I hear what you're saying man, but honestly it's a sensor. Sensors can fail all the time even without deliberate tampering. It doesn't seem to really make sense to have a single one in a single location. Besides for clarity, my question was more why was the market referenced on a single sensor rather than on multiple sensors?
lol, the OP is such a classic HN take. "Why doesn't society simply absorb a negative externality created by gambling and add high cost redundancy layers that are otherwise useless" - Not every problem is technical.
Because that wasn't what caused the problem? The reason it didn't work was some asshole intentionally tampered with the equipment, not because of where it was.
I am a bit surprised by that for a scientific application where you want high accuracy reading. Temperature sensors has a error margin, and I think they can also drift a bit.
In a bit of expensive equipment I own it happens to have 4 temperature and pressure sensors, located in two different locations on the unit. All of them generally disagree with each other on the exact temperature by around 0.1 C, which is fine for my use of it.
Why didn't you put two locks on your door? Clearly you deserved to get burgled if you only used one.
Building Byzantine fault tolerance into absolutely everything is expensive, and makes everything we do and buy more expensive. It would be better and cheaper to rely on social trust, if that's possible-- and it was possible before these gambling sites. Prediction markets are burning social trust as fuel to make profit, they should be heavily taxed, the way polluters are taxed, as destroyers of the commons.
Prediction markets and all the market manipulation are the symptom, not the cause. Our society used to have real consequences for breaching public trust, but with our mere decades old "money is speech" legal system, there have not been any consequences for moneyed interest in quite a while. And as long as there are no consequences, they keep trying more and more egregious violations of public trust to establish where the new red line (if any) actually are.
Money is speech - but speech isn't speech. What's the latest thing a US citizen said in the US that got them arrested? They said in a private WhatsApp group that Benjamin Netanyahu should come and bomb their school to get them out of an exam. Benjamin Netanyahu was not in the group chat, but they got arrested anyway.
>Why didn't you put two locks on your door? Clearly you deserved to get burgled if you only used one.
I can reasonably test that the bolt is in a locked position every time I close the door and turn handle on the lock. But a single remote sensor could have malfunctioned or simply be out of calibration a few degrees.
If the actual temperature at the airport is important to any set of users enough so that the difference between it being 18 or 22 deg C is relevant, one should expect that there be at least 3 sensors (much like clocks) and assuming variances between the 3 sensors are within tolerance an average of the 3 temperatures is taken.
> If the actual temperature at the airport is important to any set of users enough so that the difference between it being 18 or 22 deg C is relevant, one should expect that there be at least 3 sensors
Three sensors doesn't solve the problem. Manipulating becomes marginally more difficult with three sensors, but it's still very possible, and with enough monetary incentive it's still even likely that it happens again.
So why not 5 sensors? How about 10?
And what about more consequential issues like the toppling of governments or military blockades where true redundancy is impossible and people are actually harmed?
Is there not a point where you start to blame the incentives that are being chased?
This. I see much cheap naysaying without referenece to the vuln hashes. If it is smoke and mirrors, then the naysayers should loudly shout down the specific hashes and when they get revealed, or don't, then they will have done a great service to dissuading fake claims to world changing tech.
I do, inevitable, but ime the prompts force certain behaviors at similar strength (instruction following). So it's one thing that the model is biased towards any particular direction by its latent space, it's another that it is biased by an immodifiable prompt which can only be contradicted for the benefit of the lcd at the expense of the more involved operator.
Sure, but now we have to remodel whatever bias we want for our use case with every new release because the system prompt changes, whereas the underlying data does not.
Underlying data changes all the time, as do training methodologies / preferences.
You do realize that these LLMs are trained with a metric ton of synthetic examples? You describe the kind of examples / behavior you want, let it generate thousands of examples of this behavior (positive and negative), and you feed that to the training process.
So changing this type of data is cheap to change, and often not even stored (one LLM is generating examples while the other is training in real-time).
Well, I'd say it's a reasonable expectation for the model to behave similarly across releases. Am I wrong to assume that?
I imagine the system prompt can correct some training artifacts and drive abnormal behavior to the mean in the dimensions that Anthropic deems fit. So it's either that they are responding to their brittle training process, or that they chose this direction deliberately for a different reason.
Yeah for anyone seriously using these models I highly reccomend reading the Mythos system card, esp the sections on analyzing it's internal non verbalized states. Save a lot of head wall banging.
reply