HNHacker News
TopNewBestAskShowJobs

woctordho

193 karma · joined February 7, 2023

submissionscomments
woctordho··on Backing up Spotify
The worldwide median internet download bandwidth is about 100 Mbps, which is far enough for HD or bluray video. The technical barrier can be as low as 'click to search, click to download' in some user-friendly BT clients. That being said, the price of these subscriptions is a problem that actually needs to be solved.
woctordho··on Backing up Spotify
Data are basically free. Infra to store and transfer data is not.
woctordho··on How China built its ‘Manhattan Project’ to rival the West in AI chips
Their is always a political will in China to kill other Chinese since thousands of years ago. This works vastly different from the western humanitarian philosophy.
woctordho··on Russia Bans Roblox
There are plenty of ways to circumvent whitelists, such as domain fronting. Some are already battle-tested in China and we'd be happy to share them with the rest of the world.
woctordho··on Datacenters in space aren't going to work
What if we deploy reversible computing, which does not produce heat?
woctordho··on The current state of the theory that GPL propagates to AI models
How do you know I'm human?
woctordho··on Kimi K2 Thinking, a SOTA open-source trillion-parameter reasoning model
Almost all training data are on the internet. As long as the small model has enough agentic browsing ability, given it enough time it will retrieve the data from the internet.
woctordho··on Disk Prices
As long as disk prices decay faster than 1/time, we can store data forever with a finite cost.
woctordho··on Visual lexicon of consumer aesthetics from the 1970s until now
I've been using https://aesthetics.fandom.com/wiki/Aesthetics_Wiki
woctordho··on Towards a Physics Foundation Model
I guess it can be implemented in the 'sampler' part. When solving an actual PDE, project the output of the AI onto a space that preserves the invariants.
woctordho··on Don't bother parsing: Just use images for RAG
Context window extrapolation should work with hierarchical/multi-scale tokenization of images, such as Haar wavelets
woctordho··on Seedance 1.0
AI is going to watching those. AI watches much faster than humans.
woctordho··on Gemini Diffusion
Then document it. Whenever you choose one algorithm/library/tech stack but not another, write your consideration in the documents.
woctordho··on Briar: Peer to Peer Encrypted Messaging
There's deniable encryption to counter the wrench attack
woctordho··on DeepSeek open source DeepEP – library for MoE training and Inference
MoEs require less computation and more memory, so they're harder to setup in small labs
woctordho··on The Ren'Py Visual Novel Engine
We use Unity so it's cross-platform, including mobile. But we never implemented a server for cloud saves, so you need to sync the saves using other tools
woctordho··on The Ren'Py Visual Novel Engine
Reddit has it https://www.reddit.com/r/visualnovels/wiki/recommendations
woctordho··on The Ren'Py Visual Novel Engine
VNs are interesting to AI developers because they're multimodal datasets with extensive text annotations. People are already working on this https://huggingface.co/datasets/Limour/b-corpus
woctordho··on The Ren'Py Visual Novel Engine
If you prefer JS to Python, there's also WebGAL https://github.com/OpenWebGAL/WebGAL
woctordho··on The Ren'Py Visual Novel Engine
The script graph can be thought as a DFA, but a 'game state' (images showing on screen, music playing...) is different from a node of the script graph, and there is not a one-to-one mapping between them

For example, if the script loops through a node (and there is a choice to go out of the loop), and the node moves a character on screen to the left by 5 pixels each time, then there can be different game states with the same node. It happens in a common trope of VNs when the story lets you explore different places and go back to the original place each time

In the design of our VN framework, the game state is determined not by the current node, but by the node history (and other things like 'global variables'). If the author changes some script in the new version, then all saved data with the affected node history need to get updated

If there is no choice added/deleted, then the update can be implemented by simply re-running the nodes. However, if there are choices added/deleted, the update can be more complicated and involve graph isomorphism in the worst case

woctordho··on The Ren'Py Visual Novel Engine
It has a DSL specified for writing VNs
woctordho··on The Ren'Py Visual Novel Engine
Many examples can be queried from VNDB https://query.vndb.org/8e75036b2dcfe441
woctordho··on The Ren'Py Visual Novel Engine
Forward compatibility of saves is harder than people thought. VN scripts have choices and loops, so in general they are graphs, and upgrading the saves to another version requires matching two graphs. I'd be happy to know if there is a good diff algorithm for graphs

In practice graph matching can be helped by manually tagging the same nodes (labels in Ren'Py) in the two versions, but that cannot cover all the edge cases

I'm developing a VN framework for the own use of my indie VN dev group, and we mostly implemented the diff algorithm for the 'linear' part of the graphs. You can search my handle to know more

woctordho··on Why bother with privacy when I have nothing to hide? (2023)
The problem is that banks do not distinguish read and write permissions. I'm absolutely happy with giving out the read permission
woctordho··on DeepScaleR: Surpassing O1-Preview with a 1.5B Model by Scaling RL
Crawling the whole internet still helps solve problems in the long tail of the distribution. The purpose of big data is garbage in, signal out
woctordho··on LIMO: Less Is More for Reasoning
A finite-size LLM can solve the finite-size halting problem, and an infinite-size LLM can solve the infinite-size halting problem
woctordho··on Qubes OS: A reasonably secure operating system
In China we have ways to obfuscate those unusual traffic into usual ones like WeChat video calling
woctordho··on Sora is here
3e4a3ad9f05fdfb609dda6e5f512e52506f4c1053962e21bfd93f1ed81582d16ca0fef9574fb07ab62f8f5b1373b4ddd541804c0d176f4a557d900b05047e853

(This is the hash of a string randomly popped in my mind. An LLM will write this with almost 0 probability --- until this is crawled into the training sets)

woctordho··on Trellis – 3D mesh generative model
At this point maybe meshes are not the best representation for animation and editing. We can just use latents of neural networks
woctordho··on Genie 2: A large-scale foundation world model
We can already program Minecraft. Also we can already program GTA6. But imagine interpolating Minecraft and GTA6 such that all buildings are destroyable. It may be easier to achieve using AI rather than traditional programming
← PreviousPage 4 of 5Next →