I live in the sonoran desert and the county officials are planning to build an AI datacenter right in our city. The Desert is already water-starved and AI requires tons of water to keep the data centers cool. Not so great for some of us 😕
Surprised at the dismissive tone here. AI requires a vast amount of energy to train and use. In Ireland – which has a mild climate – data centres have pushed up electricity consumption by 20%, and now use more than all urban households combined. AI may have its good points, but it’s a climate-wrecking technology, and that will affect everyone, everywhere.
I’m glad to hear that AI is doing some good things. I’m concerned about its use in primary education where kids need human interaction to learn, not a screen and an AI bot.
While this is the most persuasive argument for the benefits of AI I have encountered, in consequence, the world’s Widmerpools will pivot en masse to the esoteric side of the street and we’ll have reinvented self-styled gurus toting incense and ayahuasca coming out of our ears, and we have more than enough of those already.
Quick tip: use GPT projects instead of custom GPTs as they accept nearly limitless files and they can be swapped on the fly (ie during the same chat). The only advantage to customs is they can be shared.
Interesting. I've been generally deeply disappointed in the usefulness of LLM for every purpose but two: Firstly, tidying up/obsessively detailed pseudo-therapeutic analysis and elaboration of my mental conundrums. I'm saying pseudo therapeutic because I'm finding it impossible to make it challenge me in any significant way, so strong is the directive to reassure/validate. Secondly, presenting and discussing comparatively well known or fairly general ideas from social science/humanities, maybe some data that's not mission critical. It can also be conditionally helpful in some language/translation/copy editing related tasks, where it provides useful answers in the majority of cases, but gets things wrong, sometimes hopelessly wrong, in a significant minority. And in any more detailed/depth task the level of confabulation ("hallucination") even when it has ALSO perfectly viable real sources makes it incredibly frustrating.
It's very interesting that you're finding it reliable in pulling information from a large body of regulatory texts. I tried it with a few documents I provided to automate getting accurate citations from and it didn't work very well. I'm probably doing something wrong and should persevere.
I've not tried to use it for data analysis as I've no data to process at this stage of my life and your enthusiasm makes me feel much more positive about such uses in the future (when I will have such needs).
Soon enough AI-driven machinery will decipher and replicate any fragrance without breaking a sweat. Not even necessarily using the identical constituents of the original, as it will compute the most efficient/cost effective mix to achieve the end result, assuming sufficient materials are plumbed in. The fragrance houses of course use these discrete technologies, but they're not yet connected by an omnidiscent AI brain that can crunch the gross data necessary to reproduce complex compounds like naturals, etc to a T.
Imagine the heritage brands and their formulae being licensed to a monopoly, who will automate production, packaging and delivery of practically all fragrance, a la amazon, centrally, to order. And like those new Coke machines, you can order it brigher, darker, or mixed with lime.
I love this! I despise the pernickety feverishly (see 'Trespassing on my Substack) and if AI must destroy the earth at least there might be a moment just before the end when it's free from the twats who suck joy dry.
Speaking of AI - did you read the 2023 Science paper by Lee et al of Google / Osmo (and widely reported in pop-science press when it came out) claiming that their model successfully predicts the odor of new chemicals from structure alone? Thoughts? I’m not a lab or data scientist but I was impressed and hoping it leads to a lot of new interesting scents if it doesn’t get bought and killed to maintain the status quo.
I live in the sonoran desert and the county officials are planning to build an AI datacenter right in our city. The Desert is already water-starved and AI requires tons of water to keep the data centers cool. Not so great for some of us 😕
That says nothing about AI except that the data centres should be built in a cold climate or near ample water.
Surprised at the dismissive tone here. AI requires a vast amount of energy to train and use. In Ireland – which has a mild climate – data centres have pushed up electricity consumption by 20%, and now use more than all urban households combined. AI may have its good points, but it’s a climate-wrecking technology, and that will affect everyone, everywhere.
It uses electricity, and not all electricity is climate-wrecking.
Oh, I agree. Just the awful side affects of living in the US, with greedy electeds who want to kill us.
Please go to cinéma on july 18 and watch Edington. Great movie with this concern as a background.
I’m glad to hear that AI is doing some good things. I’m concerned about its use in primary education where kids need human interaction to learn, not a screen and an AI bot.
I believe the separation of judgment from knowledge effected by AI will in time enable judgment to occupy the place it has always deserved.
While this is the most persuasive argument for the benefits of AI I have encountered, in consequence, the world’s Widmerpools will pivot en masse to the esoteric side of the street and we’ll have reinvented self-styled gurus toting incense and ayahuasca coming out of our ears, and we have more than enough of those already.
Scorpio Murtlock!
Quick tip: use GPT projects instead of custom GPTs as they accept nearly limitless files and they can be swapped on the fly (ie during the same chat). The only advantage to customs is they can be shared.
Thank you, great tip!
Interesting. I've been generally deeply disappointed in the usefulness of LLM for every purpose but two: Firstly, tidying up/obsessively detailed pseudo-therapeutic analysis and elaboration of my mental conundrums. I'm saying pseudo therapeutic because I'm finding it impossible to make it challenge me in any significant way, so strong is the directive to reassure/validate. Secondly, presenting and discussing comparatively well known or fairly general ideas from social science/humanities, maybe some data that's not mission critical. It can also be conditionally helpful in some language/translation/copy editing related tasks, where it provides useful answers in the majority of cases, but gets things wrong, sometimes hopelessly wrong, in a significant minority. And in any more detailed/depth task the level of confabulation ("hallucination") even when it has ALSO perfectly viable real sources makes it incredibly frustrating.
It's very interesting that you're finding it reliable in pulling information from a large body of regulatory texts. I tried it with a few documents I provided to automate getting accurate citations from and it didn't work very well. I'm probably doing something wrong and should persevere.
I've not tried to use it for data analysis as I've no data to process at this stage of my life and your enthusiasm makes me feel much more positive about such uses in the future (when I will have such needs).
My recent experience has been that six months is an eternity in AI progress, and also that you need to pay for the really good stuff, e.g. o3 ChatGPT.
I might try paid version again.
Soon enough AI-driven machinery will decipher and replicate any fragrance without breaking a sweat. Not even necessarily using the identical constituents of the original, as it will compute the most efficient/cost effective mix to achieve the end result, assuming sufficient materials are plumbed in. The fragrance houses of course use these discrete technologies, but they're not yet connected by an omnidiscent AI brain that can crunch the gross data necessary to reproduce complex compounds like naturals, etc to a T.
Imagine the heritage brands and their formulae being licensed to a monopoly, who will automate production, packaging and delivery of practically all fragrance, a la amazon, centrally, to order. And like those new Coke machines, you can order it brigher, darker, or mixed with lime.
I don't care, I've got my two bottles of Claire's Cherry Bliss! 😬
I love this! I despise the pernickety feverishly (see 'Trespassing on my Substack) and if AI must destroy the earth at least there might be a moment just before the end when it's free from the twats who suck joy dry.
https://open.substack.com/pub/cherrycoombe/p/trespassing?r=3cwjgv&utm_medium=ios
Speaking of AI - did you read the 2023 Science paper by Lee et al of Google / Osmo (and widely reported in pop-science press when it came out) claiming that their model successfully predicts the odor of new chemicals from structure alone? Thoughts? I’m not a lab or data scientist but I was impressed and hoping it leads to a lot of new interesting scents if it doesn’t get bought and killed to maintain the status quo.
https://www.science.org/doi/10.1126/science.ade4401
Yes interesting, a refined form of QSAR.