gwern

11 ms·

61.
by gwern 5mo ago
I understood his joke just fine. I just disagree, because I think he missed that it was AI-written in a way which is actually already implementing his suggestion of 'lowering' the original high-level human input into a more rigid,
62.
by gwern 5mo ago
This post wasn't written in English, it was written in AIglish. (For god's sake, please tell me you see it at this point and you don't need to punch the opening into Pangram to see '100% AI' to recognize it by now
63.
by gwern 5mo ago
Yes, I definitely take this as a kind of reductio of Grow-Speech, after having defined it rigorously enough to be enforceable, or as chatbots love to say now, 'auditable' ( https://gwern.net/grow-speech ). The LLMs
64.
by gwern 5mo ago
> Improving their probe of the risk-of-failure was enough. The point was it was dangerous in a way they didn't even realize was an issue, for a thin business rationale. Unless you are going to do thorough tests and understand the ri
65.
by gwern 5mo ago
Query strings break unpredictably, and that alone is enough to ban them by third parties, especially for something as minor as referral tracking. Example: The Browser is a well known link aggregation paid periodical. I subscribe, and every
66.
by gwern 5mo ago
> Then it wasn't for you. No big whoop. Ah yes. I remember this in Penny Arcade: https://www.penny-arcade.com/comic/2004/03/24/the-adventures...
67.
by gwern 5mo ago
I didn't enjoy it. It flunks my criteria for good AI images, because there is no there there: https://gwern.net/blog/2025/good-ai-samples It wasted my thoughts as I stared at it, trying to learn about the aut
68.
by gwern 5mo ago
But doesn't that make it bad? It doesn't say anything new. Unlike the software in question, which is personalized, so it's not even symbolically reflecting the topic. It's a sheer waste of pixels and time spent looking a
69.
by gwern 5mo ago
tldr: do a standard img2img workflow where you lay out a skeleton or skeleton or low-res version, and then turn it into the final high-quality photorealistic version, instead of trying to zeroshot it purely from a text prompt.
70.
by gwern 5mo ago
> Clearly if it was able to be leaked it already was being preserved Preserved by whom ? Many leaks are done by old or ex-employees who quietly kept a shall we say 'backup' of their work. More than one 'official' re-
71.
by gwern 5mo ago
Note: this is one of the less popular mirrors. The most popular mirror is the virginia.edu one ( http://www.cs.virginia.edu/~robins/YouAndYourResearch.html ), and the best is (IMO) my heavily annotated version https:&#x
72.
by gwern 5mo ago
Source for it being fictional? The original 1960 source describes it as extremely nonfictional: https://gwern.net/doc/fiction/science-fiction/1960-analog-oc...
73.
by gwern 5mo ago
> 2,000-year-old honey that's still edible? No, it's a lie. I researched it a bunch back in September 2024 (I was curious what the oldest possible edible food was*), and the Smithsonian knows it's BS (because I emailed the
74.
by gwern 6mo ago
Main discussion: https://news.ycombinator.com/item?id=47884971
75.
by gwern 6mo ago
> The problem is that subjective judgements by streaming platforms on where an AI line is drawn in music production is difficult. This may be more of an economic problem. There is a stark difference between a music track with 1% human wo
76.
by gwern 6mo ago
"hyperpigmentation on the eyelids from blue-light exposure" is a super specific query almost definitionally 'bixonimania' which probably brought up the 'bixonimania' poison at the time (the search hits for that
77.
by gwern 6mo ago
> In addition to detecting, tagging and removing AI-generated music from recommendations, Deezer has now stopped storing hi-res versions of AI-tracks Important point for anyone out there thinking about generating a lot of samples. Expect
78.
by gwern 6mo ago
> Two blog articles and two preprints of fake academic articles [0] were able to convince CoPilot, Gemini, ChatGPT and Perplexity AI of the existence of a fake disease, against all majority consensus Wrong. There are no 'majority co
79.
by gwern 6mo ago
Glad OP got the tri-state toggle right!
80.
by gwern 6mo ago
Apparent source: https://www.theinformation.com/newsletters/applied-ai/uber-c...
81.
by gwern 6mo ago
Unfortunately, that is an oversimplification for a highly RLed/chatbot trained LLM like Claude-4.7-opus. It may have started life as a base model (where prompting it with correctly spelled prompts, or text from 'gwern', woul
82.
by gwern 6mo ago
LLM APIs sell on value they deliver to the user, not the sheer number of tokens you can buy per $. The latter is roughly labor-theory-of-value levels of wrong.
83.
by gwern 6mo ago
You didn't say anything about 'episodic' and that's irrelevant to the point even if its long-term memory from training didn't count.
84.
by gwern 6mo ago
It is definitely not the first codebase an extensively RL-trained Claude has ever analyzed. How do you think it got so good?
85.
by gwern 6mo ago
It's a lot easier to look at the whitespace and paragraphs, realize it's a LLM, plug it into Pangram to see that it gets 100% (unsurprisingly), and click to close; than it is to read it with a sucker's good faith and realize
86.
by gwern 6mo ago
But is the cinematography there of any interest? Why would OP include it? They're talking about the hard shots, like the exploding spaceship where they need to find a spot in the desert to shoot dozens of mortars at it, or the crazy
87.
by gwern 6mo ago
The 2% is the camel's nose. They are establishing that they tax the Strait traffic and there is no longer freedom of navigation. Once it is a done deal, the deal will be altered...
88.
by gwern 6mo ago
My point is more that since you can expect taste's commoditization to lag behind for deep fundamental reasons, then taste does serve as a moat. Just perhaps a weaker one than one would naively expect, and where you will have to frant
89.
by gwern 6mo ago
They do improve, but the general creativity and sparkle we see with increasing scale comes mostly from scaling up pretraining/parameter-size, so it's quite slow and expensive compared to the speed (and decreasing cost) people have
90.
by gwern 6mo ago
Similar numbers for other countries: https://gwern.net/banner#they-just-dont-know Wouldn't shock me if it's even lower as people move to walled-gardens like smartphones.

More ›