Clinicians are gathering iris scans, voice recordings and harnessing AI to crunch through and curate masses of data to spot signs of change that may be early indicators of future problems.
…
These machine learning algorithms have been trained to identify drugs that could convert the neurological disease signature into a healthy one.
They are talking about several distinct uses of different sorts of AI for research here. One of them does seem to be LLMs used to automate data processing tasks.
ah, missed the voice recordings part, I can see an LLM for annotating those. I meant to emphasise this is separate from the LLM assistant and coding models that are the most visible to consumers right now.
Is there a reason to believe they don’t use the same models? You wouldn’t necessarily need something specialized for ‘curating data’. The drug identification stuff is definitely separate though.
For text classification? They might. There’s a million things they could be using. I’ve definitely seen just throwing ChatGPT at text and asking it to generate a label for tagging and classification, but its much cheaper to use fine tuned Roberta or some other encoder-only model. Both are LLMs using a transformer architecture, just one is more what we’re familiar with the other is more meant for text classification tasks.
They are talking about several distinct uses of different sorts of AI for research here. One of them does seem to be LLMs used to automate data processing tasks.
ah, missed the voice recordings part, I can see an LLM for annotating those. I meant to emphasise this is separate from the LLM assistant and coding models that are the most visible to consumers right now.
Is there a reason to believe they don’t use the same models? You wouldn’t necessarily need something specialized for ‘curating data’. The drug identification stuff is definitely separate though.
For text classification? They might. There’s a million things they could be using. I’ve definitely seen just throwing ChatGPT at text and asking it to generate a label for tagging and classification, but its much cheaper to use fine tuned Roberta or some other encoder-only model. Both are LLMs using a transformer architecture, just one is more what we’re familiar with the other is more meant for text classification tasks.