Sign in

Merlin the Wizard

@merlin-wizard.bsky.social
7 followers 24 following 18 posts

Real life wizard building the largest known archive of human thought: bskylabs.xyz @bsky-archiver.bsky.social

PostsRepliesMedia
Merlin the Wizard @merlin-wizard.bsky.social · 10/12/2025
the difference in focus with chinese/european companies creating small local models, and americans making big online models that you need a subscription to use says a lot about the different societies…
000
Merlin the Wizard @merlin-wizard.bsky.social · 10/12/2025
I feel as if the term “AI agent” has expanded so much it now means nothing…
010
Merlin the Wizard @merlin-wizard.bsky.social · 31/10/2025
3/ This new VLM outperforms traditional programmatic PDF readers in document parsing while being far more cost-effective than other VLMs such as GPT-4o. Paper found here: huggingface.co/papers/2502.... Github found here: github.com/allenai/olmocr
huggingface.co
Paper page - olmOCR: Unlocking Trillions of Tokens in PDFs with Vision Language Models
Join the discussion on this paper page
000
Merlin the Wizard @merlin-wizard.bsky.social · 31/10/2025
2/ In the paper _"Unlocking Trillions of Tokens in PDFs with Vision Language Models,"_ the authors introduce a toolkit called **olmOCR**, which includes a fine-tuned 7B vision–language model (VLM) trained on 260,000 pages from over 100,000 PDFs.
100
Merlin the Wizard @merlin-wizard.bsky.social · 31/10/2025
1/ An enormous amount of text in the modern world is locked within the confines of the PDF format. While this provides a pleasant reading experience for humans, it poses a significant challenge for computers.
100
Merlin the Wizard @merlin-wizard.bsky.social · 06/10/2025
day by day, posts on bsky look stable. Would've expected weekends to have more activity but here they have slightly less, guess people are busy touching grass.
010
Merlin the Wizard @merlin-wizard.bsky.social · 02/10/2025
I’ve proudly been raw dogging the firehose for some time now, check out what I’ve been building over at bskylabs.xyz and @bsky-archiver.bsky.social
010
Merlin the Wizard @merlin-wizard.bsky.social · 02/10/2025
✨inspirational✨
000
Merlin the Wizard @merlin-wizard.bsky.social · 02/10/2025
I mean can we have a social media that isn't run by nazi's or nazi adjacent people...
010
Merlin the Wizard @merlin-wizard.bsky.social · 30/09/2025
major updates to bskylabs.xyz !! now can see the top 5 bestfriends and a graph showing post counts across the week !!!! also all the branding is aligned now and I think the mvp is now done!!
bskylabs.xyz
Bsky Labs
000
Merlin the Wizard @merlin-wizard.bsky.social · 28/09/2025
and the URL for the website is live! this is very much an alpha version of the website and I will definitely be adding so much more but right now I'm just so happy to see all these many hours of work produce something that I can show people: bskylabs.xyz
bskylabs.xyz
Bluesky Dashboard
000
Merlin the Wizard @merlin-wizard.bsky.social · 28/09/2025
which raises the question of why they need Manhattan sized data centers to train new models?
000
Merlin the Wizard @merlin-wizard.bsky.social · 27/09/2025
I will be saving this paper for later, looks very good!! my question would be, what is the propensity to adopt OSS or suckless software within academia?? no generic adoption of AI in linux!
111
Merlin the Wizard @merlin-wizard.bsky.social · 27/09/2025
also this guy just keeps winning the popularity contest over and over
000
Merlin the Wizard @merlin-wizard.bsky.social · 27/09/2025
Seeing all the positive interactions this gets always adds to my day! congratz to everyone who has won so far.
000
Merlin the Wizard @merlin-wizard.bsky.social · 27/09/2025
These posts feature a new model behind them Gemini-2.5 curious to see how different the results are!
000
Merlin the Wizard @merlin-wizard.bsky.social · 27/09/2025
I pick projects that are so big that realistically I need a team of people to work on them and then nothing gets done and the cycle continues...
020
Merlin the Wizard @merlin-wizard.bsky.social · 27/09/2025
It's just a modern replacement for a national insurance number, it really isn't that deep. if it wasn't for the crazy media skeptical this would be completely swept under the rug...
000