Right, the most frustrating thing about this is that the complete lack of of transparency about how training works (and how the data is used) means it's impossible to confidently state how it all really works
DATA
-
ML-based spam detection requires continuous human labeling efforts
By
–
The only way you can counter it is via ML. You need to use all available signals to make a judgment call on whether an account's activity is spam or not. It's a game of cat and mouse, ever shifting. Very solvable, but it requires actual people labeling your data everyday.
-
Data importance to AI companies questioned and reconsidered
By
–
Also your data is not as important to AI companies as you think.
-
Regulated Industries Legal Obligations for AI Training Data
By
–
And of course:
1) Regulated industries & service firms have legal obligations that supersede these agreements (though some systems are HIPAA compliant, etc.)
2) Even if your data is used for training, it probably couldn't be used to reconstruct the original document in any case -
LLM Privacy Concerns: Training Data Usage and Enterprise Trust
By
–
Privacy still comes up as a big issue when I talk to firms about AI. But the privacy risk is if your inputs are used as training data. Of course, all the big LLMs have plans where they agree not to do that. People are suspicious of LLMs in way they aren't for other cloud apps.
-
Digital Surveillance and Reproductive Rights: Jessica Burgess Case
By
–
No one should have these powers. But ofc there are real diffs that manifest materially for real people. Like Jessica Burgess who's living in prison rn for accessing reproductive care for her daughter in a state that criminalized it. Facebook messages provided key evidence.
-
Dataset Cartography: Managing Training Data in AI
By
–
A year ago, William Huang of Capital One explored ""dataset cartography"" and its value in training data management. Check out this must-watch video to stay updated on the latest trends in the field:
-
LangChain PostgreSQL Integration: Vectorstore, Chat History, Checkpoints
By
–
Announcing a new `langchain-postgres` integration package, implementing: Vectorstore based on pgvector Chat Message History Checkpoint Saver for LangGraph Vector store docs: https://
python.langchain.com/docs/integrati
ons/vectorstores/pgvector/
… Chat message history docs: https://
github.com/langchain-ai/l
angchain-postgres/?tab=readme-ov-file#chatmessagehistory
… Checkpoint -
Large-scale data processing for generative AI model training
By
–
Yeah, anything that has to do with processing large amounts of data. I actually think this would be amazing for helping train new models as well. You could use large batches of image classification and then use the results to train a model at the quality level of DALL-E.
-
Misleading public sector employment percentage change graph
By
–
Yeah that original "percentage change" graph feels very deceptive to me, makes it look at first glance like the public sector employs the majority of the working population