He would surely be happy to meet a real dragon. The dragon flew to the man's house and went inside, only to find him asleep. When the man woke up, he saw the dragon coiled by the bed, its scales and teeth glittering in the moonlight. The man screamed in terror. >
There are two categories of people: those who quickly figure out that chatbots give you the answer you expect when you ask questions in a biased way, and the ascended polymaths currently out-thinking every expert on Earth
People say large language models are a watershed because, for the first time in human history, we've invented a technology we cannot control - but we haven't been able to control printers for 40 years.
“When you have eliminated the impossible, whatever remains is often more improbable than your having made a mistake in one of your impossibility proofs.” -- Steven Kass
@azeem@balajis Stress tests may not have been required for SBV due to 2018's S.2155 bill. SBV lobbied for this change and stayed below the threshold. https://t.co/erkrxn7D5Z
Incidentally this is why crypto has not yet taken off as a developer platform, it hasn’t unlocked fundamentally new powers for devs. It’s theoretically & economically interesting, but it’s not transformative for devs.
Three of the top 5 posts on Hacker News are about different but related debates on version control of open source projects: automation of software development, renumeration for open source contributions and legal implications of large language models trained on open source code
We’re facing a difficult problem with the open-source world.
There are no open-source models with comparable performance to Codex.
@StabilityAI will eventually train near data-optimal code models, but will they risk training on codebases with no permissive licenses? (5/7)