Top all-time · #11679
Svilen Todorov
Personal blog · svilentodorov.xyz/blog · on HN as Tenokeaidata
Svilen Todorov's personal blog at svilentodorov.xyz covers ai and data. Since 2019, 6 different Hacker News readers have linked it 34 times, most often “Talking to Myself: How I Trained GPT-2 1.5B for Rubber Ducking with FB Chat Data”. It is still writing: the last post was on 2025-04-09, and you can follow it by feed.
Too few sampled comments (1) to say how often HN recommends it.?
Why HN links it
Linked as an earlier example of fine-tuning GPT-2 on personal chat data for rubber-ducking, from January 2020. The only comment cites it as a 'see also' next to a Twitter walkthrough.
Summarised by Claude from a sample of 1 HN comments that link here. Each comment is marked recommending, critical or neutral (citing without judging); how this works.
Personal site for Data Scientist and Machine Learning Engineer - Svilen Todorov
Most-linked posts?
- Talking to Myself: How I Trained GPT-2 1.5B for Rubber Ducking with FB Chat Data3 people · 4 mentions · first linked 2020-01-23
- My Day in 20352 people · 2 mentions · first linked 2025-04-10
What Hacker News says?
>the electricity and carbon footprint. That's no longer a concern for an increasing number of NFTs minted today[0]. 0. https://svilentodorov.xyz/blog/nft-electricity/
Comment linking /blog/nft-electricity/
on “John Cleese Sells the Brooklyn Bridge as an NFT”Submitted How is search so bad? A case study as a story · 319 points
HN discussionI guess with this they can also record user-browser interactions to use as training data, which is one way I was envisioning for creating a human-like AGI back in the day (2019)[0]. Of course, the current paradigm has went in a different direction and training directly from all the inputs/outputs of computer usage isn't quite how this data would be used, but still. 0. https://svilentodorov.xyz/blog/human-imitating-t…
Comment linking /blog/human-imitating-task/
on “Operator research preview”Submitted Talking to myself: how I trained GPT2-1.5b for rubber ducking using my chat data as a story · 217 points
HN discussionExcept Google still displays a lot of archived very old threads as if they are new[0] so often searching for recent reviews that way is annoying. 0. https://svilentodorov.xyz/blog/bad-search/
Comment linking /blog/bad-search/
on “Amazon has ruined search and Google is in on it”Submitted Adding layers to the middle of trained network without invalidating the weights as a story · 30 points
HN discussionI just pushed it, the blog post (which includes the notebook) is here[0]. It's 14mb of data, not sure how many messages it actually comes down to but FB Messenger has been my main platform for talking to friends for the last decade. 0. https://svilentodorov.xyz/blog/gpt-15b-chat-finetune/
Comment linking /blog/gpt-15b-chat-finetune/
on “Show HN: Train a language model to talk like you”Submitted Corona Advice for People Who Go Against Advice as a story · 5 points
HN discussionI've been playing with training different sizes[0] of gpt on my own chat data precisely for this reason. Coincidentally, today I was even planning to publish my last post and notebook for training gpt2-1.5b and then chatting to oneself with the model. I left it for tomorrow though.. Maybe a mistake. There is quite a lot you can do and talking to my trained model which is responding to me as me can be real weird at t…
Comment linking /blog/gpt-345M-finetune/
on “Show HN: Train a language model to talk like you”Submitted My Day in 2035 as a story · 3 points
HN discussionSEEKING WORK | Data Scientist / Machine Learning Engineer | Remote / Berlin, DE Site: https://svilentodorov.xyz/ email: sviltodorov[at]gmail.com CV: https://svilentodorov.xyz/SvilenTodorov.pdf Python Developer with 5+ years of experiencing specializing in Machine Learning. I focus on Deep Learning, especially projects near the cutting edge, but not exclusively. Computer Vision, NLP, Tabular Data, etc. Building model…
Comment linking /SvilenTodorov.pdf
on “Ask HN: Who wants to be hired? (May 2019)”Submitted How is search so bad? A 'Case Study' (2020) as a story · 3 points
HN discussionLocation: Berlin, DE Remote: Yes (3 years of Remote experience), Berlin is also okay Willing to relocate: No Technologies: TensorFlow, PyTorch, Pandas, Numpy, AWS, GCP, Django, Flask, SQL, Docker.. Résumé/CV: https://svilentodorov.xyz/SvilenTodorov.pdf Site: https://svilentodorov.xyz/ Email: sviltodorov[at]gmail.com Developer with 5+ years of experiencing specializing in Machine Learning. I focus on Deep Learning, e…
Comment linking /
on “Ask HN: Who wants to be hired? (April 2019)”Submitted Training Google’s Reformer – takeaways, code, and weights as a story · 3 points
HN discussion>Recently I've noticed that almost every search ends with ' reddit'. Quite true, but as I ranted here[0] Google screws that up, too. If you try to limit your search to only post from e.g. the last month or year, they'll happily show you archived posts from 5 years ago as well. 0. https://svilentodorov.xyz/blog/bad-search/
Comment linking /blog/bad-search/
on “Ask HN: Is Google Search Dying?”Submitted Corona Advice for People Who Go Against Advice as a story · 3 points
HN discussionThe timeout doesn't matter much, you can periodically save to drive. I spent like a month (subjectively) training gpt2 on Colab by just letting it run for a while and save the weights to drive and then pick up from there next time. I've done the same with GANs, too over a couple of days. https://svilentodorov.xyz/blog/stylegan-for-evil/ https://svilentodorov.xyz/blog/gpt-15b-chat-finetune/
Comment linking /blog/gpt-15b-chat-finetune/
on “This Bench Does Not Exist”Submitted Tips and Pain Point Workarounds for Machine Learning Work in Google Colab as a story · 3 points
HN discussionSEEKING WORK | Data Scientist / Machine Learning Engineer | Remote / Berlin, DE Site: https://svilentodorov.xyz/ email: sviltodorov[at]gmail.com CV: https://svilentodorov.xyz/SvilenTodorov.pdf Python Developer with 5+ years of experiencing specializing in Machine Learning. I focus on Deep Learning, especially projects near the cutting edge, but not exclusively. Computer Vision, NLP, Tabular Data, etc. Building model…
Comment linking /SvilenTodorov.pdf
on “Ask HN: Freelancer? Seeking Freelancer? (April 2019)”Submitted My Day in 2035 as a story · 2 points
HN discussionFwiw, I made a simple model[0] when trying tf.js a year ago which could consistently beat me and friends if you play enough against it. Wish I had kept it training with the data from people who used it but it's on a static github pages site. 0. https://svilentodorov.xyz/rps
Comment linking /rps
on ““Rock Paper Scissors” Trained in Browser AI/ML”Submitted Talking to Myself: How I Trained GPT-2 1.5B for Rubber Ducking with FB Chat Data as a story · 2 points
HN discussionIn 2019 I proposed[0] you might reach AGI if you train on computer usage - mouse movement, keypresses, what's on the screen etc. - and it sounds like Meta are kind of trying some form of it. Huh. 0. https://svilentodorov.xyz/blog/human-imitating-task/
Comment linking /blog/human-imitating-task/
on “Meta's Embrace of A.I. Is Making Its Employees Miserable”Submitted Talking to Myself or How I Trained GPT2-1.5b for Rubber Ducking Using My Facebo as a story · 2 points
HN discussionA step-by-step instruction by the author, https://twitter.com/michellehuang42/status/15977029748891443... . See also an earlier attempt by another person, "Talking to Myself or How I Trained GPT2-1.5b for Rubber Ducking using My Facebook Chat Data Using only Google Colab", https://svilentodorov.xyz/blog/gpt-15b-chat-finetune/ (Jan 2020). Personally, I think that this technology has great use for (auto)therapy.
Comment linking /blog/gpt-15b-chat-finetune/
on “I trained ChatGPT on my childhood journal entries to talk to my inner child”Submitted The Three Factors Framework for Goal Achievement as a story · 1 points
HN discussionI finetuned one in 2020[0] to play around with it and the results still seemed a bit worse than a gpt of comparable size. 0. https://svilentodorov.xyz/blog/reformer-99m/
Comment linking /blog/reformer-99m/
on “OpenAI’s policies hinder reproducible research on language models”Submitted Exploring Lichess data on improvement using e4 vs. d4 and blitz vs. rapid as a story · 1 points
HN discussionSubmitted Stylegan for Evil: Trypophobia and Clockwork Oranging as a story · 1 points
HN discussionSubmitted Building an In-Browser Rock Paper Scissors Neural Network (2018) as a story · 1 points
HN discussion
