dontreact

This group of people may have been the first to mention the words “AI” prominently in academia but is this flag planting or are they truly foundational to the work with the same name today? If none of these people had done anything, would we really be far behind?

My sense is that modern AI has more to owe Fukushima’s neocognitron, Hubel and Wiesel, and the connectionists than any intellectual descendant of the work mentioned here.

Why would the accuracy decrease with lower temperature? Setting temperature to 0 just means at each step the model will emit the token with the highest likelihood.

I'm left curious what the 20 pound rodent was. Capybara?

"In Venezuela, capybara meat was so popular that sometime between the 16th and 18th centuries, clergymen petitioned to continue eating it during Lent. Although biology says otherwise, the Catholic Church agreed. They declared that the capybara was a “fish” and therefore acceptable to eat during Lent."

Found this after a bit of googling

I’m a diehard YIMBY, but this isn’t really “not in my backyard” it’s more like: don’t prey on my (or my neighbor’s) emotionally and financially vulnerable state in order to buy my backyard (and also the rest of my house).

So not NIMBY but

DPOMEAFVSIOTBMBY

That won't do it, also need to close other stuf

{"this": "is valid json so farrrrrrrrrrrrrr

But yeah the general idea makes sense. Once you hit a timeout, change the mask to things that will close existing open things in a valid manner (}, ), ], ")

You would also need to keep generating until the whole string is valid. And what if it gets caught in a loop?

Not sure how this can really guarantee 100%

Uber was founded in 2009, and pretty broadly available by 2014 Grubhub was founded in 2004, I was ordering from there all through college 2007-2011 Google actually put out hangouts in 2013, and Skype could do group calls in 2010

Not trying to downplay what happened after this: it was incremental change that made the products scale and be more usable to more people. I would say the things you describe as revolutions were actually evolutions of earlier products.

But as mentioned above, isn't doing so against Google's own self interest? It seems like the project is explicitly stating their goal isn't to allow for websites to do this, and they are implementing it in a manner consistent with that.

One thing about your comment above: Hulu can't start implementing attestation until Google turns the knob to 0 because they can't start randomly dropping 5% of Chrome users. So in your comment above it should be "and" not "or". If I understand correctly Hulu cannot act unilaterally with the currently planned implementation of this.

If let's say they did turn the knob for Chrome, wouldn't it take a while for websites to start implementing this? For me not knowing as much about this it feels like this is a step in an ambiguous direction which could be good or bad still. But since it's Google everyone is thinking ahead in the causal chain. Can you help me understand why this is such a big and clearly bad step against the open web? Thank you!

Is it possible for them to implement this API in such a way that it will fail 5% of the time or so, making it impossible for websites to deny individuals based on failing attestation?

https://github.com/RupertBenWiser/Web-Environment-Integrity/...

You are moving the goalposts here. The original claim I am responding to is "A tool that gives incorrect and inconsistent results shouldn’t have any part of a decision making process."

I agree that there are places where we shouldn't put AI and that checking whether something is an LLM or not is one of them. However I think the sentence above takes it way too far and breast cancer screening is a pretty clear example of somewhere we should accept AI even if it can sometimes make mistakes.

It's not a strawman. There are many fundamentally unpredictable things where we can't make the benchmark be 100% accuracy.

To make it more concrete on work I am very familiar with: breast cancer screening. If you had a model that outperformed human radiologists at predicting whether there is pathology confirmed cancer within 1 year, but the accuracy was not 100%, would you want to use that model or not?

If you were trying to predict the direction a stock will move (up or down) and it was right 99.9% of the time, would you use it or not?

I personally don’t care that much the collision rate. I care about the fatality or injury rate (deaths or injuries per miles driven)

I don’t mind picking an Uber driver if it reduces my chances of getting hurt, even if we are slightly more likely to experience a fender bender.

Humans cause fatalities 1 per 100 million miles.

There is no way to truly evaluate safety being an order of magnitude better until we have 1 billion miles driven.

You can crash less often (reduce fender benders) but kill more people (have rare but egregious errors) so I am not interested in seeing statistics about incidents or collisions.

You could avoid living in a place likely to flood during a 100 year flood

And by garbage I mean literally data/illegible/etc. that would ruin the pretraining of the model.

One of the big values of reddit is that it serves as an easy index for non-garbage websites on the internet. This is exactly how it was used in the T5 paper: they threw away all the websites from their crawl that were not linked to from Reddit.

As far back as 2015, it’s been common to take neural net inference and get a C++ version of it. That’s what this is.

It didn’t make python obsolete then and it won’t now.

Training (orchestration, architecture definition etc.) and data munging (scraping cleaning analyzing etc.) are much easier with python than C++, and so there is no chance that C++ takes over as the lingua Franca of machine learning until those activities are rare

2: The more people buy and hold Bitcoin, the more valuable it becomes.

I asked ChatGPT about the definition of common enterprise:

The 7th and 2nd Circuits use a "horizontal commonality" test, where a common enterprise is found if the profits of the investors are interdependent - i.e., if the success of the enterprise is tied to the success of other investors. Essentially, the investors' funds are pooled and they share in the profits and losses.

3: Overwhelming majority of people who buy gold buy it as a hedge against inflation. In the past 5 years everyone I personally know who has bought cryptocurrency was expecting it to increase massively in value (expecting profit). 4: There have been a huge amount of cryptocurrency pump and dump stories where the promoters have benefitted immensely. In those cases it seems pretty clear that those are securities. In fact pretty much anything that isn't a top-2 coin at this point is subject to this sort of behavior from what I've seen.

Putting this here in case it's useful. Please correct if it's wrong. What is the Howey test for whether something is a security?

""" The Howey Test asks whether a transaction constitutes an "investment contract," which is a type of security. If it is an investment contract, it must satisfy four criteria:

It is an investment of money. The investment is in a common enterprise. There is an expectation of profits from the investment. Any profit comes from the efforts of a promoter or third party.

If the answer to all these questions is "yes," the transaction is likely an investment contract and, therefore, a security. However, please note that this is a simplified explanation. The actual determination can be complex and might require a detailed analysis by legal professionals. """

To me it DEFINITELY sounds like crypto is a security, but it seems like it mostly hinges on whether people expect to make a profit by investing. Which some people do and some people don't. But most people do right?

My guess is that "making cheaper and faster GPT4" was leading to the rumors that GPT4 was nerfed

An LLM only has to download text once. You could easily coordinate a scrape across various IPs and accounts over the course of a couple of months and it would be really hard to detect. Especially if you target alternate GUI api based versions of the site.

Why wouldn’t people then just scrape using the free API?

Lol he said he would put them in his children. But he doesn’t have one?

you're going to pay "replacement cost"at the same rate or higher for someone else

How do you know this is true? This is an assertion you are making, but the evidence is to the contrary given significantly lower compensation for Microsoft or Apple's CEO.

What useful things are people able to do with the smaller more inaccurate models? I have a hard time understanding why I would build on top of this, rather than just the openaI API, since the performance is so much better.

Is this what is happening in Xinjiang with the degree of ai surveillance?

For context: even if you were able to do this regularly, it still likely wouldn't be enough to make a living as an artist.

This has been the case for a long time but my sense is that pornography addiction is a growing concern because of the accessibility and customizability increases that technology has afforded. AI accelerates that.

This is true only if all the outcomes are the same. However “all the outcomes” includes long term outcomes of swapping the real for the fake, and historically there have been unanticipated consequences aka outcomes when doing so.

In the case of AI pornography, the scary outcome to me is that more people become addicted to internet pornography and the amount of healthy connections in society decreases.