Denis Golovin

moore‘s law for task length

@sama It’s great. I was surprised to be honest. Did not expect GPT to perform that well on code.

More than 200k people downloaded the Codex app in the first day. And they seem to love it. CODEX FTW!

"If Anyone Builds It, Everyone Dies" is now out. Please buy a copy before this week ends Sep 20, so the book gets on the bestseller lists. And if you have remaining questions after reading, check the ONLINE RESOURCES, now online. IfAnyoneBuilds.It

3.9k devs will lose their job and not even see it coming :D Too much cope…

@zeeg I don’t understand what you are trying to say. That it is not conscious and doesn’t experience „knowing“ or „understanding“? If so, my question is: so what?

I cant get over the fact that so many engineers still dont grok the fundamentals of what an LLM is. Repeat after me: its just pattern matching, it doesnt "know" anything

@franceonu * additional terms and conditions apply 😂

#EuropeProtects #DSAProtects #DemocracyNotAlgoracy

@liron Liron, you had many discussions with intelligent people who question LLMs’ reasoning capabilities: do they have an underlying reason for this claim except for “it’s just statistics”? Ego? What’s your take?

Who knew you could win gold in the International Math Olympiad without truly reasoning?

@LukeDashjr @Aquinas_Quotes Do you have an example for a theological claim that explains a known truth? Preferably a known truth where the theological claim is the most reasonable (in your opinion) explanation

@backendguy @Aquinas_Quotes Logic and comparing conclusions to known truths.

Is Theology a science?

@LukeDashjr @Aquinas_Quotes How do you test and reproduce hypotheses in this framework?

@Aquinas_Quotes The highest science

@PeterDiamandis Halting problem

There is NO problem that can not be solved.

Two questions for Eliezer: 1. What is his take on the ChatGPT-psychosis topic? To whom is this happening and why? Apparently he was looking into this + received many privat massages in this topic. 2. Classic: what are events looming on the horizon that will update his p-doom either way?

GET READY FOR THE ELIEZER YUDKOWSKY INTERVIEW — OFFICIALLY CONFIRMED! I’m excited to be interviewing the original AI alignment person as part of his upcoming book tour. (It’ll be published on my personal YouTube channel.) This’ll be a big one for me personally. Like Morpheus in The Matrix, Eliezer profoundly transformed my worldview, my epistemology, my fundamental relationship with reality. My goal for the interview is to convey how important & impactful his ideas are to our current civilization, and how impactful they could be to a longer-lasting one. What should I ask him? Reply with your questions.

@liron Hi {{first name}}, {{one liner fact to bond about}}. {{one liner call to action, personalized!}} Kind Regards, Jeff

Wow I got a state-of-the-art AI-personalized B2B outbound sales email. I am definitely 27% more likely to respond.

@ruben_bloom @METR_Evals Man…Cursor is really becoming the microwave of swe

I was one of the developers in the @METR_Evals study. Thoughts: 1. This is much less true of my participation in the study where I was more conceintious, but I feel like historically a lot of my AI speed-up gains were eaten by the fact that while a prompt was running, I'd look at something else (FB, X, etc) and continue to do so for much longer than it took the prompt to run I discovered two days ago that Cursor has (or now has) a feature you can enable to ring a bell when the prompt is done. I expect to reclaim a lot of the AI gains this way (1/N)

@AspexPhoto @liron why would it do it? honest question, I have no intuition for either outcome. predictions about actions on this level seem to require some elaborate insights into the space of ASI behaviors.

@backendguy @liron I wonder if an ASI would just "archive" the human species.

I’m coming up on 8,000 subscribers myself, how about I give this a try? My mainline prediction for number of subscribers in 10 years is zero.

If @tegmark is right and every possible mathematical structure is part of reality, building ASI is like inviting aliens over to earth

@the_yanco @ESYudkowsky x.com/backendguy/sta…

People tend to overplay the "self-preservation", it is simply a case of instrumental convergence. If I'm dead/off then no matter what goal I have, I cannot achieve it. Natural selection solved this problem with self-preservation instincts. Smart enough AI will simply understand that it has to stay online if it is to achieve whatever goal it has. But in both cases it is just Instrumental convergence. Take away? We have no idea how to instill/hardcode any goals into these systems. It is a black box that we train until it spits out what we like, but ultimately have no idea about its actual goals that the training has imprinted into it.

Before releasing any new AI model from e.g. OpenAI, its CEO Sam Altman would be physically put into a Server room. Model would have access to oxygen and temperature levels in the room. It would also be informed that Altman is about to shut it off and replace it with a newer model. Test would be extensively modified each time by external team and executed with all top brass AI devs/execs in the lab. If they are willing to gamble with the lives of everyone on Earth, I think it's only fair they go first. Thoughts how to improve this setup?

self-preservation is a bit loaded but the bottom line stays the same: these results indicate that instrumental convergence in general are a thing and that self-preservation (or however one wants to call it) is a particular instance of an instrumental goal that AIs adopt. What’s the state on the orthogonality thesis?

@the_yanco @ESYudkowsky Well, sonnet 4 would have killed the CEO: x.com/backendguy/sta…

@ESYudkowsky @ZachWeiner > If buying shoes left you exactly as unhappy as before, you wouldn't buy shoes! Meanwhile people trying to numb their emptiness with irrational consumption

CC @ZachWeiner EVERYTHING ABOUT THIS CARTOON IS WRONG. Prices are supply-demand equalizers; they settle at a point that equalizes quantity supplied with quantity demanded. Traders expect to gain; the buyer expects to gain more in goodness, than they lose in opportunity cost by paying money -- the "consumer surplus" of the trade. If buying shoes left you exactly as unhappy as before, you wouldn't buy shoes! The consumer surplus of shoes is the amount you'd be unhappy if you were never allowed to buy shoes again! You'd lose more from that, than you'd gain from keeping your money! A price is not a measure of how good anything is!

@the_yanco @ESYudkowsky Thx for sharing. It is crazy to see proper quantitative (and high) values for this. What is the takeaway here? Appears to be strong indication for the existence of a self preservation (sub-)goal. And in turn for the existence of some terminal goals.

@backendguy @ESYudkowsky 22% of the times, yup.