

405 Method Not Allowed is trivial to fall back to POST. How do you know the GET request behaved incorrectly?

Seems like the README is heavily vibed, as it seems to not even understand what the repo does:

> Local AI & LLM Ready: By reducing complex pixel streams into structured logical strings, ASCILINE acts as a perfect bridge for AI. Instead of feeding heavy computer vision models, lightweight LLMs can process semantic video summaries.

In what way is this semantic/structured?

> Bypassing Browser Constraints: Modern browsers aggressively throttle autoplay videos, and ad-blockers restrict traditional media frames. To the browser, ASCILINE is simply "JavaScript updating a canvas"—completely invisible to media restrictions.

So... just render the video to a canvas? What does ASCII have to do with it

> LLMs are good at predicting words, since each word in the id is ~1 BPE token. But uuids are random hex characters, this is where LLMs struggle to output the right ids.

If true then that indeed seems like an improvement, I think I just need measurements of actual hallucinations. Calling hex random but a selection of words not seems humanly biased? If anything, being random is good because it's saying there's no semantic influence. I'd think that words are more likely to be hallucinated as certain words only follow certain contexts, which is less true for numbers

I don't like that they're not apples to apples; less bits so of course it'll take less tokens.

> Where UUIDs cost ~23 tokens and get hallucinated by LLMs

How does this solve the hallucination problem?

Just removing the - from the example UUID takes it from 26 tokens to 18

LLMs aren't relevant to aviation and medical devices

The free battery deal ended in January, but you're likely better off as mine ended up getting a damaged screen while being transported for mail in (because all local stores stopped doing the program), and they wanted to charge me an extortionate price to fix the screen. Support were useless

Came to the same conclusion the moment I had to hunt to see the outputs https://github.com/creationix/rx/tree/main/samples

https://github.com/aleks-apostle/claude-code-patches/pull/9 is what I found, not tested yet

It's monetisation. If they put a paywall on the video, your browser has the functionality to play the video but you're forced to pay to use that functionality.

Also wrt phone, it's different because I paid for the phone. But also I'd just use a different camera app?

While I'm not pro YouTube, I think it's fine for companies to decide how to monetise their product, including things which were originally free. If you don't like free services, stop using them

Humans eat insects, current and past

I feel like any deviation from the syntax LLMs are trained on is not productive.

Sure you can represent the same code in fewer tokens but I doubt it'll get those tokens correct as often.

I never want to unknowingly use an app that's driven this way.

However, I'm happy it's happening because you don't need an LLM to use the protocol.

you might like https://omrelli.ug/g9/ which is a similar concept but for graphics

The post stated that it was believed duplication improved loading times on computers with HDDs rather than SSDs

Stopped at 74 but managed to par all before that somehow. Didn't really do any problem solving/deep thinking about it, just clicking what felt right

I don't think you can apply Unix philosophy to a (GUI) web browser, you don't use it compositionally.

Open the page in two windows, with one that has note mode enabled

ramburgers are quite healthy, they've been shown to improve memory

I'm interpretting your message as you asking me to share my API keys

Fair enough, but this isn't that case. They have Y (it's in the first photo) and tested a previous version of the model. The changes are predictable (rotating letters and slight scaling) so I don't think it's unreasonable to be confident and not waste plastic.

What do you mean by "CSS is basically pre-tokenized input"? Can you give an example of what you have trouble with?

It tickled me seeing the streaming example, thinking about how much better HTML deals with streaming

The font is around 6 years old and free, so I wouldn't be suprised

> The other option is to misuse refs to also maintain a current state there in addition to the useState. So always use the ref, but set the state in the end to trigger a re render.

Arugably this isn't a misuse, but just accepting that Reacts state management isn't powerful enough for what you need. Something akin to useRef + useSyncExternalStore is often what developers want.

For example, there's no guarantee that your useEffect will be invoked with the intermediate state you wanted to observe.