To some extent, having slop in feeds is a platform’s choice, either implicit or explicit. In LI’s case it seems to be explicit since they added a slop button. YouTube has a similar problem with Shorts.
Based on the numbers here it seems there’s less than 800 tasks in the entire benchmark. That is enough for a handful of engineers to comb through in a week (which is what OpenAI eventually did here).
On the one hand, kudos to them for actually doing that work.
On the other hand, garbage in, garbage out. It’s a bit embarrassing for the original authors to have not actually checked, and it’s embarrassing for everyone downstream to have not checked either.
Also if you check the article, although an LLM did find issues, it tended to underestimate issues that professional software engineers found.
Imo the data centers is where things start to get scary. Anthropic and OpenAI aren’t themselves indispensable, but when all of the industries downstream of them taking on more and more debt to supply projected usage in the future I could see the USG forced to bail someone out, if only to shore up their creditors.
Snow might have some effect on the height a mountain, but most people believe geological activity is more important than weather.
The relevant question isn’t whether something or another might have “some effect” it is how to reduce the main factors which we already know damage men’s sexual health. And spending time on the long tail of factors which may or may not be relevant is sucking all of the oxygen out of the room for addressing the factors we already know are most of the problem.
If you are a podcast host that gives hot takes on news headlines, which one are you going to choose?
Option one: obesity and weight problems. Statistically 77% of your audience is either overweight or obese because 77% of Americans are either overweight or obese.
Option two: feminism, microplastics, anime, or literally any other thing than option one.
Unsurprisingly, the authors didn’t name “women’s rights” or any other feminism-adjacent culture war issues as a cause of declining testosterone. They did name obesity and diabetes.
In other words, if you’re looking for a boogeyman, blame sedentary lifestyles and ultra processed foods.
Before the reactions to the headline get too out of hand, the article says the study couldn’t rule out that obesity and diabetes might drive this change. Occam’s Razor leads me to lean on this more than any other exotic explanation.
Of course, PFAS and microplastics aren’t great for sperm health, but neither were leaded gasoline and DDT before they were curtailed.
Seems like CEV replaces one problem (“what does humanity want?”) with more problems that are probably even harder to answer.
First of all, calling it “coherent” extrapolated volition presupposes that there is such a thing. It doesn’t actually address the objection above, that there may be no such thing. It’s a bit like saying you solved car safety by presupposing a safe car.
Second, it assumes that such a thing can be effectively measured, and there will be no problems or controversies with the extrapolation process itself. There may be several EVs to choose from, and at that point the framework has nothing to say. Maybe we just pick at random then I suppose.
MS doesn’t need a reason to take your code down from GitHub. If they don’t like what your code does, they can take it down no matter how you word your readme.
> Often the rationalization is due to increased simulation awareness. It’s clear that the model knows that its actions don’t hurt anyone in the real world.
If this is true the entire evaluation is tainted. All of the misbehavior can be written off as justifiable under a simulation.
It might vary between tasks though. A model that’s great at abstract reasoning might be great at writing math proofs but struggle to write software in <insert language>.
Similarly, tasks that are too easy also aren’t ideal either. If a small model makes mistakes and backtracks but eventually cracks it, it will be using a lot more tokens than a bigger model that does it all with minimal mistakes.
Pricing based on tokens always seemed a little weird to me.“Tokens” was and still is an engineering concept. The fundamental unit of transformer encoding and decoding.
But I have a sinking feeling that many AI developers think “tokens” got their name from the same idea as “virtual tokens in a casino” which is more related to product pricing and business.
The reason is, customer might care a lot more about the prediction than you do, or anyone else does.
For example, your customer might really care a lot about some niche prediction like the number of car break-ins in Walmart parking lots. In practice you won’t have sufficient liquidity in a prediction market to actually profit off of that prediction. But a security company might really want to know the answer to it.