Datasets:
willtheorangeguy
/
2023-Changelog-Interviews-Transcripts

Dataset card Data Studio Files
xet
 Community
text
stringlengths
0
2.46k
[1925.58 --> 1927.58] It might be in your code review tool.
[1927.58 --> 1929.34] Uh, it might be in your production logs.
[1929.34 --> 1934.42] And so we want to build integrations into Cody that just pull in all this context.
[1934.42 --> 1937.58] And I think the best way to do that is just to make this, this kind of like platform,
[1937.58 --> 1942.54] uh, this orchestrator of sorts like open source and accessible to everyone.
[1942.54 --> 1947.46] Um, the other dimension that, that is very exciting to us is, is going deeper into the
[1947.46 --> 1948.00] model layer.
[1948.00 --> 1951.88] So we've already started to do this for the embeddings based, uh, like code retrieval.
[1951.88 --> 1956.50] Um, but I think we're, we're exploring some, uh, models that are related to code generation
[1956.50 --> 1959.38] and potentially even like the chat based completions at some point.
[1959.38 --> 1964.46] Um, and that's going to be interesting cause it's, it's going to allow us to incorporate pieces
[1964.46 --> 1966.42] of source graph into the actual like training process.
[1966.42 --> 1970.30] And there's been some research there that shows that, uh, incorporating like search
[1970.30 --> 1975.34] engines into training, uh, language models actually, uh, you know, yields very nice, uh,
[1975.34 --> 1978.22] properties in terms of like lower latency, but, uh, higher quality.
[1978.72 --> 1981.84] Um, and it's also important to a lot of our customers because a lot of them are, you know,
[1981.84 --> 1987.96] large corporations, they deploy on premises and, uh, even the zero retention policy where,
[1987.96 --> 1992.16] you know, the, the code is forgotten as soon as it's, uh, you know, sent back over, uh,
[1992.16 --> 1994.02] is not good enough for, for some of our customers.
[1994.02 --> 1998.14] So they want to completely be able to self host this and, uh, you know, we plan to serve
[1998.14 --> 1998.64] them as well.
[1999.16 --> 2004.92] How high up the stack, like the conceptual stack, do you think Cody can get or maybe
[2004.92 --> 2010.42] any AI tooling with code gen with regards to like how I instructed as a developer?
[2010.66 --> 2011.06] Yeah.
[2011.18 --> 2013.92] You know, cause right now we're very much like, okay, it's autocomplete.
[2014.18 --> 2015.60] There's a function here, right?
[2015.64 --> 2020.28] I can tell it, write me a thing that connects to an API and parses the JSON or whatever.
[2020.28 --> 2023.94] And I can do, you can spit that out, but like how high up the stack can I get?
[2024.00 --> 2030.08] Can I say, you know, write me a Facebook, Facebook for dogs, you know, and be done for
[2030.08 --> 2033.12] instance, or like user stories, kind of write some user stories and go from there.
[2033.30 --> 2033.64] What do you think?
[2033.74 --> 2034.28] That's a great question.
[2034.38 --> 2038.54] I mean, we've all seen, uh, the Twitter demos by now where, you know, someone is like,
[2038.56 --> 2042.40] you know, GPT four, like build me an app and you know, it creates a working app and
[2042.40 --> 2043.34] whole website.
[2043.34 --> 2047.94] I think if you actually gone through and tried that in practice yourself, you soon realize
[2047.94 --> 2053.28] like, Hey, you can get to like a working app pretty quickly just through like instructing
[2053.28 --> 2055.38] it using English or natural language.
[2055.38 --> 2058.96] But then you get a little bit further down that path and you're like, Oh, I wanted to
[2058.96 --> 2059.40] do this.
[2059.48 --> 2060.28] I wanted to do that.
[2060.36 --> 2061.30] Can you add this bell and whistle?
[2061.48 --> 2065.82] There's kind of this like commentatorial complexity that emerges as you add like different features
[2065.82 --> 2068.42] and you're kind of diverging from like the common path.
[2068.42 --> 2070.62] And then, and then it falls apart.
[2070.70 --> 2071.74] Like I actually tried this myself.
[2071.74 --> 2076.68] Like I tried to write a complete app, uh, is actually a prototype for, for the next
[2076.68 --> 2077.24] version of Cody.
[2077.42 --> 2077.68] Okay.
[2077.84 --> 2082.52] Um, I tried to do it by not writing a single line of code just by writing English.
[2082.52 --> 2086.30] And I got like 80% of the way there in like 30 minutes.
[2086.30 --> 2087.50] And I was like, this is amazing.
[2087.70 --> 2088.70] Like this is the future.
[2088.70 --> 2090.10] Like I'm never going to code again.
[2090.20 --> 2095.42] And then the remaining 20% literally took like four hours and I was banging my head against
[2095.42 --> 2099.14] the wall because I asked it to do one thing and then it did, did it.
[2099.20 --> 2102.78] But then it kind of like screwed up this other thing and it became kind of like this like
[2102.78 --> 2103.52] whack-a-mole problem.
[2103.58 --> 2105.10] So we're not all the way there yet.
[2105.20 --> 2109.32] But I think, I think the way we think about it is like Cody right now is at the point where
[2109.32 --> 2112.50] if you ask it, uh, this is another thing I tried the other day.
[2112.50 --> 2114.16] Like I wanted to add a new feature to Cody.
[2114.58 --> 2118.60] Uh, Cody has these things called recipes, which are kind of like templated interactions with,
[2118.60 --> 2119.62] uh, Cody.
[2119.74 --> 2123.94] So like write a unit test or generate a doc string or, you know, smell my code, you know, give
[2123.94 --> 2124.56] me some feedback.
[2124.56 --> 2128.84] Like I wanted to add a new recipe and I basically asked Cody, Hey, I want to add a new recipe
[2128.84 --> 2129.40] to Cody.
[2129.94 --> 2131.72] Uh, what parts of the coach I modify?
[2131.86 --> 2134.66] And it basically showed me all the parts of the code that were relevant.
[2134.76 --> 2139.46] And then it generated the code for the new recipe using the existing recipes as like a
[2139.46 --> 2140.04] reference point.
[2140.52 --> 2143.92] Uh, and I basically got it done like five minutes and it was amazing.
[2143.92 --> 2145.82] So like, I was still obviously in the hot seat there.
[2145.82 --> 2150.60] I was still calling the shots, but it turned something that probably would have been, uh, at
[2150.60 --> 2154.12] least 30 minutes, maybe an hour, you know, if I got frustrated or, or, or distracted
[2154.12 --> 2155.74] into something that was like five minutes.
[2156.02 --> 2159.92] And that was actually, that was actually the interview question we were using for interviewing
[2159.92 --> 2161.00] on the AI team.
[2161.08 --> 2163.26] So after that we had to go back and like revamp that.
[2163.34 --> 2164.28] It's like, this is too easy.
[2164.40 --> 2165.02] Too easy now.
[2165.76 --> 2166.90] Everything just got easier.
[2167.12 --> 2167.34] Yeah.
[2167.34 --> 2172.46] Do you think this is like a, a step change in what we can do?
[2172.46 --> 2176.68] And then we're going to plateau right here for a while and like refine and, you know,
[2176.68 --> 2180.14] do more stuff, but kind of like stay at this level of quote unquote intelligence.
[2180.40 --> 2183.56] Or do you think it's like, just the sky's the limit from here on out?
[2183.56 --> 2186.72] Like, which I mean, obviously just conjecture at this point.
[2186.96 --> 2187.42] Challenging to predict.
[2187.56 --> 2189.40] I mean, it's, it's very challenging to predict.
[2189.72 --> 2194.00] Uh, you know, I might be eating my words, um, in, in another six months, but like, uh, you
[2194.00 --> 2199.16] know, on the spectrum of, you know, oh, it's just like glorified autocomplete and it doesn't
[2199.16 --> 2202.70] really know anything to all, all the way to like, you know, AGI doomer, you know, let's,
[2202.70 --> 2204.62] let's nuke the GPU data centers.
[2204.80 --> 2205.08] Right.
[2205.24 --> 2205.70] Oh my gosh.
[2205.86 --> 2208.68] Um, I just, where do you fall?
[2208.84 --> 2209.12] Yeah.
[2209.28 --> 2210.56] Don't give him ideas.