Rank: #10 out of 4859 in peer accuracy at Metaculus for the time period of 2016-2020.
ChristianKl
Karma: 37,110
Why is the UI of ChatGPT and co so bad? It seems like it should be relatively easy to create web apps that are responsive and load the information they need in the background in 2026. It also seems like OpenAI isn’t the only company that doesn’t provide responsive apps where the amount the user needs to wait is minimized.
The vast majority of CeSIA’s impact (99%)
How do you know? Even when the intended impact is not via the website it’s still possible that the website has more impact that you intend to and that impact is negative.
Treating humanity and AIs as separate factions is probably not a good mental model conflicts between humans and AI.
That question sounds confused and poses the issues as a binary yes/no question. A model can certainly engage in some reasoning on it’s own. That however does not tell you how good it’s evaluation skills are. They are probably not perfect but also not non-existent.
Religion is a complex topic and is a tradition that does provide benefits like community cohesion. If you do accept that you want to pursue those goals candles are a useful tool which is why the secular solstice that we rationalist have uses candles.
It might not be your taste but probably it’s just something we you lack the understanding of the value it provides.
Explicitly questioning things you don’t understand like Cassava root treatment does not allow you do understand all their benefits.
Generally, understanding cause and effect of dismantling tradition is hard. That’s what the Cassava root example demonstrates.
It doesn’t follow from the text, it’s just based on my anecdotal experience.
It’s not clear to me how anecdotal experience would form a basis for that. Anecdotal experience can tell you that you are valuing tradition less then the average person, but not really whether most of your experiences of getting rid of tradition have complex unforeseen negative experiences that are more like getting rid of Cassava root prep.
I think most people don’t question traditions sufficiently
I’m not sure how this follows from your text, especially given the Cassava root example.
In most cultures, the traditional treatment of acute respiratory tract infections is done via given chicken bone broth. The studies we have suggest that this is more effective than what mainstream medicine has to offer for most acute respiratory tract infections (even when of course it’s not studied well enough to draw definite conclusions).
Even in the West there’s still the cultural memory for that so the relevant review was titled Were Our Grandmothers Right? Soup as Medicine—A Systematic Review of Preliminary Evidence for Managing Acute Respiratory Tract Infections. That’s the kind of thing you expect to see if most people don’t value traditions enough, not that people don’t question them enough.
Even if you take something like no sex before marriage, evaluating it is hard. We do see high divorce rates and low childbirth which both seem to be societal problems.
Did you write the title or was it given to you by the magazine?
As if the central story it is trying to tell is that Leverage fell apart due to no fault of any of its members because they encountered ancient magic or something.
It doesn’t read to me that way. They tried to learn the “ancient magic” for someone who according to “multiple former Leveragers” thought they were a force for evil.
Why might he believed that? The article says “Masters often found themselves at loggerheads. Geoff and David each bragged to their colleagues about how they psychologically manipulated the other, or complained about issues for which they blamed the other. ”
An environment where the leader brag about psychologically manipulating each other is quite obviously an environment if you run into a lot of problems if you experiment with nonverbal techniques that can be used to manipulate each other.
They also engaged in practices to increase their sensitivity to the relevant phenomena in an act of hubris. Then when one person, tried to implement at least some safety protocols they report that Geoff said they did something bad because it discourages investigation.
If someone would want to learn these things in most spiritual traditions both the hubris and the low morals involved in the consent problems around using nonverbal techniques to manipulate each other are recipes for disaster. Both of these things are faults of the members.
I do find it quite plausible that this is the key dynamic that lead them to shut down their group house when they did it.
It somehow completely fails to cover Leverage deploying spies into other organizations and trying to take over CEA. It fails to cover the enormous number of straightforward lies told by Leverage staff in that and other contexts.
That’s probably an interesting story to tell there but the CEA people don’t seem to want to talk about it and Goeff says he doesn’t want to talk about it because he doesn’t want more conflict with CEA. It’s might be harder to get the ground truth there.
The article does say “But almost all Leveragers devoted hours to studying self-examination using techniques forged by its founder”. I’m not sure that a lot of that went under the term debugging.
Both charting and belief reporting are key terms that leverage uses.
While you could do an uncharitable description where a person with imposter syndrome who believes that they are not good enough to drop that believe by a higher ranking member as “High-ranking member tells low-ranking member, ‘Why do you not agree with me? Let’s spend hours digging into what is wrong with you that you would think that, until you decide you agree with me after all.’”, you are losing quite a lot of understanding when doing so.
Belief reporting happens to be much more effective than a framework like Bryon Katie’s The Work for that purpose.
Ultrasound is quite noisy 4D data. I don’t think there’s a good reason to assume that what an AI can do with it is limited by what kind of conclusions humans that look at ultrasound tomography.
The ACX post treats it like the goal is to do what an fMRI does. This misses the potential of having technology that can answer questions about the causes of pain that your fMRI does not answer (or at least does not answer till someone invest the money into building good AI for it).
Claude’s Constitution lists hard constraints that entail behaviors forbidden to Claude. They include providing serious uplift with CBRN weapons
Yet, at the same time doing so was not in the red lines that a had for their military deal this year where they offered to drop limits around helping with weapons creation that their earlier deal seems to have. I wonder what Claude thinks about Anthropic betraying it like that.
The current hard constraints on Claude’s behavior are as follows. Claude should never:
That would be a lie. “Create cyberweapons or malicious code that could cause significant damage if deployed” is done by Claude today in the service of the US military.
Midjourney Medical Ultrasonic CT does have the potential to become an AlphaFold like project. I think the main reason the project wasn’t started earlier by someone else is that it costs a lot of capital to develop and not so much that the foundation wasn’t there.
There’s probably a product that uses any smartphone camera to estimate the 3D surface of a person and produce a better metric for being overweight than BMI.
There’s probably a product that using brainwave information to draw new inferences.
When it comes to commercial projects that I’m publishing on Etsy, I’m right now working on more of a general pipeline to create lots of objects, so I don’t have anything to show.
However for personal use, I had a problem: I want my pan to hang at the wall of my small kitchen. The existing Command Hook didn’t really fit with the pan and the solution I create with Sugru and the existing Command Hook isn’t ideal. Here’s a ChatGPT conversation about it creating my new hook (I did give it a base for the hook from Thingiverse). I probably did not have that conversation in an optimized way, so I think it’s good at showing how ChatGPT deal with this problem and prompts to refine the solution.
Besides the ChatGPT part, I used to think creating a new 3D product to sell means I have to contract a Chinese or Indian factory to produce a bunch of inventory and ship that to a warehouse and handle inventory. The fact that you can now just do it print-on-demand and have an etsy store that only allows return when the item is broken, makes the process of putting the 3D objects into the real world a lot simplier.
I think a lot of nerds write a lot of code and have code projects because all the infrastructure is simple. I think it’s valuable to know that the infrastructure is now also here for 3D objects, and there are going to be massive opportunities in the space.If you use imagen to create another T-shirt that looks just like all the other T-shirts via print-on-demand there’s huge competition. On the other hand, right now the competition for print-on-demand physical objects is relatively low. It likely still needs a certain amount of creativity, but if someone like tinkering and inventing things and previously thought that all the infrastructure just makes it too hard to put into practice, now here’s a space where a new online business can potentially thrive.
I haven’t tested it with both, but I would expect that Claude can do it as well. Anthropic did a relatively large donation to Blender to make it have a better interface for agents, so it’s on their list of things that Claude should be able to do.
But I don’t how else to gesture at actually being uncertain, especially given practical time constraints!
If it’s about the whole post, epistemic status declaration are can be helpful in that regard.
Just if anyone is unaware of current technology. You can use ChatGPT to let it design 3D objects, 3D printed out of multiple different plastics by printie.com and sell those objects via print-on-demand via etsy.
The barrier to entry to producing a new 3D product and selling it got really low and that knowledge doesn’t seem to be widely understood so there’s a market idea if you have innovative idea that can be solved within that tech stack.
Scott says something like, let’s assume we aren’t living in reality when he says “We’ll very optimistically assume that after enough tests, the smartest doctors can distinguish cancers that should be treated and cancers that shouldn’t be treated with 100% accuracy”.
We do have studies that looked at how survival rates change after various cancers screenings. To ignore that randomized studies and make arguments based on what spherical cow analysis suggest, is not a good basis for making simplified claims.
We might not have the big studies for full body scanning for cancer detection, but the results of the studies for cancer screening we do have can only be explained with doctors making many bad treatment decisions or there be other negative effects that come from cancer screening tests.
Allowing people to become vicious enemies, without allowing either side to resort to personal violence, means that people can do more ambitious things while risking others hating them for it. Now that I write it out, I’m not quite sure why this is good? If you do things that are worthy of people hating you and wanting to kill you, perhaps you should in fact not do them?
Duels are a way to punish your enemies that requires high personal sacrifice. I think we have plenty of ways in today’s society where a person can take actions to be annoying to people they dislike. Some of them legal others like squatting aren’t legal but hard to prosecute and probably less risky then asking someone to a duel to the death.
The main thing that our society does, is that it doesn’t make it a honorable course of action to go to extreme length to punish your enemies, the power to punish is still there.
Factions happen because of organization. OpenAI is an organization made up of both humans and AI that cares about OpenAI having more power.