Will there be ice cream?
tl;dr: Theyâre all wankers.
Iâd argue that there arenât two cults, but two factions of the same cult, that share a common base of garbage ideas and assumptions.
Their debates will be as productive as Christian schisms over the make-up of the Trinity. i.e. Did the Son descend from the Father alone, partake of the Holy Spirit as well, or are the Three really One? Supremely important to Christian sects, causing war and slaughter, but confusing and pointless to outsiders.
Interesting that a whole paragraph was devoted to the Friedrich Hayek story about an obscure footnote of aristo-socialism. Was that his favorite after-dinner story at the Mont Pelerin Society?
âWhen my graduate students showed me some of the results that were in this paper, I actually thought it must be a mistake,â said Marzyeh Ghassemi, an MIT assistant professor of electrical engineering and computer science, and coauthor of the paper, which was published Wednesday in the medical journal The Lancet Digital Health. âI honestly thought my students were crazy when they told me.â
At a time when AI software is increasingly used to help doctors make diagnostic decisions, the research raises the unsettling prospect that AI-based diagnostic systems could unintentionally generate racially biased results. For example, an AI (with access to X-rays) could automatically recommend a particular course of treatment for all Black patients, whether or not itâs best for a specific person. Meanwhile, the patientâs human physician wouldnât know that the AI based its diagnosis on racial data.
Yes, sue the plagiarism machines!
Initially the AI companies tried to give the impression that all the source material was processed into some concentrated holographic quantum AI paste during training, so that there was no copyright infringement. Now it looks like theyâre brute-forcing the whole damned sample set!
No Fâing wonder they need so many energy-sucking planet-warming data centers and all the GPU chips to run it!
This reminds me of a gadget from a Vernor Vinge story: an FTL communicator that works in a Slow Zone where thatâs normally impossible, but the cheat significantly dims the local sun for only teletype data rates.
Cool that the trick works, but a helluva cost, and it doesnât scale.
The valley bros have to know this, but want to raise their number to the moon before this catches up with them.
Quite. I believe the ability to brute force the process is the actual scientific breakthrough behind all this bullshit. In 2012 and it took a few years for scaling to occur. Thatâs Stephen Wolframâs timeline anyway if you have time for a long read this is a good one.
The question of whether many of the things people are suggesting it can do next are even computable is moot. Personally I find the image making and code writing from gen AI to be better than the text writing. This is because Iâm shit at programming and art but Iâm actually literate I think.
I quoted the summary from Gary Marcus in the NYT thread so Iâll copy it here so people donât need to go to the Nazi funding stack:
â The cat is out of the bag:
- Generative AI systems like DALL-E and ChatGPT have been trained on copyrighted materials;
- OpenAI, despite its name, has not been transparent about what it has been trained on.
- Generative AI systems are fully capable of producing materials that infringe on copyright.
- They do not inform users when they do so.
- They do not provide any information about the provenance of any of the images they produce.
- Users may not know when they produce any given image whether they are infringing.
§
My guess is that none of this can easily be fixed.
Systems like DALL-E and ChatGPT are essentially black boxes. GenAI systems donât give attribution to source materials because at least as constituted now, they canât .â
Theyâre massively increasing the size of their models to try to fix inherent problems, but thatâs going to require a non-linear increase in the computing requirements to handle it. (Even if they throw a Really Big hash at it! 
)
Eventually thatâll eat us all for tiny improvements.
If thereâs no one-way transformation of the source material, then they might be able to. They donât want to, because it would be death to their corporate/investment model that depends on pillaging for free.
So I just noticed this new feature on the BBS today and Iâm not sure that I like itâŚ
It does seem rather useless when few bb posts go over 150 words or so.
I think this will interest you @docosc
No surprise. Every atempt to make peds into a checklist, automated procedure has failed. I will have a job for a while yet. Still, 83% failure rate? 83% success rate would be unacceptably low. This is ridiculous.
