text
stringlengths
0
2.35k
[2767.54 --> 2768.78] So a follow-up question
[2768.78 --> 2769.28] to that.
[2769.28 --> 2771.12] for code generation,
[2771.94 --> 2772.56] some languages
[2772.56 --> 2773.56] are better than others.
[2773.86 --> 2774.32] For example,
[2774.52 --> 2775.84] Go is performing
[2775.84 --> 2776.76] even better
[2776.76 --> 2777.84] than Python and such
[2777.84 --> 2778.50] just because
[2778.50 --> 2779.48] Go has this
[2779.48 --> 2780.24] built-in linter
[2780.24 --> 2781.56] and there's many things
[2781.56 --> 2782.52] that it's not
[2782.52 --> 2783.08] either or,
[2783.22 --> 2783.60] but you know,
[2783.62 --> 2784.68] it's definitely tabs.
[2784.82 --> 2785.52] It's definitely
[2785.52 --> 2786.62] curly brackets
[2786.62 --> 2787.36] and a new line.
[2787.54 --> 2788.78] So the AI
[2788.78 --> 2790.20] has a more
[2790.20 --> 2791.18] consistent data
[2791.18 --> 2792.92] set to be trained on
[2792.92 --> 2793.58] versus Python
[2793.58 --> 2794.04] that you or
[2794.04 --> 2794.82] many other languages
[2794.82 --> 2795.46] that you can write
[2795.46 --> 2796.60] in whatever way.
[2796.60 --> 2797.74] So it just sees
[2797.74 --> 2798.80] lots of different examples
[2798.80 --> 2799.66] and it might
[2799.66 --> 2800.72] in best case
[2800.72 --> 2801.70] generate inconsistent
[2801.70 --> 2803.00] code of,
[2803.58 --> 2803.76] you know,
[2803.82 --> 2804.44] one file is different
[2804.44 --> 2804.92] from another,
[2805.02 --> 2805.78] but sometimes even
[2805.78 --> 2807.68] just sometimes wrong
[2807.68 --> 2809.28] following two different
[2809.28 --> 2810.52] paradigms in one file.
[2811.48 --> 2812.80] Do you think that
[2812.80 --> 2814.22] for the reverse part of it,
[2814.64 --> 2815.50] will this benefit
[2815.50 --> 2816.36] kind of keep rolling
[2816.36 --> 2816.98] or not
[2816.98 --> 2818.48] for the AI perspective
[2818.48 --> 2818.96] of it?
[2819.38 --> 2820.36] Yes and no.
[2820.42 --> 2821.22] So what you are referring
[2821.22 --> 2821.94] to is the fact
[2821.94 --> 2823.20] that the more strict
[2823.20 --> 2824.12] languages are
[2824.12 --> 2824.76] then the easier
[2824.76 --> 2825.38] for an AI
[2825.38 --> 2827.06] it is to understand
[2827.06 --> 2827.82] what is meant
[2827.82 --> 2828.66] and to generate code
[2828.66 --> 2829.56] and I think it makes sense
[2829.56 --> 2831.36] because Python,
[2831.64 --> 2831.98] for instance,
[2832.10 --> 2832.66] as a language
[2832.66 --> 2833.58] is extremely ambiguous,
[2833.86 --> 2834.04] right?
[2834.36 --> 2835.14] Not as ambiguous
[2835.14 --> 2835.84] as JavaScript
[2835.84 --> 2837.02] or maybe as PHP,
[2837.64 --> 2837.78] you know,
[2837.82 --> 2838.16] which are,
[2838.32 --> 2839.14] as far as I'm concerned,
[2839.46 --> 2840.20] they don't qualify
[2840.20 --> 2841.22] as actual languages
[2841.22 --> 2842.02] because they're not
[2842.02 --> 2842.82] rigorous at all.
[2843.34 --> 2844.16] But when you have
[2844.16 --> 2844.90] stuff like
[2844.90 --> 2845.64] duct typing
[2845.64 --> 2846.48] and a variable
[2846.48 --> 2847.14] that you don't have
[2847.14 --> 2848.60] to type at all,
[2849.20 --> 2850.18] well, of course,
[2850.40 --> 2851.02] for an AI
[2851.02 --> 2852.18] to figure out
[2852.18 --> 2852.90] what is going on
[2852.90 --> 2853.66] is going to require
[2853.66 --> 2854.46] much more effort
[2854.46 --> 2855.18] and it's going to
[2855.18 --> 2856.64] probably not be possible
[2856.64 --> 2857.30] all the time
[2857.30 --> 2857.90] because I think
[2857.90 --> 2858.80] that in many cases
[2858.80 --> 2859.54] a lot of stuff
[2859.54 --> 2860.16] gets determined
[2860.16 --> 2860.70] at runtime
[2860.70 --> 2862.12] which is not something
[2862.12 --> 2862.62] that an AI
[2862.62 --> 2863.74] is probably going
[2863.74 --> 2865.32] to be able to do
[2865.32 --> 2866.88] at least not that soon.