Regarding tools, I use Python. I wrote the backtesting software many, many, many years ago during my Master's degree, and I've been refining it ever since.
It's an event-driven engine (they are slower than vector-based engines, but they are easier to write strategies for, understand, and debug) with all the bells and whistles, similar to the late Zipline. In fact, I tried most of the Python backtest engines that exist, and that's why I prefer to use what I built over the years: I have 100% understanding of what’s happening and 100% control.
I’m thinking about open-sourcing it… anyway, the logic is not that complicated.
Ultimately advertisers want a return on their spend and it's a closely watched metric for marketers. Them continuing to spend is indicative that there's growing value in ads, ie bots/ai agents cannot be the reason for their growth.
You could argue fb and particularly twitter are incentivized to include it in their DAU counts but market cares more for revenue for large companies.
SMEs spend a lot on advertising, in aggregate. These smaller businesses are more sensitive to lowered conversation rates and will bail early - Meta took a huge hit in earnings in the aftermath of Apple's privacy changes due to poor conversions. Advertisers didn't keep pumping money in - they stopped campaigns.
So do Android and iOS. They're still the primary indicators of the strength of the smartphone market. You probably can't imagine a world without advertisements or smartphones, but that's the point from the post you're responding to that you're missing. If the advertising industry looks healthy, it helps everyone by keeping investors satisfied and share prices up, even if it's a ruse. It is possible for smartphones and the concept of advertising to cease existing some day. There's a spectrum between the current reality and absolute zero, and staying as far away from zero is the goal, even once we enter an inescapable decline.
Arguably true, but that's a rather different point than "there is no alternative". There are arguments to be made about market "health" and the relative sizes of the largest players, but what the upthread comment was doing was resorting to "anti trust" shorthand that clearly doesn't apply. There may be problems with this market, but lack of competition isn't one of them.
The point is not that he shouldn’t be allowed to unilaterally ban social media accounts.
The point is that he shouldn’t be allowed to do that in secrecy, without providing any public justification, and not respecting the right of the accused to defend themselves.
People are getting silenced without knowing why they are getting silenced, and without proper due process/right to respond.
Of course, it's ok to persecute my opponents, because they are the intolerant bad people, but it's not ok to prosecute me, because I am extremely tolerant and only prosecute my opponents, who, as we already established, are bad people and are ok to prosecute.
Popper's proposed remedy isn't a carte blanche for institutional censorship. Some excerpts from Popper's book that those who cite the paradox of tolerance typically ignore:
"I do not imply, for instance, that we should always suppress the utterance of intolerant philosophies; as long as we can counter them by rational argument and keep them in check by public opinion, suppression would certainly be most unwise."
"All these paradoxes can easily be avoided if we frame our political demands in the way suggested in section ii of this chapter, or perhaps in some such manner as this. We demand a government that rules according to the principles of equalitarianism and protectionism; that tolerates all who are prepared to reciprocate, i.e. who are tolerant; that is controlled by, and accountable to, the public. And we may add that some form of majority vote, together with institutions for keeping the public well informed, is the best, though not infallible, means of controlling such a government. (No infallible means exist.)"
Notes: By equalitarianism he meant the classic liberal notion of equal rights to everyone. By protectionism he meant that the state should ensure people's rights (protect people).
What the Brazilian supreme court is doing is the opposite, it's unaccountable and widespread censorship by non-elected judges.
And what if both side in this dispute "bolsonaristas" and "petistas" are intolerant?
Personally I saw worst cases of intolerance from "petistas" than from "bolsonaristas". Like accusing me of being a "bolsonarista" just because I didn't agree with his extremist political or was wearing and green & yellow shirt. Like leftists openly inciting violence against those they disagree with and receiving praises from their peers.
The worst I received from "bolsonaristas" was being called a "communist" (which I'm not and deem offensive). And ironically the groups they are mostly intolerant towards (besides leftists) are criminals, corrupt politicians, pedophiles (those that are intolerant). Were "petistas" are typically intolerant towards businessmen, policemen, Christians (specially from evangelic congregations), the wealthy and famous (but only those that don't share their views lol). But despite all the majority of "bolsonaristas" and "petistas" aren't bad, just normal people brainwashed with vicious ideologies.
You know, during the entire 90s when the people that wrote it were around, they used to say that the Article 5 said this. But nowadays it's consensus over all the influential judges that it doesn't and only registered journalists can say things, and only congresspeople can have their opinions. So, who knows...
Anyway, it still says that decisions should be informed to the punished party and people should be able to defend themselves in a court.
The issue with this approach is that for all but the most simple apps it is not possible to deduce the runtime element information needed to write traditional UI tests given just the source code. This can only be done reliably at runtime which is what we do. We run your app and iteratively build UI tests that can be reused later.
> We observed that all the VLMs tend to be confident while being wrong. Interestingly, we observed that even when the entropy was high, models tried to provide a nonsensical rational, instead of acknowledging their inability to perform the task
It looks like all current models suffer from an incurable case of Dunning–Kruger effect cognitive bias.
But they can also only do negation through exhaustion, known unknowns, future unknowns, etc...
That is the pain of the Entscheidungsproblem.
Even in Presburger arithmetic, Natural numbers will addition and equality, which is decidable, still has a double factorial time complexity to prove. That is worse than factorial time for those who've not dealt with it.
Add in multiplication then you are undecidable.
Even if you decided to use the dag like structure of transformers, causality is very very hard.
LLMs only have cheap access to their model probables which aren't ground truth.
So while asking for a pizza recipe could be called out as a potential joke if add a topping that wasn't in its training set, through exhaustion, It can't know when it is wrong in the general case.
That was an intentional choice with statistical learning and why it was called PAC (probably approximately correct) learning.
That was actually a cause of a great rift with the Symbolic camp in the past.
PAC learning is practically computable in far more cases and even the people who work in automated theorem proving don't try to prove no-instances in the general case.
There are lots of useful things we can do in BPP (bounded probabilistically polynomial time) and with random walks.
But unless there are major advancements in math and logic, transformers will have limits.
The parameters don't store any information about what inputs were seen in the training data (vs being interpolated) or how accurate the predictions were for those specific inputs.
And even if they did, the training data was usually gathered voraciously, without much preference for quality reasoning.
I don't know for sure, but here's a plausible mechanism for how:
Multiple sub-networks detect the same pattern in different ways, and confidence is the percent of those sub-networks that fire for a particular instance.
There's a ton of overlap and redundancy with so many weights, so there are lots of ways this could work
That’s good. Also maybe an architecture that runs the query through multiple times and then evaluates similarity of responses, then selects (or creates) the most-generated one, along with a confidence level of how many of the individual responses were aligned.
Actually you can get a very good proxy by looking at the probability distrobution of the "answer" tokens. The key here is you have to be able to identify the "answer" tokens.
Phind gives me ChatGPT answers with relatively authoritative references to works on the web that (usually!) support the answer. Could it have a post-filter to fact check against the references?
I guess that is a slight variation of the sibling (@habitue's) answer; both are checks against external material.
I wonder if best resources could be catalogued as the corpus is processed, giving a document vector space to select resources for such 'sense' checking.
IIRC confidence in video is related to predicting what happens next vs what actually happens. If the two seem to correlate to the model it would give it a higher confidence ranking, which would then be used further for self-reinforced learning.
> We find inspiration not from the size of a market, but from the importance of the work. Because the importance of the work is the early indicator of a future market.
> You are probably one ArXiv paper away from figuring this thing out.
> I used to be a dishwasher. I’ve cleaned a lot of toilets. I’ve cleaned more toilets than all of you combined. And some of them you can’t unsee. That’s life.
https://data.nasdaq.com/databases/SFA
It's a great survivorship-bias-free dataset.
Regarding tools, I use Python. I wrote the backtesting software many, many, many years ago during my Master's degree, and I've been refining it ever since.
It's an event-driven engine (they are slower than vector-based engines, but they are easier to write strategies for, understand, and debug) with all the bells and whistles, similar to the late Zipline. In fact, I tried most of the Python backtest engines that exist, and that's why I prefer to use what I built over the years: I have 100% understanding of what’s happening and 100% control.
I’m thinking about open-sourcing it… anyway, the logic is not that complicated.