Hacker Newsnew | past | comments | ask | show | jobs | submit | theblazehen's commentslogin

Even just giving the image and "do image analysis to figure out what it says" to your LLM will get it done


How does this work, technically?

Does it try guess the upstream by SNI / Host header and forward the data on? That'd only work for http(s) though.


Probably. I was trying to figure out if it could be done with an address family translation, and while ip4 is trivially embedded in an ip6 address, going the other way just does not have enough information. but if you know the domain, that is a simple lookup.

The fun part is, with a dns entry like that they could technically get certs issued for peoples domains if they wanted to. I will be charitable and assume it is just a neat relay someone put together, but they could get up to some real shenanigans if they had a malicious bent. Really nothing worse than cloudflare where much their business is ssl hijacking(correction)termination.


Probably no need to guess at all. If it's given a domain, it can perform a DNS lookup for the domain - if its A entry matches, it can tunnel to the IP that the respective AAAA points to. Clever.


I use them for game AI


have any details, writups on a blog, or videos?

sounds interesting!


Still a work in progress, but planning on publishing a post when it's complete


There's significant amounts of traffic going through transfer stations [1], which is then re-sold to the model labs. For a while the DeepSeek API was serving Fable as the DS 4 Pro model even. [2]

[1] https://www.chinatalk.media/p/how-to-buy-cheap-claude-tokens... [2] https://xcancel.com/synthwavedd/status/2078514339552628880



It's not replacing the firmware.


Not necessarily. The RTS/CTS pins on the serial port are there for the device to tell the host "stop sending data, I need to process the data already sent"

The reason this doesn't work is that USB adapters don't have those pins, and thus assume you can send it at 30kbaud - not respecting when the device says it must pause sending data.

By introducing a static flow rate below what the CPU can handle, it should theoretically keep the flow rate below the threshold that the CPU starts to choke at


MIDI doesn't have RTS/CTS pins...


Well that's on me for assuming :/ I assumed that it did, based on it being explained in the blog post


dezgeg is correct, MIDI has no hardware flow control pins. The synth's serial receiver chokes because the 8-bit CPU takes too long to write the SysEx buffer to SRAM.


I've implemented the same idea with OpenCL: https://github.com/theblazehen/vramblk

There is originally https://github.com/Overv/vramfs however that has the overhead of a FUSE filesystem + loop device when using as a swap device.

The performance is rather lackluster however, it's far from a miracle "now you effectively have more ram for a 90% performance drop" - it definitely feels like traditional swapping


For me personally, it's because it retains the same window position, and I get back what I had before when it closes. If GUIs worked more like they did with windows in plan9, I'd definitely use them more


300 prompts in a day isn't that unreasonable to achieve on a heavy day? And Opus has a significant multiplier as well


That is a lot of prompts. What kind of prompts do you usually give?

Mine are usually giving specification and telling Claude to implement some part of it, referring to existing code base, writing unittests and running e2e until it passes. This can easily take 4-5 hours.

Then again, I have seen colleagues prompt “are you sure?” And other nonsense like that


I think we use it differently then, I tend to go heavy on the back and forth.

For speccing things out, I have a back-and-forth with the grill-me skill, break things down into tickets, as well as kicking off subagents. That said, I significantly overestimated the number of human messages I send.

My daily 90th percentile wrt number of prompts sent is sitting at 160 queries / day and average at 97 queries / day.

Ran an analysis of my last 2000 messages, with the following breakdown

Task delegation / execution: 23% Investigation / diagnosis / “what’s going on?”: 21% Planning / architecture / brainstorming: 15% Testing / verification / release ops: 10% Review / cleanup / quality control: 9% Course-correction / constraints / preferences: 8% Agent / ticket / workflow orchestration: 6% Providing context / evidence / pasted material: 4% Social reactions / acknowledgements / vibes: 2% Other: 2%


Yeah that’s not good for current Copilot usage since you are paying per request not by token


Yeah, copilot is only my backup - Codex and Claude are my primary, where it's per-token. Just looked up my user prompt count from my opencode db


Yeah, thats 20 Opus 4.7 prompts.


> models are judged by GPT-4

An interesting choice


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: