User:16dvnk
Wikipedia editor
From Wikipedia, the free encyclopedia
| This is a Wikipedia user page. This is not an encyclopedia article or the talk page for an encyclopedia article. If you find this page on any site other than Wikipedia, you are viewing a mirror site. Be aware that the page may be outdated and that the user whom this page is about may have no personal affiliation with any site other than Wikipedia. The original page is located at https://en.wikipedia.org/wiki/User:16dvnk. |
 | This user is a WikiDragon: making massive, bold edits everywhere. |
 | This user has been to Hong Kong. |
 | This user has been to China. |
.svg){kind=small} | This user has been to the UK. |
 | This user has been to New Zealand. |
 | This user has made two wiki articles. |
 | This user is an AI researcher focused on minimalism and emergence. |
 | This user is a member of the Technology WikiProject. |
 |
This user is a Reichstag climbing patroller. |
 | This user is a member of the Guild of Copy Editors. |
 | This user has extended confirmed rights on the English Wikipedia. (verify) |
 |
Self Introduction
16dvnk is my username as an independent developer and thinker working at the intersection of artificial intelligence and political logic. My focus is on building language models with precision, emergent capabilities.
Work
I design and train my own models under the name AaI, aiming for clean, interpretable behavior with minimal hallucinations. My current release is:
- AaI mini+ preview 260311
A self-trained 14M parameter model with decent stopping and reasoning. It’s designed for controlled output, clarity, and logical discipline even at small scale. It is trained on a Python library and by a Nvidia Geforce RTX 4080 GPU.
Articles I've created
Below is the articles I have created, ranking from first made to last made:
User pages
- 16dvnk "
" - AaI (model) ""
- 16dvnk
Wiki pages
- Knowledge cutoff
- Differentially Private Stochastic Gradient Descent (Do People - Still Get Drunk?)
Philosophy
I approach AI with a layered view of cognition:
- Awareness → Consciousness → Understanding → Truth → Wisdom (It was generated by my own model)
This framework guides how I structure training, evaluation, and behavior modeling. I value strategic minimalism—fewer parameters, more insight.
Forecasting
Using ensemble techniques and original logic models, I’ve projected trends in economics and politics. For example:
- 2028 U.S. election forecast (outdated)
Harris/Shapiro: 303 EVs Vance/Haley: 235 EVs
Update pending
DeepSeek V4 Pro Preview Max's Senate prediction (not mine!)
D: 48 R: 52
Republican Victory (we'll see, I personally disagree)
- Long-term projection (with very brief / no researching)
China is likely to surpass U.S. GDP by ~2035, based on compound modeling and growth asymmetry.
The People's Republic of China is a superpower. --- Upon re-read, I realize this is a massive issue as I was in a dispute involving Superpower.
500 edits celebration pending (not actually happening anytime soon...)
Coincidentally I reached my thousandth edit while making that. Pretty ironic if you ask me. Hopefully I get to celebrate this properlly.
Practices
I value complete information, especially in code. I prefer full implementations over snippets. I train my AI Model with clean datasets like Simple English Wikipedia, with strict constraints against hallucination. However, the anti-hallucination practise is not working yet, as 14M prams is not enough. I will continue to work on the project until it reaches my standards. --- update: a 120M model is in testing, and matches or exceeds GPT-2 small in some areas. However, it is pretty clear to me that this size is too large and I have not added ways to stabilize training. May involve some vibe coding.
I also edit and structure Wikipedia-style pages for clarity, formatting accuracy, and durable knowledge representation with assistance of AI tools by checking formatting and fixing bugs. I have edited in the Scratch Wiki for over a year, so i have experience with making wiki articles.
To do:
For article Knowledge cutoff
Main concern
- *hopefully gets accepted* To do (wishful thinking)
- upgrade the sources to the time level ongoing
- PR done
- GAN after DPSGD
SuggestBot
I never read this...
Counter Vandalism Log
I ignore MOS:AT because this heading looks a lot cooler.