a_wild_dandan

7 ms·

61.
by a_wild_dandan 2y ago
So...nothing will change?
62.
by a_wild_dandan 2y ago
This conclusion is the lump of labor fallacy. It's not that simple.
63.
by a_wild_dandan 2y ago
Let's just define AI as "whatever computers still can't do." That'll show those dumb statistical parrots!
64.
by a_wild_dandan 2y ago
An additional citation substantiating that 10 MB/s figure: the 3rd sentence of TFA.
65.
by a_wild_dandan 2y ago
Expected utility, just like anything else. It's why our ancestors ate willow tree bark (aspirin), for instance.
66.
by a_wild_dandan 2y ago
Animation can look real though, e.g sci-fi vfx. But maybe you’re concerned about how prolific it may be? I could see that. Personally I think it’ll be fine. It’s just that disruptive tools create uncertainty. Or maybe I’m overcompensating
67.
by a_wild_dandan 2y ago
Reasons that I can't take this benchmark seriously: 1. Existing brute force algorithms solve 40% of this "reasoning" and "generalization" test. 2. AGI must evidently fit on a single 16GB, decade-old GPU? 3. If ARC f
68.
by a_wild_dandan 2y ago
AI researchers get huge checks working anywhere. So how does your company distinguish its appeal? Play to a researcher's desire for recognition in their field. Lean on open source. (I'm just paraphrasing Zuckerberg.)
69.
by a_wild_dandan 2y ago
You can make highly accurate predictions of what contrarians will say by assuming that they define AI as "whatever computers can't do yet." LLMs aren't truly intelligent. [No True Scotsman fallacy...] They don't
70.
by a_wild_dandan 2y ago
All of machine learning is AI. It's a subset, by definition.
71.
by a_wild_dandan 2y ago
Important testing excerpts: - "...for the closed (OpenAI) models I tried generating up to 10 times and if it still couldn’t come up with a legal move, I just chose one randomly." - "I ran all the open models (anything not fro
72.
by a_wild_dandan 2y ago
The model meets/beats Llama despite having an order-of-magnitude fewer active parameters (52B vs 405B). Absolutely bonkers. AI is moving so fast with these breakthroughs -- synthetic data, distillation, alt. architectures (e.g. MoE&#x2
73.
by a_wild_dandan 2y ago
This could be revolutionary. The PPL/compute graphs are damning. If the Transformer is a function, then the TokenFormer feels like a higher-order function. Perhaps this approach is a natural direction for producing System Two reasoning
74.
by a_wild_dandan 2y ago
What were the major differences?
75.
by a_wild_dandan 2y ago
I suppose Valve, who trained a neural network to detect/ban cheaters exhibiting unnatural behavior.
76.
by a_wild_dandan 2y ago
The "statistical parrot" parrots have been demonstrably wrong for years (see e.g. LeCun et al[1]). It's just harder to ignore reality with hundreds of millions of people now using incredible new AI tools. We're approachi
77.
by a_wild_dandan 2y ago
Evaluating cross-lingual transfer learning approaches in multilingual conversational agent models [1] Cross-lingual transfer learning for multilingual voice agents [2] Large Language Models Are Cross-Lingual Knowledge-Free Reasoners [3]
78.
by a_wild_dandan 2y ago
Yes, because transfer learning works. A specialized model for X will be subsumed by a general model for X/Y/Z as it becomes better at Y/Z. This is why models which learn other languages become better at English. Custom models
79.
by a_wild_dandan 2y ago
Cursor is too! Mixing and matching specialized & flagship models is the way forward.
80.
by a_wild_dandan 2y ago
The speed with which AI models are improving blows my mind. Humans quickly normalize technological progress, but it's staggering to reflect on our progress over just these two years .
81.
by a_wild_dandan 2y ago
Are there any other companies which surprised you with their ubiquity?
82.
by a_wild_dandan 2y ago
I'm so jazzed to comfortably run Llama 3 405b on my Mac with less quantization.
83.
by a_wild_dandan 2y ago
I’m continually surprised at JavaScript’s speed. Seeing JS sometimes nipping at the heels of C/rust/etc in performance benchmarks blows me away. V8 is such an incredible piece of engineering. In my work, it’s hard to justify using
84.
by a_wild_dandan 2y ago
Running a (405B) flagship model locally on a Mac Studio? Otherwise, no. Most of these products are fairly similar.
85.
by a_wild_dandan 2y ago
"Commoditize your complement."
86.
by a_wild_dandan 2y ago
Lina Khan's response to this argument. [1] [1] https://www.youtube.com/live/L_QaZk5iJOA?si=ZkxBe1CHgagmcBcW...
87.
by a_wild_dandan 2y ago
Could Apple have Google-like restrictions in the future? Or are we kinda fucked because Apple already "won"?
88.
by a_wild_dandan 2y ago
More citations from their blog[1] too. [1] https://www.liquid.ai/blog/liquid-neural-networks-research
89.
by a_wild_dandan 2y ago
"The Llama jumped over the ______!" (Fence? River? Wall? Synagogue?) With 1-hot encoding, the answer is "wall", with 100% probability. Oh, you gave plausibility to "fence" too? WRONG! ENJOY MORE PENALTY, SCRUB!
90.
by a_wild_dandan 2y ago
That retcon was bizarre. There are myriad simple pretexts for enslaving humanity. Choosing "let's maintain lukewarm meat bags for energy" is so contrived and absurd.

More ›