blueblimp

7 ms·

31.
by blueblimp 3y ago
The proposed replacement definitely makes more sense (and I've always found the absence of a "failed query" to be puzzling in standard attention), but, in deep learning, things that make more sense don't always actually
32.
by blueblimp 3y ago
For those who aren't aware, this blog is infamous for sloppy work. https://www.lesswrong.com/posts/7iAABhWpcGeP5e6SB/it-s-proba...
33.
by blueblimp 3y ago
What's problematic is that there are big models that adopt truly open source licenses, such as MPT-30b and Falcon-40b. As grateful as I am for having access to the Llama2 weights, it feels unfair that it gets credit for being "ope
34.
by blueblimp 3y ago
This is a good way to look at it. The deployment of the LLM technology we currently have is one thing. The impact of AGI, when it's created in the future, is a different thing. And people are often not clear on which of the two they&#x
35.
by blueblimp 3y ago
I found the visual justification for partitioning a cube into three pyramids to be a bit confusing. For me, an easier way to think of it is that there are three coordinates (x,y,z) and each pyramid represents a region where a particular coo
36.
by blueblimp 3y ago
Cutting edge models are quite expensive to train, so it'd be quite possible to restrict them (unfortunately).
37.
by blueblimp 3y ago
From what I understood, the problem is less the change itself but more the short notice. 30 days isn't much to redesign your app and monetization, especially if you had many users on year-long subscription plans.
38.
by blueblimp 3y ago
This brought to mind another blog post I liked on the topic of finishing, by Derek Yu (of Spelunky fame): https://makegames.tumblr.com/post/1136623767/finishing-a-gam... .
39.
by blueblimp 3y ago
I liked the trick of having them make excuses instead of revealing they're a bot. Clever.
40.
by blueblimp 3y ago
AI isn't nuclear weaponry. It's bits, not atoms. It's more like encryption.
41.
by blueblimp 3y ago
Personalization, customization, etc.: by aligning AI systems to many users, we benefit from the already-existing diversity of values among different people. This could be achieved via open source or proprietary means; the important thing is
42.
by blueblimp 3y ago
There is a way, in my opinion: distribute AI widely and give it a diversity of values, so that any one AI attempting takeover (or being misused) is opposed by the others. This is best achieved by having both open source and a competitive ma
43.
by blueblimp 3y ago
This is way better than the open letter. It's much clearer and much more concise, and, maybe most importantly, it simply raises awareness rather than advocating for any particular solution. The goal appears to have been to make a state
44.
by blueblimp 3y ago
Some time ago, I listened to Switched-On Bach out of historical interest, and I was surprised to find that it's largely gimmick-free. It's a (good) classical music performance that just happens to be performed using synthesizers.
45.
by blueblimp 3y ago
I wonder how much of that comes from gacha games filling a similar role.
46.
by blueblimp 3y ago
I hate ads and use an ad blocker, but I also pay for ad removal on any site I use regularly that offers the option (which includes YouTube). This seems like a reasonable balance between having the internet be usable while also supporting th
47.
by blueblimp 3y ago
It seems to me that right now is an especially awkward time to launch a new programming language competing with Python, because LLMs are great at writing Python, and presumably can't write your new language at all (since there's n
48.
by blueblimp 3y ago
Interesting resource. I had been wondering whether anyone had tried to compile such a list.
49.
by blueblimp 3y ago
Schulman's talk is great. I had been thinking about this problem, and he covers almost everything I thought of, with great clarity. One thing he didn't mention though is that there's potentially a bit of a trick to get the fi
50.
by blueblimp 3y ago
> If it understands language at all, an LLM only does so in a statistical, rather than a grammatical, way. It is much more like an abacus than it is like a mind. This analogy puzzled me. An abacus does not strike me as statistical in the
51.
by blueblimp 3y ago
It's interesting to hear your experience as a pedestrian, because it makes sense that cautious self-driving cars would cause fewer problems for pedestrians than for traffic. After all, a pedestrian has no problem dealing with a station
52.
by blueblimp 4y ago
The recommendation to support conversational interactions is very good. After all, if you consult with a human expert, one of the main things they'll be doing is conversing with you to figure out your requirements. I find it a bit frus
53.
by blueblimp 4y ago
I've always thought of them a bit like improv, since they tend to follow the "yes, and..." rule, by happily continuing whatever direction you want to go. Now that the base models have been fine-tuned to avoid some topics, tha
54.
by blueblimp 4y ago
AI has a privacy problem right now where the most capable models are available only on cloud services that censor and access your input and output data. The gap between the state-of-the-art and private alternatives is especially large in te
55.
by blueblimp 4y ago
My grandfather wrote an autobiography (for family) and I found it interesting to read.
56.
by blueblimp 4y ago
The key quote: > [Wozniak] was overjoyed when he learned that the skill that put so much effort into suddenly became massively easier. He wasn't worried about not being able to earn an above-average salary from this anymore, he was
57.
by blueblimp 4y ago
Yes, it goes beyond even just extensive usage restrictions and restricts _who_ can use it. https://jamesturk.github.io/scrapeghost/LICENSE/#3 It seems, for example, that (by 3.1.12) if you are a person who is invo
58.
by blueblimp 4y ago
SEO garbage is such a problem these days that, if I were Google, more than using AI as a new frontend for search, I'd be trying to find a way to use it to defeat SEO.
59.
by blueblimp 4y ago
It seems still unclear how much quality loss there is compared to the best models. What's really needed is systematic evaluation of the output quality, but that's tricky and relatively expensive (compared to automated benchmarks),
60.
by blueblimp 4y ago
NovelAI's image generation uses a fine-tune of Stable Diffusion.

More ›