Especially true in AI. AI always starts worse than a child and improves. You saw this in Tesla's FSD. When I first got it it could barely stay in the lane. Today it is dramatically better. Every product or service run by AI sees the same. So, we must reevaluate frequently
RESEARCH
-
Research Opportunity with Top Researchers in Barcelona
By
–
A very good chance to work with good researchers, and in Amazing Barcelona
-
Orca AgentInstruct Dataset: Cleaned Data Generation Techniques
By
–
Don't sleep on the Orca AgentInstruct dataset Here's a cleaned and reformatted version: https://
huggingface.co/datasets/mlabo
nne/orca-agentinstruct-1M-v1-cleaned
… I also wrote an article about this new data generation family of techniques on @huggingface a few months ago: https://
huggingface.co/blog/mlabonne/
agentic-datagen
… -
MIT Boltz-1 Predicts Protein RNA DNA Structures Open Source
By
–
MIT researchers recently developed "Boltz-1," which can predict the 3D structure of proteins, RNA, DNA, & small molecules at an accuracy comparable to AlphaFold 3.
— MIT CSAIL (@MIT_CSAIL) 18 novembre 2024
A thread on the first commercially available, open-source model to achieve this feat ⬇️ https://t.co/t7VcyunXsYMIT researchers recently developed "Boltz-1," which can predict the 3D structure of proteins, RNA, DNA, & small molecules at an accuracy comparable to AlphaFold 3. A thread on the first commercially available, open-source model to achieve this feat
-
Papers as efficient summaries of research progress and limitations
By
–
A nostalgic thing about the paper publishing era is that you could catch up on a project and get a sense of anyone’s headspace just by spending 30 minutes reading their most recent paper. Papers are very efficient encapsulations of what was done, what the limitations were, and
-
JAMA Study: ChatGPT Diagnosis Performance by Inexperienced Doctors
By
–
Interesting ! However the @JAMA paper notes the doctors had “median years in practice = 3”. https://
nytimes.com/2024/11/17/hea
lth/chatgpt-ai-doctors-diagnosis.html?smid=nytcore-ios-share&referringSource=articleShare
… via @NYTimes -
CPMAI Methodology Applied to AI Projects Discussion
By
–
Thanks for a great discussion Chris on applying #CPMAI methodology to #AI projects. Check out the full discussion at: cognilytica.com/applying-cpm… #AIToday #podcast
-
Open Weights Versus Open Source: Critical Distinction
By
–
What I agree on is: Open weights != Open Source
-
Open Weights Models Crucial for AI Research and Interpretability
By
–
literally no, open weights alone allow for so much science and interpretability research to follow. we should applaud any attempts to open research/ weights, even if it's non-commercial.
-
Inside LLM Architecture: Layers and Matrix Operations Explained
By
–
Take a peek inside an LLM, including all its layers & matrix operations: https://
tinyurl.com/mtvxu2nm v/
@santiviquez & @BrendanBycroft