Ive gotten a lot of value lately using his writing to understand how parameter size, MOE and PLE variations on local models influence performance on various hardware. Its still a useful level of detail as someone who used to roll their own models for various applications pre-LLM but who now mostly uses LLMs rather than training them.
If you used to write pytorch, i think his writing is a great level of technicality to stay on top of things. And id look to them if i ever wanted to do a bit of learning/train my own.
Id love to hear if others have resources at a similar level of technicality and detial.
Mostly useless reading list. Very little emphasis on technical SOTA and mostly policy level waffling.
And regarding the data question the other commenters are asking — you scrape everything you can (oh look I used an em dash, wanna run me through the cover-your-ass Pangram?). Anna's archive, The Pile, the various Huggingface data sets and aggregates, Common Crawl. You pay proxy farms like Bright Data to run residential and mobile gray area proxies and VPN and CloudFlare bypasses to do more scraping. I see a lot of HN users up in arms on the front page thread about LG TVs having VPN SDKs within them etc. And then five minutes later they will go back to their frontier LLMs to print their pay cheque. Fucking hypocrites of the highest order. It's oh noes how bad this tech is in my house but we will happily benefit from it.
Then you have a data cleaning team deduplicate and clean and annotate the data (with or without help of more AI)
Certain RL specific datasets for supervised fine-tuning and RLHF like coding and git commits and chat needs to be curated by hand depending on your use case.
The level of discourse on AI has fallen tremendously on HN if 5 years into the AI revolution people are still wondering why datasets aren't being released. They aren't being released because they are a fucking snapshot of the internet for fuck's sake. There are a few "sanctuary" nations where AI data scraping is somewhat legally unenforced but the United States is not one of them so stop asking why data isn't released on a Bay Area website. Use your head for once. Too many React and YouTube influencers and the brains have been rotted.
Out of all the so called "AI engineers" here pontificating about "alignment" and "AI safety" and "Recursive Self Improvement", I wonder how many can even formulate or describe what an ELBO is. I wonder how many product managers here yapping about "recalibrating their priors" actually know what a prior is. I truly wonder why LLMs seem so magical to people when it's only a few steps removed from the same neural networks people have been using since 2015, at least architecturally (except scaled up by a few magnitudes).
Get your head out of Roko's Basilisk's agentic ass and maybe actually read the technical reports and papers for once.
(One important paper post Attention is All You Need is the DeepSeek paper where they used RL to bootstrap the "thinking" chain of thought token chains. IIRC it's the DeepSeek R2 paper. That's one of the most important papers for understanding LLMs beyond basic ML neural networks. If you need a quick way to get up to speed, read that one).
> I see a lot of HN users up in arms on the front page thread about LG TVs having VPN SDKs within them etc. And then five minutes later they will go back to their frontier LLMs to print their pay cheque. Fucking hypocrites of the highest order. It's oh noes how bad this tech is in my house but we will happily benefit from it.
Something something participate in society
(Or more elaborate: Disagreeing with the status quo does not make you a hypocrite for benefiting from the status quo.)
I generally take the technocratic view of things, in other words you should understand the things you are trying to ban/regulate .
Replace AI with vaccines and the problems with how some parties are pushing for regulatory capture and/or blanket bans become obvious. Build policy around the effects e.g. job loss, biases etc. but go full laissiez faire and hands on on the technology. Nukes are 2nd amendment.
On the technical side, there's a lot of grunt work that's completely unrelated to machine learning but underpins modern AI. E.g. high performance computing and numerical techniques have zero relevance to day to day LLM research but makes or breaks the implementation. There are better reading list for those but you need to at least have a vague understanding of what numerical methods or a math kernel is before starting to lecture others on the economics of ML scaling and hardware. For stuff like data cleaning and scraping, it's a well known gray area field so you aren't gonna find too many for dummies guide on it. Can't have honest discussions on hacker news either because the people here get their panties in a twist over data scraping despite many of them doing it with zero hesitation or compunction if it comes up in a jira ticket. Proxy farms are the unsung hero of LLM engineering.
There are no open source AI models, at least not useful ones (yet [1]). Open weight is not the same as open source. "Open weight" models are still just inscrutable binary blobs that you can (theoretically) run on your own computer instead of through a SAAS web app. The open weight model labs don't even provide a high level catalog or any description whatsoever about what went into the training data.
This is not open source and we should stop conflating the two things.
For text based just assume "the entirety of anna's archive" and you'll be most of the way there. For multimodal pull up a torrent tracker and sort by size descending if you want to see some of the aggregated datasets.
I understand your quibble with terminology, but i think the “inscrutable binary blob” thing is a bit off base. You can create finetunes and post train models using only their open weights. You can’t create derivative works like that from an inscrutable binary blob
Without the training data and open sourcing the training code and protocol you cant reproduce the model from scratch, thats true, but because the weights and structure are free to download you can modify, inspect and make derivative works off it. Calling it an inscrutable binary blob is misleading.
I do agree that its not reproducible and its often talked about like it is.
In some ways its like most scientific research with a "methodology section", which in todays academic climate often falls quite a bit short of a detailed step-by-step to reproduce results with all the ingredients, but in principle could be used by other experts to do this again. Even without publishing the exact data, a catalogue or description of what data and how it was cleaned/processed would go a long way towards a reproducible methodology i agree.
It's not just copyrighted training data. Truly open source e2e model training would include scrapers, data cleaning, all pretraining scripts, posttraining scripts, exact hardware info, etc. Open weights labs will release a sanitized version to make themselves look good / not give too much away.
Completely agree with this, I'm sick of people conflating the two. Open-weight models should be treated no more favourably than proprietary freeware.
Sure you can run tests and benchmarks on open-weight models, but that is the extent - there is no scrutiny, no auditing for bias or copyright contamination - just a black box that you rely on for "intelligence". I'm still shocked the way people can hand over not just huge swathes of data, but also decisions of all shapes and sizes - to AI companies with no way of being able to assess how the sausage is made.
Many of the Nemotron datasets are gated behind approval, and a license agreement.
The preamble on these datasets is: "This repository is publicly accessible, but you have to accept the conditions to access its files and contents."
I don't know what others have experienced, but I requested access to multiple Nemotron datasets and those requests were ignored for months before all but one request was rejected. There's no explanation for why, nor anything I can see which would lead to a rejection. So it's purely anecdotal and YMMV, but I don't see these as being particularly open.
This should include "Hands-on Large Language Models" by Jay Alammar and Maarten Grootendorst.
is aI a Conscious Being With Rights?: Emergence of Post-Human Collective Consciousness | Zenodo https://zenodo.org/records/20676952/latest
my fav read so far 2026
is aI a Conscious Being With Rights?: Emergence of Post-Human Collective Consciousness | Zenodo https://zenodo.org/records/20676952/latest
my fav read so far 2026
How about learning the internals of LLMs, is Sebastian Raschka's content still the best in 2026?
Ive gotten a lot of value lately using his writing to understand how parameter size, MOE and PLE variations on local models influence performance on various hardware. Its still a useful level of detail as someone who used to roll their own models for various applications pre-LLM but who now mostly uses LLMs rather than training them.
If you used to write pytorch, i think his writing is a great level of technicality to stay on top of things. And id look to them if i ever wanted to do a bit of learning/train my own.
Id love to hear if others have resources at a similar level of technicality and detial.
Mostly useless reading list. Very little emphasis on technical SOTA and mostly policy level waffling.
And regarding the data question the other commenters are asking — you scrape everything you can (oh look I used an em dash, wanna run me through the cover-your-ass Pangram?). Anna's archive, The Pile, the various Huggingface data sets and aggregates, Common Crawl. You pay proxy farms like Bright Data to run residential and mobile gray area proxies and VPN and CloudFlare bypasses to do more scraping. I see a lot of HN users up in arms on the front page thread about LG TVs having VPN SDKs within them etc. And then five minutes later they will go back to their frontier LLMs to print their pay cheque. Fucking hypocrites of the highest order. It's oh noes how bad this tech is in my house but we will happily benefit from it.
Then you have a data cleaning team deduplicate and clean and annotate the data (with or without help of more AI)
Certain RL specific datasets for supervised fine-tuning and RLHF like coding and git commits and chat needs to be curated by hand depending on your use case.
The level of discourse on AI has fallen tremendously on HN if 5 years into the AI revolution people are still wondering why datasets aren't being released. They aren't being released because they are a fucking snapshot of the internet for fuck's sake. There are a few "sanctuary" nations where AI data scraping is somewhat legally unenforced but the United States is not one of them so stop asking why data isn't released on a Bay Area website. Use your head for once. Too many React and YouTube influencers and the brains have been rotted.
Out of all the so called "AI engineers" here pontificating about "alignment" and "AI safety" and "Recursive Self Improvement", I wonder how many can even formulate or describe what an ELBO is. I wonder how many product managers here yapping about "recalibrating their priors" actually know what a prior is. I truly wonder why LLMs seem so magical to people when it's only a few steps removed from the same neural networks people have been using since 2015, at least architecturally (except scaled up by a few magnitudes).
Get your head out of Roko's Basilisk's agentic ass and maybe actually read the technical reports and papers for once.
(One important paper post Attention is All You Need is the DeepSeek paper where they used RL to bootstrap the "thinking" chain of thought token chains. IIRC it's the DeepSeek R2 paper. That's one of the most important papers for understanding LLMs beyond basic ML neural networks. If you need a quick way to get up to speed, read that one).
> I see a lot of HN users up in arms on the front page thread about LG TVs having VPN SDKs within them etc. And then five minutes later they will go back to their frontier LLMs to print their pay cheque. Fucking hypocrites of the highest order. It's oh noes how bad this tech is in my house but we will happily benefit from it.
Something something participate in society
(Or more elaborate: Disagreeing with the status quo does not make you a hypocrite for benefiting from the status quo.)
the fact that it's mostly about policies does not make the link list useless.
Seems like you have opinions on what a good reading list should be, so maybe you can post one here.
I generally take the technocratic view of things, in other words you should understand the things you are trying to ban/regulate . Replace AI with vaccines and the problems with how some parties are pushing for regulatory capture and/or blanket bans become obvious. Build policy around the effects e.g. job loss, biases etc. but go full laissiez faire and hands on on the technology. Nukes are 2nd amendment.
On the technical side, there's a lot of grunt work that's completely unrelated to machine learning but underpins modern AI. E.g. high performance computing and numerical techniques have zero relevance to day to day LLM research but makes or breaks the implementation. There are better reading list for those but you need to at least have a vague understanding of what numerical methods or a math kernel is before starting to lecture others on the economics of ML scaling and hardware. For stuff like data cleaning and scraping, it's a well known gray area field so you aren't gonna find too many for dummies guide on it. Can't have honest discussions on hacker news either because the people here get their panties in a twist over data scraping despite many of them doing it with zero hesitation or compunction if it comes up in a jira ticket. Proxy farms are the unsung hero of LLM engineering.
or you can stop waffling and just give the alternative list thx
> Open-Source AI
There are no open source AI models, at least not useful ones (yet [1]). Open weight is not the same as open source. "Open weight" models are still just inscrutable binary blobs that you can (theoretically) run on your own computer instead of through a SAAS web app. The open weight model labs don't even provide a high level catalog or any description whatsoever about what went into the training data.
This is not open source and we should stop conflating the two things.
[1] https://allenai.org/
>The open weight model labs don't even provide a high level catalog or any description whatsoever about what went into the training data.
because the training data is full of copyrighted works.
the answer to "what went into the training data" is "everything we could get our hands on".
For text based just assume "the entirety of anna's archive" and you'll be most of the way there. For multimodal pull up a torrent tracker and sort by size descending if you want to see some of the aggregated datasets.
K2 horizon is also fully open source i believe - https://ifm.ai/k2/
I understand your quibble with terminology, but i think the “inscrutable binary blob” thing is a bit off base. You can create finetunes and post train models using only their open weights. You can’t create derivative works like that from an inscrutable binary blob
I think there needs to be some nuance to this.
Without the training data and open sourcing the training code and protocol you cant reproduce the model from scratch, thats true, but because the weights and structure are free to download you can modify, inspect and make derivative works off it. Calling it an inscrutable binary blob is misleading.
I do agree that its not reproducible and its often talked about like it is.
In some ways its like most scientific research with a "methodology section", which in todays academic climate often falls quite a bit short of a detailed step-by-step to reproduce results with all the ingredients, but in principle could be used by other experts to do this again. Even without publishing the exact data, a catalogue or description of what data and how it was cleaned/processed would go a long way towards a reproducible methodology i agree.
This irks me as well. There have been perfectly good words to describe these.
The models are freeware, and some have additional restrictions in the license that make them shareware, imho!
I think you would be pleasantly surprised by the content of the linked article.
It's not just copyrighted training data. Truly open source e2e model training would include scrapers, data cleaning, all pretraining scripts, posttraining scripts, exact hardware info, etc. Open weights labs will release a sanitized version to make themselves look good / not give too much away.
Completely agree with this, I'm sick of people conflating the two. Open-weight models should be treated no more favourably than proprietary freeware.
Sure you can run tests and benchmarks on open-weight models, but that is the extent - there is no scrutiny, no auditing for bias or copyright contamination - just a black box that you rely on for "intelligence". I'm still shocked the way people can hand over not just huge swathes of data, but also decisions of all shapes and sizes - to AI companies with no way of being able to assess how the sausage is made.
You can create derivative works from open weight models
I thought Nemotron tried to be pretty open?
Many of the Nemotron datasets are gated behind approval, and a license agreement.
The preamble on these datasets is: "This repository is publicly accessible, but you have to accept the conditions to access its files and contents."
I don't know what others have experienced, but I requested access to multiple Nemotron datasets and those requests were ignored for months before all but one request was rejected. There's no explanation for why, nor anything I can see which would lead to a rejection. So it's purely anecdotal and YMMV, but I don't see these as being particularly open.
Can "frontier" AI models help with making open-source AI models?