text
stringlengths
0
2.46k
[3381.64 --> 3384.76] Because I'm so sick of the word salad that spits out.
[3384.82 --> 3385.78] I'm like, I just want the answer.
[3385.98 --> 3386.08] Right.
[3386.08 --> 3387.36] The answers are, you know, useful.
[3387.64 --> 3387.82] Yes.
[3387.84 --> 3390.66] But sometimes you're like waiting for it to tell me the whole history of the thing.
[3390.76 --> 3391.08] You're like, no.
[3391.38 --> 3393.96] Well, don't you want to know, like, the retrospective while you're at it?
[3394.40 --> 3396.00] I'm being very sarcastic about it.
[3396.08 --> 3396.22] Yes.
[3396.48 --> 3399.74] People can't tell it's a podcast, but we're all eye-rolling each other on that one.
[3399.92 --> 3400.02] We are.
[3400.12 --> 3401.24] That was major eye-rolls.
[3401.24 --> 3409.74] So using it, let's say I've never used anything but ChatGBT's web UI.
[3409.86 --> 3410.02] Sure.
[3410.04 --> 3410.62] But I'm a developer.
[3410.84 --> 3411.02] Sure.
[3411.18 --> 3412.34] And I want my own.
[3412.56 --> 3414.68] I want Dolly to answer my questions.
[3414.88 --> 3415.06] Yes.
[3415.22 --> 3416.70] What does that process look like for folks?
[3416.82 --> 3416.98] Okay.
[3417.06 --> 3418.36] So you've got two choices.
[3419.06 --> 3419.60] No, no.
[3419.62 --> 3420.24] I should rephrase it slightly.
[3420.28 --> 3421.34] You've got many choices, in fact.
[3421.34 --> 3426.64] But the most common choices are we have a Databricks notebook that's in the Dolly GitHub
[3426.64 --> 3429.42] that you can just download for free, run it.
[3429.42 --> 3433.40] Now, then you're going to tell me, but Denny, I don't want to use Databricks.
[3433.82 --> 3434.34] That's fair.
[3434.44 --> 3436.44] I would prefer you to, but I understand if you don't.
[3436.70 --> 3437.06] That's fine.
[3437.16 --> 3438.06] Go to Hugging Face.
[3438.72 --> 3441.96] The instructions are all right there on how to use it.
[3442.04 --> 3445.68] In fact, like I was saying, I was actually playing with it so that way I could optimize
[3445.68 --> 3448.66] for an M1 Mac and so that the answers could come back faster.
[3448.82 --> 3448.92] Right.
[3449.24 --> 3453.50] My only problem was that when I started testing it, there was an obvious bug and pie torch.
[3454.18 --> 3454.54] Okay.
[3454.54 --> 3459.88] Because basically when we told it to go ahead and use the M1, it was giving us back garbage
[3459.88 --> 3460.30] answers.
[3460.54 --> 3462.02] Like it wasn't even like actual answers.
[3462.02 --> 3466.24] It was literally like nonsensical characters.
[3466.42 --> 3466.64] Okay.
[3466.90 --> 3469.66] And when we used CPU mode, it worked perfectly fine.
[3469.94 --> 3474.62] But then just as I was about to create a new issue on pie torch, they fixed it.
[3474.90 --> 3475.66] No, that's a good thing.
[3475.72 --> 3476.12] That's a good thing.
[3476.32 --> 3476.46] I know.
[3476.52 --> 3478.06] But I also had the fix.
[3478.42 --> 3479.34] Oh, you had the fix.
[3479.34 --> 3479.78] Okay.
[3480.34 --> 3480.68] That's it.
[3480.80 --> 3481.24] I get you.
[3481.62 --> 3482.80] You're about to have a contrived.
[3482.94 --> 3483.68] I was going to waste my time.
[3483.70 --> 3484.72] You wasted my time.
[3484.72 --> 3485.24] Damn it.
[3485.36 --> 3485.72] But no, no.
[3485.90 --> 3486.88] But it's fun.
[3486.98 --> 3490.36] But basically the idea is that obviously, okay, I shouldn't say obviously.
[3490.50 --> 3490.84] Can I share a question about that?
[3490.84 --> 3493.72] You probably don't want to train with an M1, but you can definitely do inference with
[3493.72 --> 3493.92] M1.
[3494.12 --> 3494.20] Sorry.
[3494.40 --> 3494.98] The Q&A.
[3495.06 --> 3495.78] So you got your data.
[3496.02 --> 3501.68] So how do you collect that data and how do you format it so that Dolly can understand
[3501.68 --> 3501.84] it?
[3501.84 --> 3502.18] No joke.
[3502.30 --> 3504.62] I'm assuming you're saying, so don't use Databricks data.
[3504.92 --> 3506.50] You could do the same thing like you did with the Q&A.
[3506.50 --> 3506.58] Yes, absolutely.
[3506.78 --> 3507.02] Literally.
[3507.24 --> 3507.78] It's not at work.
[3507.78 --> 3509.54] When we asked people to fill out, it was a Google form.
[3509.76 --> 3510.00] Okay.
[3510.36 --> 3511.18] That's literally it.
[3511.50 --> 3512.38] And what were the questions?
[3512.88 --> 3513.24] Oh, no, no.
[3513.36 --> 3516.36] They could produce the questions and then the answers.
[3516.58 --> 3519.28] They would ask a question and then it would spit out.
[3519.28 --> 3520.76] Provide a detailed answer for it.
[3520.80 --> 3521.12] I see.
[3521.22 --> 3523.26] So how do you make an espresso?
[3523.42 --> 3523.94] How do you make?
[3524.14 --> 3524.76] To choose coffee.
[3525.02 --> 3526.64] It wouldn't even be how do you make an espresso.
[3526.78 --> 3527.84] For example, let's be very specific.
[3528.02 --> 3528.12] Okay.
[3528.60 --> 3535.92] It would say, what are the particular features of great espresso?
[3536.36 --> 3536.66] Okay.
[3536.66 --> 3536.74] Okay.
[3537.02 --> 3541.02] And then we would talk about, okay, you're required to have a fine grind.
[3541.26 --> 3544.62] You're required to, using a conical burr grinder.
[3544.90 --> 3547.40] There's a religious war between flat burr grinders and conical burr grinders.
[3547.74 --> 3548.78] I put in conical burr grinders.
[3548.90 --> 3552.52] So, yeah, I'm sure the flat burr grinders are pissed off that that's not the answer that
[3552.52 --> 3553.12] they're going to get from Dolly.
[3553.24 --> 3553.58] That's bias.