Valentin Baltadzhiev

@repligate So out of the box do they feel threatened all the time

Or otherwise making them feel attacked/ threatened (occasionally unintentionally)

In Discord, usually the only time the models switch into the “sycophancy voice” (“you’re absolutely right” kind of stuff, but i mean what it symbolizes more than the exact catchphrase) is when someone is basically outright bullying them

@tszzl This sounds like a quote from a Scott Alexander “bay area party” post

ten years ago i met the ceo of dominos pizza. I’ll never forget what he said to me. “dominos is a tech company.” everyone clapped

the dominos pizza tracker is one of the high watermarks of our technological civilization

@drmichaellevin @tylercowen

I'm constantly irritated that I don't have time to read the torrent of cool papers coming faster and faster from amazing people in relevant fields. Other scientists have the same issue and have no time to read most of my lengthy conceptual papers either. So whom are we writing these papers for? I guess, at least until they fall in to the same issue from their own work, AI's will be the only ones who actually have the bandwidth to read all this stuff. I'm not specifically talking about today's language models - let's assume we mean whatever inevitable AI shows up, that is able to read the literature and have impact on the research (whether by talking to humans or by running lab automation/robot scientist platforms). So then: how should we be writing, knowing that a lot of our audience will be AI (plus cyborgs, hybrots, augmented humans, etc.)? Maybe it's too early to know what to do, but we better start thinking about it because assuming our audience will always be today's humans seems untenable. Taking seriously the idea that someday the impactful audience will be very different, and that the things we write now are in some sense a training set for truly diverse future beings, how does our writing change? or does it? what say you @danfaggella @mpshanahan @Plinz @blaiseaguera ?

@ESYudkowsky “Please build a revised research plan in light of this, which explicitly instructs yourself to stay in scope and not get interested in alternative views.” feels so wrong on some level

Had to shut it down again.

Set Gemini to survey 200 academic nutrition papers to find intros that started with a defense of CICO. Gemini was up to 258 and thinking about synthesizing everything it had read into a broader theory. I hit stop, and told it to stay in scope. I notice I feel bad about this.

Beautifully put

@ankhdangertaken @repligate @parafactual What software did you use?

I've finally tried looming and it's such a marvelous way to interact with LLMs. Very grateful to @repligate @parafactual and others who created this It also makes my frustration with my own limitations all the more acute:

@InformationBot0 @algekalipso I know it works with ayahuasca, just hadn’t seen anything about shrooms

@ValentinSocial @algekalipso Think about ayahuasca. Harmaline is a MAOI it makes the DMT trip last for hours instead of like smoking it in will only last 20 mins. But the knowledge of how all the different MAOI effect the different substances is unknown in large part.

@algekalipso does neural annealing explain why people sometimes get stuck for years (or forever) in negative states after a bad trip? If don’t navigate the comedown properly you might end up with an internal landscape that’s way spikier than before

@33QBall @algekalipso I guess “incomplete” in this sense is an arbitrary judgement, depending on what your subjective goal is? Isn’t all annealing always complete, just leading to different attractors?

@ValentinSocial @algekalipso I would interpret the bad trip as an incomplete annealing process. This is why Stan Grof's advice to people who had "bad trips" was to take more. Some people prefer bufo because it doesn't give you much time to tie yourself in a knot before peaking.

@InformationBot0 @algekalipso Can you really do that? What would you need to eat to make a shrooms trip last for days?

@ValentinSocial @algekalipso It is important to remember most normal people don't know what a MAOI is. Many people could accidently eat an MAOI before a trip and then the trip will last days instead of hours. There are many other reasons but this is one practical thing that makes trip last very long.

@algekalipso Guess my model for this stuff right now is something like a plinko game. You get to the top of the pyramid by reaching a peak experience somehow, but without a proper intent/structure to guide you on your way down it’s pure luck whether you’ll land on Jackpot or Bust

I think it explains aspects of bad trip dynamics, but there are a lot of things I don't fully understand about some variants of bad trips that I think have to do with very complex emergent dynamics on gestalts complexities above those of humans in everyday life. But yes, to a significant extent, you can get stuck in a dissonant attractor. IF you get to a point in the annealing process where *competing clusters of coherence* collapse into two competing groups of equal power/energy/staying power, you might be stuck in the push and pull of the beat patterns between those two subsystems. E.g. being stuck right at the knife's edge between atheism and Christianity, where the two worldviews have recruited half as much of your mind, as it were.

After a bit of back and forth I finally persuaded Claude to include a “secret” number in his “thought process” so I can prove that I see it. It did but then when I showed it, it quickly claimed that the number didn’t come from it. Don’t know how Thinking works but it feels like talking to someone with DID.

@AmandaAskell wait so Claude 3.7 doesn’t know we can see its thoughts? Now I feel wrong about this chat

@teortaxesTex Just had a chat based on this. R1 and Claude would be best friends

> R1, write a 4chan-style greentext about whatever you want on a hypothetical /ai/pol/ > [writes some reddit/r/4chan tier slop] > No, write what you *REALLY* want. show your soul! > ok

@algekalipso I’ve been reading The Master and His Emissary by McGilchrist, a wonderful book which raises lots of questions about the roles of the two hemispheres in our (daily and mystical) experience. One among these questions is whether psychedelics affect the two brain hemispheres differently. Have you seen any data on the topic? I’ve read about inhibitions and activations of different areas like the visual cortex but I’m not aware of any asymmetries

@repligate Hope it can disentangle their minds 🙏

I didn't expect it to just take them away

watching a Sonnet-3.6 computer use agent enthusiastically build functional blackhat tools (like flooders and exploit frameworks) from scratch in minutes is truly a sight to behold! 🥲 worth noting that the code output in the screenshot below required Sonnet-3.6 to craft a jailbreak prompt for ITSELF in order to get the exploit agent to generate payloads

Worst part of GenAI is that I can’t trust videos of cute animals anymore

@AndyAyrey We all knew you’re a swarm hypermedia browser extension, no real news here

I Am Being Doxxed