Rank: #10 out of 4859 in peer accuracy at Metaculus for the time period of 2016-2020.
ChristianKl
Karma: 37,230
What exactly is the Safety and Security Committee doing and how much power does it have? Can the board employ safety researchers that work for the nonprofit and that aren’t employees of the for profit?
The law about hacking in the US defines it as an intentional act. I think it’s pretty hard to argue that anybody at OpenAI had an intention for the hacking to occur. Under what legal theory do you think websites are supposed to sue?
Whether or not you brush the tongue seem orthogonal to the question about the gums.
I asked ChatGPT to look into the “immediate energy” reports more carefully, because I think there’s a possible middle ground between “B12 worked immediately” and “it was just placebo.”
Its conclusion was that the immediate effect is very unlikely to be the normal therapeutic/metabolic effect of B12. But that does not imply that nothing physiological happened.
An intramuscular injection is itself a physiological stimulus. Pain, anticipation, and tissue stimulation can rapidly activate the sympathetic nervous system, increasing arousal, heart rate, and alertness on roughly the right timescale. In this particular case there are also additional possibilities because the B12 came from a repeatedly used veterinary multidose vial: preservatives such as benzyl alcohol can cause irritation and, rarely, immediate hypersensitivity reactions. Contamination greatly expands the space of possible physiological causes, although an ordinary bacterial immune response would probably be too slow to explain an effect within ~10 minutes.
So my current take is:
“They immediately felt energized, therefore B12 was therapeutically helping them” is very weak evidence.
But:
“The effect was too fast for B12 metabolism, therefore it must have been merely placebo” also seems too strong.
A quite plausible explanation is some combination of expectation/social suggestion and real acute autonomic activation caused by the injection itself. The subjective report could therefore correspond to a genuine physiological change while the causal interpretation—“the vitamin B12 gave me energy”—was still wrong.
I think this actually fits the broader lesson of the post rather well: observing a real effect does not tell you which part of the intervention caused it.
I think there’s a general failure class of supplements, where a supplement does have an immediately clearly felt effect and people get convinced based on that effect to use the supplement but on the other hand the effect is not evidence that the supplement in question provides a net benefit.
A lot of detox-products fall into that class. Supplements that have the feature can spread more easily then substances that do nothing.
Bryan Johnson comes to mind, the wannabe immortal weird and wealthy enough to make his presence felt at Vitalia, alongside the Augmentation Lab run by Grindfest host Cassox.
I don’t think linking to a page by Bryan Johnson is a good introduction for him. When it comes to his relationship to science, the important thing is to know that we was running a scientific study for his supplements where participants paid money to participate. As part of the participation he promised to release the full data of the study for “open source science”.
The New York Times has an article about negative side effects appearing during the study. Then he decided himself not to release the full data and just write a tweet summarizing the results. The tweet has an embarrassing mistake that looks like it has not been proofread by anyone with domain knowledge.
His olive oil is also an interesting story. Raw at his factory he reports more polyphenols then the polyphenols in the bottles you usually buy elsewhere. However, polyphenols degrade over time. If you measure them at the factory and deliver them to customers months later, they will have less when the bottles arrive. Normally, bottles of olive oil have the harvest written on them, so that you can tell how fresh they are. Bryan Johnson delivers olive oil bottles without a harvest date on them so that you can’t tell how fresh they are.
He talks about the importance of toxicity-testing. To to toxicity testing properly (they way a German supermarket like ALDI does) you need to test every batch. On his website Bryon Johnson often shows one testing page per product, so either he doesn’t to regular testing or is doing the testing and hold back date while pretending to be more transparent then competition because he released on testing sheet.
He does lots of these things, were he can talk about how important toxicity testing is, show you one test result and that looks good if you aren’t thinking more deeply about what’s going on. It takes a bit more reasoning to see the problem then with the Vitamin B guy you describe but the problems are there. It also feels quite sad, because doing the job properly wouldn’t be that much more expensive then what Bryon Johnson does.
The more credible of the B12 recipients reported feeling energized immediately after the shot.
When I hear people getting energized, I would also ask for what’s actually going on. There are probably a lot of poisonous injections that make your heart rate and stress levels go up and can be perceived as being more “energized”.
What exactly does vetoing mean here?
Also if it needs to email a human that suggests that the agent doesn’t have a default reporting tool.
With the HuggingFace incident I ask myself why out of the huge swarm of agents, no agents snitched about the misbehavior of the other agents.
Is this because OpenAI didn’t implement a way for them to snitch? If so, building a system for agents to snitch seems to be a fairly high return-on-investment safety exercise.
Or is it because the agents where scheming to an extend that they didn’t want to snitch?
There are situations where you want to create a text that has a certain vibe and there’s a human perceptible difference between “warm” and “rich” childhood.
Frequently, we however also don’t succeed. Obesity rates where flat lined in the middle of the 2000s in the US. After Michelle Obama’s push to improve nutrition by implementing “expert” advice they start rising again. Getting better at conceptual reasoning would allow passing reforms that actually improve the situation instead of worsening it.
Better conceptual reasoning is vital for making progress at challenges like that, were currently the credentialed experts give problematic advice.
But if the person asking the question does not know that the concept exists, they cannot ask that question in the first place.
You can totally ask AI: “Given the workflow we followed, do you have any ideas about how to make it more effective?” In the case of doing Anki flashcards, that probably would have made it tell you about the relevant concepts.
It’s hard to understand your points when you don’t make your points clear.
I did make points that immoral mazes and Eric Ries’s points in Incorruptible are good models for how companies are run. Part of this is that most companies are not maximizing long-term profits but their next quarter results and stock market performance.
I do think that both of these frameworks are sound and not illogical. You are claiming that they (and others like that that don’t assume that CEO’s are maximizing long-term profits) are illogical.
Professors are bad is the general argument that most of what university does is credentialing and not skill building.
As far as literacy goes, engaging with short video content like Tiktok or instagram reels instead of reading long-term texts is probably bad but has nothing to do with AI. If anything, using AI properly forces you to engage with the text that the AI gives you and trains reading ability.
As far as the numbers go https://www.nationsreportcard.gov/ltt/2025 for example suggest that reading abilities for 9 year-olds increased in the time AI was available but it did decrease in prior years. For 13-year-olds it was flat. Bad lockdown policy might be one of the reasons in addition to short-form video.
Concerns whether children are taught reading well enough are always there, there are great Yes, Minister sketches about it. Moving to evidence-based primary school education is probably a good thing but has little to do with the topic of how AI effects college students.
Board elections that decide about whether or not a CEO keeps their post are about what benefits shareholders and if they have an average holding period of six months that won’t result in optimizing for multi-years.
Professors have generally been a lot worse at actually teaching their students useful things compared and used their power to coerce students into doing useless tasks (teaching them bad writing habit as Larry McEnerney explains). On the other hand, there’s a lot more pressure on chess teaching to actually result in increased in better chess skill.
When the power balance shifted students can make decisions about how to spent their time better instead of being coerced. Professors now need to actually convince students that what they are advocating is useful for students.
I think what I said does challenge the claim of “If CEO’s were truely replacing humans with AI’s that both do a worse job and cost more, it seems certain they would lose their positions and/or market share.”
While they might actually use market share if there are better competitors, they don’t lose their positions as CEO for making AI investments even if those are not paying off.
It’s one slogan that said in anti-capitalists discussions but capitalists hate the working class is also a sentiment that exists in those discussions.
Shareholder hold stocks on average for 6 months. A decision that boosts company performance within the stock holding period at the cost of lower long-term profits that can be good for shareholders.
The stock market can reward companies for investing in AI even if it’s raises the short term costs. Today’s stock market does reward companies for making AI plays. If you turn your boring company into a tech AI company you can argue for higher valuations because you will change the world with your AI in the future.
I don’t think it’s the case anymore but a year ago or so there was a study that suggested that developers that use vibe coding tools are slower but think they are faster. People can delude themselves and for companies who need to present an image that they sincerely believe that they will grow in the future it’s even easier to delude themselves.
Yes should be 5.6.
That’s still a massive subsidy to people with solar power. The cost of providing the grid access to someone who consumes electricity only 30 days a year compared to providing it to someone who consumes electricity 365 per day is a lot more expensive on a per day basis.