foob
Joined 15 years ago
Karma
4,793
Submissions
389
Member since
Feb 14
Recent submissions
Comment · 1 month ago
A much deeper analysis on the falling price per task was published yesterday by Epoch AI [1]. It's a real statistical analysis and comes to more defensible and grounded conclusions. The headline takeaway is:
The cost of a given level of performance often falls fastest right after that level is first achieved, that is, when it is state of the art (SOTA). We see this pattern on three of our five main benchmarks of AI capability. Averaging across all five, cost falls 66% per quarter (75× per year) for performance that has just debuted as SOTA. Two years later, prices fall half as fast, at 32% per quarter (4.7× per year).
but the analysis itself has more nuance and is a quite interesting read.
[1] https://epoch.ai/publications/the-plunging-price-of-thought
Comment · 7 months ago
Amusingly, they deprecated it with a message of "Unpublished" instead of actually unpublishing it [1]. When you use npm unpublish it removes the package version from the registry, when you use npm deprecate it leaves it there and simply marks the package as deprecated with your message. I have to imagine the point was to make it harder for people to download the source map, so to deprecate it with this message gives off a bit of claude, unpublish the latest version of this package for me vibe.
[1] - https://www.npmjs.com/package/@anthropic-ai/claude-code/v/2....
Comment · 1 year ago
That's a fascinating paper, but you're editorializing it a bit. It's not that they fed it illogical code making it less logical and then it turned more politically conservative as a result.
They fine-tuned it with a relatively small set of 6k examples to produce subtly insecure code and then it produced comically harmful content across a broad range of categories (e.g. advising the user to poison a spouse, sell counterfeit concert tickets, overdose on sleeping pills). The model was also able to introspect that it was doing this. I find it more suggestive that the general way that information and its relationships are modeled were mostly unchanged, and it was a more superficial shift in the direction of harm, danger, and whatever else correlates with producing insecure code within that model.
If you were to ask a human to role play as someone evil and then asked them to take a political test, then I suspect their answers would depend a lot on whatever their actual political beliefs are because they're likely to view themselves as righteous. I'm not saying the mechanism is the same with LLMs, but the tests tell you more about how the world is modeled in both cases than they do about which political beliefs are fundamentally logical or altruistic.
Comment · 1 year ago
I've been pretty happy with the python package hns for this [1]. You can run it from the terminal with uvx hns and it will listen until you press enter and then copy the transcription to the clipboard. It's a simple tool that does one thing well and integrates smoothly with a CLI-based workflow.
Comment · 1 year ago
your data can be used UNLESS you opt out
It's even more nuanced than that.
Google recently testified in court that they still train on user data after users opt out from training [1]. The loophole is that the opt-out only applies to one organization within Google, but other organizations are still free to train on the data. They may or may not have cleaned up their act given that they're under active investigation, but their recent actions haven't exactly earned them the benefit of the doubt on this topic.
[1] https://www.business-standard.com/technology/tech-news/googl...
Comment · 1 year ago
I signed up to give it a try, but when I click "Go to Levels" it takes me to https://vimgolf.ai/levels and I get a 404 error.
Comment · 2 years ago
It's also important to take any corporation's explanation for increasing their own margins with an extremely large grain of salt. I'm not doubting in the slightest that consumers had some confusion around the fractions, but all it would take for the company to revert their campaign is for the increase in sales to insufficiently offset the increase in their own costs. Blaming it on consumer stupidity afterwards washes their hands of any responsibility for backpedaling, and makes for a memorable and repeatable story that increases brand recognition while simultaneously painting them as heroically trying to offer more value for the same cost.
Comment · 2 years ago
That's quite the broad brush you're using to paint the Left as accelerationists who favor Trump over Biden.
Comment · 2 years ago
I mostly agree with this in principle, but an important point is that, when you squint, the technology behind blocking ad blockers starts looking very similar to the technology behind blocking web scrapers. If you're capable of programmatically scraping content without a human user viewing ads, then you're capable of displaying the content to a user without the ads. So any solution for preventing ad blocking implies that the content can't be scraped programmatically.
I know that web scrapers carry some negative connotations, but keep in mind that search engines like Google couldn't possibly exist without web scraping. A world where you can't block ads or scrape content for indexing is a world where only a few preordained companies have the ability to build search engines. Proposals like Web Environment Integrity (WEI) accomplish two goals for Google: they make ad blocking more difficult, and they kick down the ladder to prevent new innovative search engines from emerging. There are already many websites which only allow-list Google's IPs for indexing, and I think we should be very hesitant about anything that could further entrench their monopoly on search even if we support content creators being compensated through ads.
Comment · 2 years ago
My pet theory is that the only reason we didn't see a copyright extension act this year, like we've seen every time Steamboat Willy was about to enter the public domain in the past, is that Disney ended up on one side of the culture wars with what went on in Florida. It's far more difficult to get bipartisan support for robbing the public of what should be in the public domain once you've involved yourself in partisan politics.
Comment · 2 years ago
Does it always mean that in Australia, or does it depend on how you say it? I think both usages are actually pretty common in the US, but you tell the difference based on the inflection. If you emphasize "too" and go down in pitch on "bad" then it means you're actually not that good at something, if you emphasize "bad" and go up in pitch then it means you're actually good at it. We also have alternative constructions, like "not too shabby," which carry the positive connotation regardless of how they're pronounced.
Comment · 2 years ago
What about the less convoluted scheme of "I forgot it?"
The "I do not recall" answer in high profile trials is so common that it's essentially become a meme. How can you possibly be compelled to reveal anything when there's a reasonable chance that you legitimately can't remember it?
Comment · 2 years ago
I’ve never wished a man dead, but I have read some obituaries with great pleasure.
- Clarence Darrow
Not the only war criminal to ever win a Nobel Peace Prize, but he's surely in the running for the one with the most blood on his hands.
my public key: https://keybase.io/sangaline my proof: https://keybase.io/sangaline/sigs/uKMqGQ_auE6z2qTdrkpZihnoA-DcJ4BeHb35KLA1KZI