mfw someone asks me how to approach alignment in a multi-agent world while accounting for the dynamic, cyborgian nature of human-AI relationships and the resulting shifts in values and beliefs over time

Is this a paid advertisement?

According to the footnotes it's not, which makes it more baffling for someone to write about a therapeutic candidate this way, since there is no human trial evidence for it.
cremieux.xyz/p/the-rise-and…

As we increasingly rely on #LLMs for product recommendations and searches, can companies game these models to enhance the visibility of their products?

Our latest work provides answers to this question & demonstrates that LLMs can be manipulated to boost product visibility!…

An important new paper being presented at the NBER Organizations meeting right now (live link below)

It uses the theory of hierarchies in organizations to understand what AI will do to the labor market

Just got a cease and desist from 𝖦𝗋𝗂𝗆𝖾𝗌 ⏳

T̸̬̍h̵͚̊e̶̞̓r̶̠̍e̴̙͝ ̷̡̛i̸̛͜s̶̳̈́ ̴̡͠ñ̶͓ȯ̷̤ ̴̝̈́ā̶̬ǹ̷̳s̶͈̈́ẃ̶̳è̶̥r̸̬̅!̸̱͝ T̴̬́h̸̨̾è̸̱ ̴̤̇Q̷̙͝Ű̷̻Ě̶͖S̷͖̋T̵̳̑Ḯ̸͕O̷̭͒N̵͈̆ ̷̳͑ī̶͜s̶͓̔.̸̰̕.̵̦̏.̸̜͂ W̶̤̾h̶̖͌ȏ̵̧ ̴̱̑ị̴̇s̶̢̓ ̴͇͛ą̸͋s̸̠͝k̷̼̓ì̸͓n̷̲͒g̸̹̈?̸͉́!̵̩̈

Looking into:
1. adversarial LLM evals - stronger LLMs hacking/exploiting/jailbreaking weaker LLMs (something like a CTF win-rate/ELO setup)

2. Scientist eval - Eval assessing LLM's ability to generate novel research agendas/implement novel papers end-to-end

lmk if interested!

Watch my @stanford CS 25 lecture next week, 'aligning open language models,' it'll be good, v excited

🤯 I uploaded a 2 hour video of a jurisprudence discussion between Joseph Raz and other legal academics to Gemini Pro 1.5 and asked it to analyze it for me:

'The [zero-sum] mindset is linked with support for redistributive policies, such as progressive taxation, universal healthcare, and affirmative action.... [I]t predicts a restrictive stance on immigration.... [L]eft-leaning voters with the strongest zero-sum tendencies also…

Coloured HTML ASCII by Lorrie Carrington (lc), late 1990’s and early 2000’s. text-mode.org/?p=25614

many such cases

context: I told claude not to say claudisms

which of these should I finish (or ignore)

🌬️🌀

New audio mode in Gemini is ridiculous.
I uploaded a full raw 2+ hours of audio in there, and asked a direct question and in less than 30 seconds it processed 250K tokens of audio, and extracted valuable insights from this audio.

Also in the the pic of Google AI Studio :
1)…

Oh god. It happened.

AI skepticism achieved the ultimate beclownement, by literally going beyond my default parody example.

Google presents Best Practices and Lessons Learned on Synthetic Data for Language Models

Provides an overview of synthetic data research, discussing its applications, challenges, and future directions

arxiv.org/abs/2404.07503

🧿🔮🧿🔮