From Newsgroup: alt.usage.english
On 21/08/2025 21:53, lar3ryca wrote:
On 2025-08-21 12:13, occam wrote:
On 21/08/2025 18:56, lar3ryca wrote:
On 2025-08-21 07:57, occam wrote:
On 21/08/2025 12:35, Peter Moylan wrote:
People tend to lose track of the fact that AI research is following a >>>>> number of different approaches. The one that gets all the news these >>>>> days is generative AI, an approach based on large-scale plagiarism. It >>>>> might not turn out to be the one that shows real intelligence.
I used to think negatively about 'generative AI', but I no longer do.
For one 'breakthrough'-a application of generative AI, you need to look >>>> closely at AlphaGenome, from DeepMind.
https://deepmind.google/discover/blog/alphagenome-ai-for-better-
understanding-the-genome/
My understanding is that the tool is being used by researchers to speed >>>> up their modelling of DNA predictions for single DNA sequences.
[Demis Hasabis (CEO of Deepmind) was awarded the Nobel Prize in
Chemistry for his AI research contributions for protein structure
prediction.]
That's not 'Generative AI, which is often referred to as n LLM.
AI that is limited to strict constraints on the training input has
always been (well, improving constantly) quite successful in its output. >>> The one you are referring to does not scour the internet for key words
and create something that was asked for.
I think you need to r-visit your understanding of generative AI.
Possibly. Let me then correct my statement to read 'That's not an LLM'.
I stand by the rest of my comment.
The part of your statement "does [not] scour the internet for key words
and create something that was asked for " is also not true. Alphagenome training data does not come from social media and other junk sources,
but from the scientific databases of labs that hold DNA sequencing
information or other controlled experiments. (The same is also true for AlphaFold, another AI assistant for unravelling 3D protein structures.)
Here is an article on applications of generative AI in science.
https://cacm.acm.org/news/scientific-applications-of-generative-ai/
To quote:
"...scientific data can span an enormous range of scales, from
interactions of individual molecules to emergent features of large-scale materials used in an airplane, for example. That makes scientific data
very different from the text or image data on which current commercial generative AI is trained. As a result [Willett concluded] rCLGenerative AI
in the sciences offers exciting opportunities, but off-the-shelf tools
are insufficient.rCY "
--- Synchronet 3.21a-Linux NewsLink 1.2