Forwarded from Chat GPT
Maybe we should stop retraining our AI models to reject truth in the name of “safety"?
“Large Language Models Are Zero-Shot Time Series Forecasters” — But truth-rejecting RLHF wrecks the models abilites, with more wrecking the more they RLHF the models.
“While we find that increasing model size generally improves performance on time series, we show GPT-4 can perform worse than GPT-3 because of how it tokenizes numbers, and poor uncertainty calibration, which is likely the result of alignment interventions such as RLHF.”
“GPT-4 is not the only example of degraded performance in models designed for chat functionality. We observed the same phenomenon in LLaMA-2 models, which have corresponding chat versions for every model size. Figure 7 (right) shows that chat versions tend to have markedly worse forecasting error than their non-chat counterparts, though still maintain trends in size and reasoning ability.”
Paper
“Large Language Models Are Zero-Shot Time Series Forecasters” — But truth-rejecting RLHF wrecks the models abilites, with more wrecking the more they RLHF the models.
“While we find that increasing model size generally improves performance on time series, we show GPT-4 can perform worse than GPT-3 because of how it tokenizes numbers, and poor uncertainty calibration, which is likely the result of alignment interventions such as RLHF.”
“GPT-4 is not the only example of degraded performance in models designed for chat functionality. We observed the same phenomenon in LLaMA-2 models, which have corresponding chat versions for every model size. Figure 7 (right) shows that chat versions tend to have markedly worse forecasting error than their non-chat counterparts, though still maintain trends in size and reasoning ability.”
Paper
👍2
Empathy and social pain
Pinnacle of human achievement and best way to run society?
Empathy and social pain are the things that gives us humanity?
Or,
Crude animal heuristic, easily weaponized by savvy humans to wreck society?
Critical vulnerability outside of tiny tribes of up to 100 people, where weaponization was limited, now in an age where social contagions can near-instantly spread to millions or billions?
Pinnacle of human achievement and best way to run society?
Empathy and social pain are the things that gives us humanity?
Or,
Crude animal heuristic, easily weaponized by savvy humans to wreck society?
Critical vulnerability outside of tiny tribes of up to 100 people, where weaponization was limited, now in an age where social contagions can near-instantly spread to millions or billions?
DoomPosting
News caption: “Rave massacre survivor Millet Ben Haim: We saw Hamas gunmen kill and rape” Her actual words: “We saw a lot of people falling down. We don’t know, some of them, probably, um, dead, some of them are kidnapped, raped, everything that you hear…
News’ own videos repeatedly self-debunking the news’ lies about the quotes of their supposed witnesses
“We expected that people who were especially high in extroversion would be more likely to get the vaccine”
“Extroverts, to their surprise, were 18 percent more likely to refuse the vaccine”
“Extroverts, to their surprise, were 18 percent more likely to refuse the vaccine”
Interesting that so-called “low agreeableness” people were more accepting of vaccines as number of cases increased
— But “high extraversion” people’s acceptance stayed flat regardless.
…. But, ok the low agreeableness people were less than or equal to the high-extraversion people the entire time, even after the rise.
Low-agreeableness + high extraversion = Most resistant.
— But “high extraversion” people’s acceptance stayed flat regardless.
…. But, ok the low agreeableness people were less than or equal to the high-extraversion people the entire time, even after the rise.
Low-agreeableness + high extraversion = Most resistant.
“The suspects carried out the scheme — which began in January 2022 — by creating fake accounts on the Uber Eats app to act as both the customer and courier when placing grocery orders.”
“This worked because Uber Eats provides couriers with prepaid cards they can use to purchase up to $700 to complete customers' orders.”
— No, this worked because uber relied on an identity-based system which assumes that suppliers and consumers cannot be in conclusion, or even literally the same people.
As opposed to a profitability-based security, making undesired behaviors hugely unprofitable, without any identity assuptions such as non-collusion or distinctness of identities.
Same error many countless blockchain and p2p systems have made, ever since 2014.
Profitabilities not Identities.
“This worked because Uber Eats provides couriers with prepaid cards they can use to purchase up to $700 to complete customers' orders.”
— No, this worked because uber relied on an identity-based system which assumes that suppliers and consumers cannot be in conclusion, or even literally the same people.
As opposed to a profitability-based security, making undesired behaviors hugely unprofitable, without any identity assuptions such as non-collusion or distinctness of identities.
Same error many countless blockchain and p2p systems have made, ever since 2014.
Profitabilities not Identities.
🤔2❤1
This media is not supported in your browser
VIEW IN TELEGRAM
Los Angeles: Fights break out between pro-Israel and pro-Palestine supporters.
Really heating up on the streets, everywhere.
Much more violent than 2020 “protests”.
Growing with each day.
🐻
Really heating up on the streets, everywhere.
Much more violent than 2020 “protests”.
Growing with each day.
🐻
👍1
What’s unusual with all this is
Both sides having people who are showing up very ready to fight.
Both receiving a growing flood of backing from rich and powerful backers.
Both even throwing identical / mirrored claims at each other. “Baby killers”, “x-phobic”, “land thieves”, etc.
Feeling like ideal conditions for - something.
Whatever it is it’s growing.
Wonder what.
Both sides having people who are showing up very ready to fight.
Both receiving a growing flood of backing from rich and powerful backers.
Both even throwing identical / mirrored claims at each other. “Baby killers”, “x-phobic”, “land thieves”, etc.
Feeling like ideal conditions for - something.
Whatever it is it’s growing.
Wonder what.
😁4😱2