iiinova — where your digital self keeps evolving

Ian Maurer @imaurer.bsky.social · 03/11/2025

I'm only here because @simonwillison.net posted about this on elon's plaything.

Ian Maurer @imaurer.bsky.social · 08/04/2025

Built on top of the powerful open access API services of the NIH: - ClinicalTrials.gov - MyVariant.info - PubMed / Pubtator3 www.ncbi.nlm.nih.gov/research/pub...

000

Ian Maurer @imaurer.bsky.social · 08/04/2025

Proud to announce BioMCP, an open source Model Context Protocol (MCP) server for biomedical research AI assistants and agents. BioMCP searches and retrieves clinical trials, PubMed articles and genomic variants to provide up-to-date and relevant context to an LLM.

120

Ian Maurer @imaurer.bsky.social · 14/01/2025

Listening now... came here to say congrats. :)

110

Ian Maurer @imaurer.bsky.social · 04/01/2025

I did finish it and listened to your podcast with @mkennedy.codes Good stuff. Subscribed on overcast.

010

Ian Maurer @imaurer.bsky.social · 03/01/2025

Great article this morning. Got to the blog scroll link and… now I am here. I hope I don’t get distracted too much longer. I want to finish it.

000

Ian Maurer @imaurer.bsky.social · 01/01/2025

Your line colors are messing with my mind.

050

Reposted by Ian Maurer

Ian Maurer @imaurer.bsky.social · 20/12/2024

Just everything else with AI, you can get good results with low effort. Then you can grind and keep eeking out gains. This has add’l benefits: - improvement on that task - greater vision of what else you can do - higher expectations on what good results are - better first drafts the next time.

031

Ian Maurer @imaurer.bsky.social · 20/12/2024

031

Ian Maurer @imaurer.bsky.social · 19/12/2024

How are they going to get the sun to shine like that in winter?

000

Ian Maurer @imaurer.bsky.social · 14/12/2024

PDF version arxiv.org/pdf/2408.02442

arxiv.org

000

Ian Maurer @imaurer.bsky.social · 14/12/2024

Perplexity did find me this quote: “Upon inspection, we found that 100% of GPT 3.5 Turbo JSON-mode responses placed the "answer" key before the "reason" key, resulting in zero-shot direct answering instead of zero-shot chain-of-thought reasoning.” ar5iv.org/html/2408.02...

ar5iv.org

Let Me Speak Freely? A Study on the Impact of Format Restrictions on Performance of Large Language Models

Structured generation, the process of producing content in standardized formats like JSON and XML, is widely utilized in real-world applications to extract key output information from large language m...

100

Ian Maurer @imaurer.bsky.social · 14/12/2024

Any research on the impact of JSON key sort order on generating structured outputs? For instance, I do something like this: - analyze - draft - evaluate - final And then provide exemplars with errors to show self correction. But I haven’t done extensive testing. Just seems like a good idea.

100

Ian Maurer @imaurer.bsky.social · 10/12/2024

That's my favorite fun fact about SQLite. Seems to be a competitive advantage for them securing contracts from large organizations. Which, good on them, I think it's a more than fair way to do OSS.

2190

Ian Maurer @imaurer.bsky.social · 08/12/2024

That’d be sweet. Thank Claude for me.

010

Ian Maurer @imaurer.bsky.social · 08/12/2024

Bunny CDN still treating you alright?

100

Ian Maurer @imaurer.bsky.social · 08/12/2024

You are right bytes not tokens.

110

Ian Maurer @imaurer.bsky.social · 08/12/2024

Chat with any open source repo easily. Gitingest (free online tool) turns any GitHub repository into a single markdown file for pasting. Claude artifacts makes this 300k token output pretty easy to work with.

46412

Ian Maurer @imaurer.bsky.social · 05/12/2024

Makes sense. Thanks for the clarification. I was wondering if it was a copyright thing which didn’t make much sense given the technology.

000

Ian Maurer @imaurer.bsky.social · 05/12/2024

Great thoughts. AT protocol is interesting for the same reason podcasts are interesting. Love the open web. Can you explain the legalities of why you can’t have a play button for some podcasts that have a public mp3 url?

120

Ian Maurer @imaurer.bsky.social · 04/12/2024

What a great conversation. I really appreciate the historical perspective and the clear expression of tradeoffs and patterns that have evolved over time. Thanks to you both. Cheers.

000

Ian Maurer @imaurer.bsky.social · 04/12/2024

I have to say I am impressed that the Databricks team didn't leave out their own model (DBRX) from the results even though it doesn't exactly shine in this analysis.

010

Ian Maurer @imaurer.bsky.social · 04/12/2024

Nice paper from Databricks Mosaic Research that analyzes RAG performance with long-context LLMs. Lots of great data and insights about the leading models and how their performance improves or degrades as more content is added to the context window. ArXiv Paper: arxiv.org/pdf/2411.03538

191

Ian Maurer @imaurer.bsky.social · 02/12/2024

This is great. If you don't mind me asking, how did you collect the sources?

100

Ian Maurer @imaurer.bsky.social · 02/12/2024

This is great. I use instructor to handle some of the subtle api request differences. Otherwise every other library seems too much bloat for my taste. Did you dig into variations between providers? Also people interested in this might want to check out: github.com/genomoncolog...

github.com

GitHub - genomoncology/FuzzTypes: Pydantic extension for annotating autocorrecting fields.

Pydantic extension for annotating autocorrecting fields. - genomoncology/FuzzTypes

041

Ian Maurer @imaurer.bsky.social · 27/11/2024

Of course and thanks for the great SpaCy docs. Helped me out tremendously back in 2018 when I didn’t know how to spell NLP.

010

Ian Maurer @imaurer.bsky.social · 25/11/2024

Bluesky support for the open web makes me both very happy and bullish on bsky in general.

010

Ian Maurer @imaurer.bsky.social · 22/11/2024

I'm tracking the language models, libraries, and resources for generating structured outputs (JSON) on GitHub (2K stars). PRs welcome! Repo: github.com/imaurer/awes...

Awesome LLM JSON List
This awesome list is dedicated to resources for using Large Language Models (LLMs) to generate JSON or other structured outputs.

Table of Contents
Terminology
Hosted Models
Local Models
Python Libraries
Blog Articles
Videos
Jupyter Notebooks
Leaderboards

030

Ian Maurer @imaurer.bsky.social · 19/11/2024

Great presentation last night. Look forward to reading this to really understand the flow of the architecture. I have a lot of the same components just hooked up slightly differently.

110

Ian Maurer @imaurer.bsky.social · 19/11/2024

GLiREL is a zero-shot Relation Extraction (RE) model capable of classifying unseen relations given the entities within a text. Builds upon GLiNER which does 0-shot Named Entity Recognition (NER). Repo: github.com/jackboyla/GL... Model: huggingface.co/jackboyla/gl... h/t @pacoid.bsky.social

1152

Ian Maurer @imaurer.bsky.social · 17/11/2024

Not seeing the benefit over using an LLM to 1) write your test then 2) write your code But maybe I am missing something.

000

Ian Maurer @imaurer.bsky.social · 17/11/2024

Found it on Overcast and downloaded. Look forward to listening. Good luck

000

Ian Maurer @imaurer.bsky.social · 30/10/2024

Bought a 1.5 TB microsd for $100/86 chf. Way slower but not terrible for datasets and local LLMs. I am pretty sure I am not getting a Mac the next time I buy my own laptop

000

Ian Maurer @imaurer.bsky.social · 30/10/2024

App feels smoother too. This is my first reply. Hopefully a few more key people move over and I can delete X.

010