HNHacker News
TopNewBestAskShowJobs

xiphias2

8,476 karma · joined September 7, 2018

submissionscomments
xiphias2··on Show HN: PageQL – Embed SQL directly in HTML
Right now it's using Python DB API with SQLite, so it can be easily modified to use any database, as they are using the same standard Python API.

In the future though I'm planning to add reactive query support where it is really advantagous to have low latency between the PageQL rendering engine and the database.

For MySQL/PostgreSQL for example an extension would run much faster, and give back the results, that's why I don't want to add async / http downloading / scripting support inside PageQL, just keep it simple to implement / embed.

xiphias2··on Matrix-vector multiplication implemented in off-the-shelf DRAM for Low-Bit LLMs
This woule be a cool way to make a cheap inferencing device for the largest LLMs
xiphias2··on Run LLMs on Apple Neural Engine (ANE)
I guess it's a hard choice as it's 5x more energy efficient than GPU because it uses systolic array.

For laptops, 2x GPU cores would make more sense, for phones/tablets, energy efficiency is everything.

xiphias2··on Expanding on what we missed with sycophancy
Probably you are right. Early adopters prefer not to be bullshitted generally, just like how Google in the early days optimized relevancy in search results as opposed to popularity.

As more people adopted Google, it became more popularity oriented.

Personally I pay more not to be bs-d, but I know many people who prefer to be lied to, and I expect this part of the personalization in the future.

xiphias2··on Expanding on what we missed with sycophancy
I'm quite happy thar they mention mental illness, as Meta and TikTok wouldn't ever take responsibility of how much part they took in setting unrealistic expectations for people to life.

I'm hopeful that ChatGPT takes even more care together with other companies.

xiphias2··on Redis is open source again
,,for companies rooted in open source, it has posed a fundamental challenge: how do you keep innovating and investing in OSS projects when cloud providers reap the profits and control the infrastructure without proportional contributions back to the projects that they exploit?''

I don't see any exploitation happening. As DHH said, the main reason engineers open source their work is to give a gift to the world.

xiphias2··on All four major web browsers are about to lose 80% of their funding
Great, thanks for correcting me on the details
xiphias2··on All four major web browsers are about to lose 80% of their funding
Maybe you are right, but sone things I really dislike are:

- Getting rid of native SQLite in Chrome (Firefox was forced to follow), with the main reason is that ,,there are no 2 different implementations, all browsers use the same''.

- While there is file system implementation, there's no mmap, so there is no fast app start (just have SQLite with mmap, nothing fancy).

- No file system persistency guarantee, the OS can just wipe out the data

- All the persistent page APIs are just super hard to use compared to simple HTML/javascript/css as they are in a different process and need communication...it's overcomplicated instead of embracing a simple page as an app as an option as well.

You are right that WASM is a great improvement, and file system API is slowly coming back (still I'm not sure about mmap API which is crucial for fast app startup), but I'm talking about what was possible 10 years ago then reversed.

xiphias2··on All four major web browsers are about to lose 80% of their funding
The main inhibitor of browsers advancing was not lack of funding, but lack of will.

It would be relatively straightforward to make web browsers competitive with Java/Swift mobile apps, but 2 specific companies would lose a lot of money on it.

xiphias2··on Qwen3: Think deeper, act faster
When I tested Qwen with different sizes / quants, generally the 8-bit quant versions had the best quality for the same speed.

4-bit was ,,fine'', but a smaller 8-bit version beat it in quality for the same speed

xiphias2··on How to quickly charge your smartphone: fast charging technologies in detail
Most multichargers solve it by having faster and slower charging ports. This may be not an app, but a nice / simple hardware based solution
xiphias2··on Sapphire: Rust based package manager for macOS (Homebrew replacement)
It's not _just_ the infrastructure that is awesome for homebrew. The help I got from the team, answering in real-time when I didn't know how to get through the CI bugs is amazing. It's the boring maintainence work that makes it so special for me.

I also feel that there could be a lot of automation in the backend part, catching bugs early (maybe even on local machine before CI run) for example.

xiphias2··on Show HN: Open Codex – OpenAI Codex CLI with open-source LLMs
Maybe it was part of the reason that they accepted the PR. The fork would happen anyways if they don't allow any LLM.

A bit like how Android came after iPhone with open source implementation.

xiphias2··on Florida's New Social Media Bill Explicitly Demands an Encryption Backdoor
I guess people don't understand what it really means: government workers passing around naked selfies of their daughter that they are sending anyways to their boyfriends as part of the dating process.
xiphias2··on You might not need WebSockets
Sure, it would be just cool to not have to do that
xiphias2··on You might not need WebSockets
With HTTP streaming the browser shows that it's still loading data. Is there some mitigation for it after the initial loading?
xiphias2··on Justice Dept. scales back crypto cases in line with Trump administration memo
He didn't even tell whether ETH, the second highest value cryptocurrency is considered security or not, even though he was asked about it all the time.
xiphias2··on Less Htmx Is More
As you can see even the basic documentation is compromised at this point:

https://unplannedobsolescence.com/blog/behavior-belongs-in-h...

<button onclick="alert('I was clicked!')">Click me</button>

,,You can find HTML attribute equivalents for many of the event handler properties; however, you shouldn't use these — they are considered bad practice. It might seem easy to use an event handler attribute if you are doing something really quick, but they quickly become unmanageable and inefficient.''

,,You should never use the HTML event handler attributes — those are outdated, and using them is bad practice.''

And then proceeds to suggest this:

<button>Click me</button>

<script> const btn = document.querySelector("button")

btn.addEventListener("click", () => { alert('I was clicked!') }) </script>

After this, next step would be putting javascript to a separate file, and on a slow mobile connection in a random airport with thousands of people using the same wifi, a simple working code gets timeouts.

20 years ago all I needed to know was this, and it worked great:

<button onclick="alert('I was clicked!')">Click me</button>

xiphias2··on Recent AI model progress feels mostly like bullshit
You're right, sorry...I just tested Qwen models, not QwQ, I see QwQ only has 32B.
xiphias2··on Recent AI model progress feels mostly like bullshit
Have you compared it with 8-bit QwQ-17B?

In my evals 8 bit quantized smaller Qwen models were better, but again evaluating is hard.

xiphias2··on I was a British tourist trying to leave America. Then I was detained, shackled
All you are writing is true, but I would still put the main responsibility to workaway. Writing that ,,you need to obtain the correct visa'' is just not enough for a big company thats main profile is arranging work.
xiphias2··on I was a British tourist trying to leave America. Then I was detained, shackled
,,Some hosts may give a paid allowance to ensure they are offering at least the minimum wage in their country''

It's sad, but workaway is similarly to airbnb (hotel service) and uber (taxi service) uses ,,work'' and ,,help'' interchangebly on its website on purpose.

I am a tourist in USA (not working), and I haven't experienced any problem at the border control.

xiphias2··on LLM providers on the cusp of an 'extinction' phase as capex realities bite
There seems to be some part of the LLM capabilities being lost by hacking some benchmarks.

Claude 3.7 is a great example of a model clearly beating 3.5 in all benchmarks, but slowly destroying my code base by adding lots of extra lines or hacking around my instructions (adding ,,if'' statements when I want it to change the code to handle a case instead of understanding what change is really needed to be done for it).

I still prefer o1 pro and a lot of those leapfrogging in benchmarks don't translate to being smarter anymore.

xiphias2··on OpenAI releasing new open model in coming months, seeks community feedback
Something even cooler would be a model trained for 4 or less (1.33) bit weights instead of quantized after pretraining.

Math units are completely underutilized when I'm inferencing with batch size of 1, and post-training quantization under 8 bits loses too much of the precision to make a real difference compared to smaller models with higher precision.

xiphias2··on Apple's AI isn't a letdown. AI is the letdown
Putting generative AI inside tools without giving deep understanding of those tools to the AI generally made me more confused and frustrated than outside of it:

for example Gemini forced itself on me on my SMS app, so I thought I ask it to search something simple inside the messages, and it just started generating some random text about searching and saying that it doesn't have access to the messages themselves.

When I use ChatGPT, of course I know they don't have access to my SMSs (it would be weird).

I can give ChatGPT the exact context I want to, and I know it will work with it as long as the context is not too big.

xiphias2··on Jeremy Howard taught AI and helped invent ChatGPT. He fears he's failed
2023
xiphias2··on US administration revokes $11B in funding for addiction, mental health care
The problem here is that the same can be said to spending trillions of dollars _after_ COVID was gone.

It's quite common for goverments that cut spending aggressively getting hated (I have seen in other country where I was born and raised), but that's the only way to do it.

Politicians know that they usually don't get reelected after cutting spending, that's why they usually put in a temporary prime minister for that job.

xiphias2··on I genuinely don't understand why some people are still bullish about LLMs
,,By my personal estimate currently GPT 4o DeepResearch is the best one. ''

If the o3 based 3 month old strongest model is the best one, it's a proof that there were quite significant improvements in the last 2 years.

I can't name any other technology that improved as much in 2 years.

O1 and o1 pro helped me with filing tax returns and answered me questions that (probably quite bad) tax accountants (and less smart models) weren't able to (of course I read the referenced laws, I don't trust the output either).

xiphias2··on Gemini 2.5
Also reusing key/values for different queries can compress the KV cache, it can be an 1000x or 10000x improvement in bandwidth if the model is trained for it.
xiphias2··on Qwen2.5-VL-32B: Smarter and Lighter
I wish they would start producing graphs with quantized version performances as well. What matters is RAM/bandwidth vs performance, not number of parameters.
← PreviousPage 10 of 34Next →